Intelligent Systems Reference Library 209

Akash Kumar Bhoi

Pradeep Kumar Mallick
Mihir Narayana Mohanty
Victor Hugo C. de Albuquerque Editors

Hybrid Artificial
Intelligence
and IoT in
Healthcare
Intelligent Systems Reference Library

Volume 209

Series Editors
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
Lakhmi C. Jain, KES International, Shoreham-by-Sea, UK
The aim of this series is to publish a Reference Library, including novel advances
and developments in all aspects of Intelligent Systems in an easily accessible and
well structured form. The series includes reference works, handbooks, compendia,
textbooks, well-structured monographs, dictionaries, and encyclopedias. It contains
well integrated knowledge and current information in the field of Intelligent
Systems. The series covers the theory, applications, and design methods of
Intelligent Systems. Virtually all disciplines such as engineering, computer science,
avionics, business, e-commerce, environment, healthcare, physics and life science
are included. The list of topics spans all the areas of modern intelligent systems
such as: Ambient intelligence, Computational intelligence, Social intelligence,
Computational neuroscience, Artificial life, Virtual society, Cognitive systems,
DNA and immunity-based systems, e-Learning and teaching, Human-centred
computing and Machine ethics, Intelligent control, Intelligent data analysis,
Knowledge-based paradigms, Knowledge management, Intelligent agents,
Intelligent decision making, Intelligent network security, Interactive entertainment,
Learning paradigms, Recommender systems, Robotics and Mechatronics including
human-machine teaming, Self-organizing and adaptive systems, Soft computing
including Neural systems, Fuzzy systems, Evolutionary computing and the Fusion
of these paradigms, Perception and Vision, Web intelligence and Multimedia.
Indexed by SCOPUS, DBLP, zbMATH, SCImago.
All books published in the series are submitted for consideration in Web of Science.

More information about this series at http://www.springer.com/series/8578

Akash Kumar Bhoi · Pradeep Kumar Mallick ·
Mihir Narayana Mohanty ·
Victor Hugo C. de Albuquerque
Editors

Hybrid Artificial Intelligence

and IoT in Healthcare
Editors
Akash Kumar Bhoi Pradeep Kumar Mallick
Department of Computer Science School of Computer Engineering
and Engineering Kalinga Institute of Industrial Technology
Sikkim Manipal Institute of Technology Deemed to be University
(SMIT), Sikkim Manipal University Bhubaneswar, Odisha, India
Majitar, Sikkim, India
Victor Hugo C. de Albuquerque
Mihir Narayana Mohanty Graduate Program on Teleinformatics
Department of Electronics Engineering
and Communication Engineering Federal University of Ceará
Siksha ‘O’ Anusandhan (Deemed to be Fortaleza, Brazil
University)
Bhubaneswar, Odisha, India

ISSN 1868-4394 ISSN 1868-4408 (electronic)

Intelligent Systems Reference Library
ISBN 978-981-16-2971-6 ISBN 978-981-16-2972-3 (eBook)
https://doi.org/10.1007/978-981-16-2972-3

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Singapore Pte Ltd. 2021
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface

Hybrid artificial intelligence (AI) with bio-inspired techniques, genetic algorithm,

neuro-fuzzy algorithms and soft computing approaches significantly improves the
prediction of critical cardiovascular abnormalities and other healthcare solutions to
the ongoing challenging research. AI utilizes measurable methods to enable computer
frameworks to “learn” with approaching information and to distinguish examples
and settle on choices with negligible human bearing. The extent of conversation
of the potential ramifications of AI and IoT in future human services is practically
boundless. This book intended to cover the applications for hybrid AI and IoT for
an integrated approach and problem solving in the areas of radiology, drug interac-
tions, creation of new drugs, imaging, electronic health records, disease diagnosis,
telehealth and mobility-related problems in healthcare.
Chapter “Hybrid Cloud/Fog Environment for Healthcare: An Exploratory Study,
Opportunities, Challenges, and Future Prospects” discusses the areas of applica-
bility in healthcare systems of hybrid cloud/fog computing. The several extraordinary
opportunities brought by the technologies in the healthcare system with research chal-
lenges in deployment are discussed. Chapter “Hybrid Intelligent System for Medical
Diagnosis in Health Care” by Moolchand Sharma et al. focuses on the need for
a hybrid intelligent system in the healthcare industry and their medical diagnosis
applicability. Chapter “Remote Patient Monitoring Using IoT, Cloud Computing
and AI” illustrates the concept of cloud and AI-based IoT for remote healthcaring.
It also discusses different decision-making systems using AI and the principle of
operation of several cloud infrastructures used to access secured medical records.
Chapter “An Analytical Study of the Role of M-IoT in Healthcare Domain” by
Bidisha Chanda et al. brings to light several applications of M-IoT in healthcare in
measuring body temperature, monitoring blood glucose level, ECG, etc. The effect
of Internet of things has been evolving in every aspect of living; however, the effect
it has at the healthcare industry is massive because of its growing need and accuracy.
Chapter “Hybrid AI and IoT Approaches Used in Health Care for Patients Diagnosis”
proposes the AIWAC emotion interaction system, which has been designed to build

v
vi Preface

the cognitive low-power wide area zone (LPWAN) and further test the proposed AI-
enabled LPWA hybrid method. Chapter “RADIoT: The Unifying Framework for IoT,
Radiomics and Deep Learning Modeling” by Sakinat Oluwabukonla FOLORUNSO
et al. discusses the different types and sources of radiological data, feature extrac-
tion and selection method for image analysis. The chapter also presents different
ML models ideal for the radiomics and parameter tuning. Chapter “Hybrid Artificial
Intelligence and IoT in Health care for Cardiovascular Patient in Decision-Making
System” discusses on how to create an IoT device that collects body area sensor
(BAS) data with the intent of providing early warning about an imminent heart
arrest.
C. Aravindan et al. in the chapter “A Smart Assistive System for Visually Impaired
to Inform Acquaintance Using Image Processing (ML) Supported by IoT” proposed
to provide a new interface support system for a visually impaired person to inde-
pendently handle everyday life activities like recognizing and greeting a friend
during a walk in the street. Chapter “Internet of Things in Health Care: A Survey”
by Ahmed Izzat Alsalibi et al. surveys current advances in IoT-based healthcare
focused in the aspects of privacy and security techniques, e-health and m-health,
system design and architecture, cloud, fog and evolutionary computing, and network
and communication techniques. Chapter “Disease Diagnosis System for IoT-Based
Wearable Body Sensors with Machine Learning Algorithm” proposes a framework
for IoT-WBN based with a machine learning algorithm (ML). The data collected
from different wearable sensors like body temperature, glucose sensors, heartbeat
sensors and chest were transmitted through IoT devices to the integrated cloud
database. Chapter “Integration of Machine Learning and IoT in Healthcare Domain”
discusses on the integration of machine learning and IoT in healthcare domain. The
combination of these two technologies can be of great help in clinical sector in the
generation of a receptive and inter-related environment, thereby providing various
services to healthcare staffs and patients. Adriana Trus, culescu et al. in the chapter
“Managing Interstitial Lung Diseases with Computer-Aided Visualization” presented
flexible modeling of diffuse lung interstitial diseases (DLID), which is especially
important in Idiopathic pulmonary fibrosis (IPF), in which the accepted computer
tomography diagnosis criteria allow extremely diverse individual variations. Chapter
“Use of Machine Learning Algorithms to Identify Sleep Phases Starting from ECG
Signals” by Giovanna Sannino and Ivanoe De Falco presents the identification of the
different sleep phases a subject is experiencing by using heart rate variability (HRV)
values. These are computed starting from the signals gathered from electrocardio-
gram (ECG) sensors placed on the subject and online classification in an IoT-based
fully automated e-health system. Chapter “Emerging Technologies for Pandemic
and Its Impact” discusses about the emerging technologies for pandemic and its
impact and their remedies with the latest technological developments whether it is the
security, tracking, IoT-based healthcare systems, healthcare seminars, conferences,
etc. Finally, the chapter “Impact of Artificial Intelligence in Health care: A Study”
presents the role of AI in this critical healthcare sector along with some popular
existing research works in the healthcare domain. Software projects involving AI in
Preface vii

this sector are also summarized, and a real-time implementation of medical imaging
using different computational methods is demonstrated.
The editors wish to thank all the authors who have contributed to this book and
sincerely thank the Springer Nature editorial and production team for the constant
support. We send our best wishes to our readers!

Majitar, India Akash Kumar Bhoi

Bhubaneswar, India Pradeep Kumar Mallick
Bhubaneswar, India Mihir Narayana Mohanty
Fortaleza, Brazil Victor Hugo C. de Albuquerque
Contents

Hybrid Cloud/Fog Environment for Healthcare: An Exploratory

Study, Opportunities, Challenges, and Future Prospects . . . . . . . . . . . . . . . 1
Joseph Bamidele Awotunde, Akash Kumar Bhoi, and Paolo Barsocchi
Hybrid Intelligent System for Medical Diagnosis in Health Care . . . . . . . 21
Moolchand Sharma, Akanksha Kochhar, Deepak Gupta, and Jafar Al Zubi
Remote Patient Monitoring Using IoT, Cloud Computing and AI . . . . . . 51
M. V. V. Prasad Kantipudi, C. John Moses, Rajanikanth Aluvalu,
and Sandeep Kumar
An Analytical Study of the Role of M-IoT in Healthcare Domain . . . . . . . 75
Bidisha Chanda, Pradeep Kumar Mallick, and Gyoo-Soo Chae
Hybrid AI and IoT Approaches Used in Health Care for Patients
Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Shruti Mathur, Akhilesh Kumar Sharma, and Phayung Meesad
RADIoT: The Unifying Framework for IoT, Radiomics and Deep
Learning Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Sakinat Oluwabukonla Folorunso, Joseph Bamidele Awotunde,
Femi Emmanuel Ayo, and Khadijah-Khuburah Adebisi Abdullah
Hybrid Artificial Intelligence and IoT in Health care
for Cardiovascular Patient in Decision-Making System . . . . . . . . . . . . . . . . 129
M. Safa, A. Pandian, T. Kartick, K. Chakrapani, G. Geetha, and G. Saranya
A Smart Assistive System for Visually Impaired to Inform
Acquaintance Using Image Processing (ML) Supported by IoT . . . . . . . . 149
C. Aravindan, R. Arthi, R. Kishankumar, V. Gokul, and S. Giridaran
Internet of Things in Health Care: A Survey . . . . . . . . . . . . . . . . . . . . . . . . . 165
Ahmed Izzat Alsalibi, Mohd Khaled Yousef Shambour,
Muhannad A. Abu-Hashem, Mohammad Shehab, and Qusai Shambour

ix
x Contents

Disease Diagnosis System for IoT-Based Wearable Body Sensors

with Machine Learning Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Joseph Bamidele Awotunde, Sakinat Oluwabukonla Folorunso,
Akash Kumar Bhoi, Paul Olujide Adebayo, and Muhammad Fazal Ijaz
Integration of Machine Learning and IoT in Healthcare Domain . . . . . . . 223
Ananya Chattopadhyay, Sushruta Mishra, and Alfonso González-Briones
Managing Interstitial Lung Diseases with Computer-Aided
Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Adriana Trus, culescu, Laura Broască, Versavia Maria Ancus, a,
Diana Manolescu, Emanuela Tudorache, and Cristian Oancea
Use of Machine Learning Algorithms to Identify Sleep Phases
Starting from ECG Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
Giovanna Sannino and Ivanoe De Falco
Emerging Technologies for Pandemic and Its Impact . . . . . . . . . . . . . . . . . 291
Dinesh Kumar Saini, Sandhiya Bala, Akhilesh Kumar Sharma,
and Kashif Zia
Impact of Artificial Intelligence in Health care: A Study . . . . . . . . . . . . . . . 311
Devyani Bairagya, Hrudaya Kumar Tripathy, Akash Kumar Bhoi,
and Paolo Barsocchi
Editors and Contributors

About the Editors

Akash Kumar Bhoi (B.Tech., M.Tech., Ph.D.) is

working as Assistant Professor (Research) in the Depart-
ment of Computer Science and Engineering at Sikkim
Manipal Institute of Technology (SMIT), India, since
2012. He is also working (Period: 20 January 2021–
19 January 2022) as Research Associate at Wireless
Networks (WN) Research Laboratory, Institute of Infor-
mation Science and Technologies, National Research
Council (ISTI-CRN) Pisa, Italy. He is a University Ph.D.
Course Coordinator for “Research & Publication Ethics
(RPE).” He is a member of IEEE, ISEIS, and IAENG, an
associate member of IEI, UACEE, and an editorial board
member reviewer of Indian and international journals.
He is also a regular reviewer of repute journals, namely
IEEE, Springer, Elsevier, Taylor and Francis, Inder-
science, etc. His research areas are Biomedical Tech-
nologies, the Internet of Things, Computational Intelli-
gence, Antenna, Renewable Energy. He has published
several papers in national and international journals and
conferences. He has 100+ documents registered in the
Scopus database by the year 2020. He has also served
on numerous organizing panels for international confer-
ences and workshops. He is currently editing several
books with Springer Nature, Elsevier and Routledge
and CRC Press. He is also serving as Guest editor for
special issues of the journal like Springer Nature and
Inderscience.

xi
xii Editors and Contributors

Dr. Pradeep Kumar Mallick is currently working as

Senior Associate Professor in the School of Computer
Engineering, Kalinga Institute of Industrial Technology
(KIIT) Deemed to be University, Odisha, India .He
has also served as Professor and Head Department of
Computer Science and Engineering , Vignana Bharathi
Institute of Technology, Hyderabad . He has completed
his Post Doctoral Fellow (PDF) from Kongju National
University South Korea , Ph.D. from Siksha ‘Ó’
Anusandhan University, M.Tech. (CSE) from Biju
Patnaik University of Technology (BPUT), and MCA
from Fakir Mohan University Balasore, India. Besides
academics, he is also involved in various adminis-
trative activities, Member of Board of Studies to C.
V. Ramman Global University Bhubaneswar, Member
of Doctoral Research Evaluation Committee, Admis-
sion Committee etc. His area of research includes
Data Mining, Image Processing, Soft Computing, and
Machine Learning. Now he is the editorial member
of Korean Convergence Society for SMB .He has
published 12 edited books, one text Book and more than
80 research papers in National and international journals
and conference proceedings to his credit.

Dr. Mihir Narayana Mohanty is presently working

as Professor in the Department of Electronics and
Communication Engineering, Institute of Technical
Education and Research, Siksha ‘O’ Anusandhan
University, Bhubaneswar, Odisha. He has published
over 300 papers in International/National Journals, Book
Chapters, and Conferences along with approximately
25 years of teaching experience in UG and PG level.
He is the active member of many professional soci-
eties like IEEE, IET, ISTE, IRED, EMC and EMI
Engineers India, ISCA, ACEEE, IAEng, etc. Also he
is a Fellow of IE (I) and IETE. He has received
his M.Tech. Degree in Communication System Engi-
neering from Sambalpur University, Sambalpur, Odisha.
Also he has done his Ph.D. work in Applied Signal
Processing. He was working as Associate Professor and
Head in the Department of Electronics and Instrumen-
tation Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan Deemed to be Univer-
sity, Bhubaneswar, Odisha. His area of research inter-
ests includes—Applied Signal and Image Processing,
Editors and Contributors xiii

Digital Signal/Image Processing, Biomedical Signal

Processing, Microwave Communication Engineering
and Bioinformatics. He has worked as Guest lecturer
in many universities. Simultaneously has given many
invited talks. He has reviewed many Springer and
IEEE based conference papers as well as for some
International Journal Papers. Currently seven research
scholars along with two number of PG students are
under his guidance on Antenna, Signal, Image & Speech
Processing and Communication Engineering.

Victor Hugo C. de Albuquerque (M’17, SM’19) is a

collaborator Professor and senior researcher at the Grad-
uate Program on Teleinformatics Engineering at the
Federal University of Ceará, Brazil. He has a Ph.D.
in Mechanical Engineering from the Federal University
of Paraíba (UFPB, 2010), an M.Sc. in Teleinformatics
Engineering from the Federal University of Ceará (UFC,
2007), and he graduated in Mechatronics Engineering
at the Federal Center of Technological Education of
Ceará (CEFETCE, 2006). He is a specialist, mainly,
in Image Data Science, IoT, Machine/Deep Learning,
Pattern Recognition, Robotic.

Contributors

Khadijah-Khuburah Adebisi Abdullah Department of Mathematical Science,

Olabisi Onabanjo University, Ago-Iwoye, Nigeria
Muhannad A. Abu-Hashem Department of Geomatics, Faculty of Architecture
and Planning, King Abdulaziz University KAU, Jeddah, Saudi Arabia
Paul Olujide Adebayo Department of Computer Science, Federal Polytechnic,
Nasarawa, Nigeria
Ahmed Izzat Alsalibi Department of Information Technology, Faculty of Engi-
neering and Information Technology, Israa University, Gaza, Palestine
Rajanikanth Aluvalu Vardhaman College of Engineering, Hyderabad, India
Versavia Maria Ancus, a Department of Computer and Information Technology—
Automation and Computers Faculty, “Politehnica” University of Timisoara,
Timis, oara, RO, Romania
C. Aravindan Department of ECE, SRM Institute of Science and Technology,
Chennai, Tamil Nadu, India
xiv Editors and Contributors

R. Arthi Department of ECE, SRM Institute of Science and Technology, Chennai,

Tamil Nadu, India
Joseph Bamidele Awotunde Department of Computer Science, University of
Ilorin, Ilorin, Nigeria
Femi Emmanuel Ayo Department of Computer Science, McPherson University,
Ogun State, Nigeria
Devyani Bairagya School of Computer Engineering, Kalinga Institute of Industrial
Technology (KIIT), Deemed To Be University, Bhubaneswar, Odisha, India
Sandhiya Bala Nielsen Global Connect Chennai, Chennai, India
Paolo Barsocchi Institute of Information Science and Technologies, National
Research Council, Pisa, Italy
Akash Kumar Bhoi Department of Computer Science and Engineering, Sikkim
Manipal Institute of Technology (SMIT), Sikkim Manipal University (SMU),
Majitar, Sikkim, India;
Institute of Information Science and Technologies, National Research Council, Pisa,
Italy
Laura Broască Department of Computer and Information Technology—Automa-
tion and Computers Faculty, “Politehnica” University of Timisoara, Timis, oara, RO,
Romania
Gyoo-Soo Chae Division of Smart IT Engineering, Baekseok University, Cheonan-
si, South Korea
K. Chakrapani Department of Information Technology, School of Computing,
SASTRA Deemed University, Tanjore, Tamil Nadu, India
Bidisha Chanda School of Computer Engineering, Kalinga Institute of Industrial
Technology, Deemed to be University, Bhubaneswar, India
Ananya Chattopadhyay School of Computer Engineering, KIIT Deemed To Be
University, Bhubaneswar, India
Ivanoe De Falco Institute for High-Performance Computing and Networking
(ICAR), National Research Council of Italy (CNR), Naples, Italy
Sakinat Oluwabukonla Folorunso Department of Mathematical Science, Olabisi
Onabanjo University, Ago-Iwoye, Nigeria
G. Geetha School of Computing, SRM Institute of Science and Technology,
Kattankulathur, Chennai, Tamil Nadu, India
S. Giridaran Department of ECE, SRM Institute of Science and Technology,
Chennai, Tamil Nadu, India
V. Gokul Department of ECE, SRM Institute of Science and Technology, Chennai,
Tamil Nadu, India
Editors and Contributors xv

Alfonso González-Briones Research Group on Agent-Based, Social and Interdisci-

plinary Applications (GRASIA), Complutense University of Madrid, Madrid, Spain;
BISITE Research Group, University of Salamanca, Salamanca, Spain;
Air Institute, IoT Digital Innovation Hub, Salamanca, Spain
Deepak Gupta Maharaja Agrasen Institute of Technology, MAIT, Delhi
(GGSIPU), India
Muhammad Fazal Ijaz Department of Intelligent Mechatronics Engineering,
Sejong University, Seoul, South Korea
M. V. V. Prasad Kantipudi Sreyas Institute of Engineering and Technology,
Hyderabad, India
T. Kartick School of Computing, SRM Institute of Science and Technology,
Kattankulathur, Chennai, Tamil Nadu, India
R. Kishankumar Department of ECE, SRM Institute of Science and Technology,
Chennai, Tamil Nadu, India
Akanksha Kochhar Maharaja Agrasen Institute of Technology, MAIT, Delhi
(GGSIPU), India
Sandeep Kumar Sreyas Institute of Engineering and Technology, Hyderabad, India
Pradeep Kumar Mallick School of Computer Engineering, Kalinga Institute of
Industrial Technology, Deemed to be University, Bhubaneswar, India;
Division of Smart IT Engineering, Baekseok University, Cheonan-si, South Korea
Diana Manolescu Department of Radiology, Center for Research and Innovation in
Precision Medicine of Respiratory Diseases (CRIPMRD), “Victor Babes, ” University
of Medicine and Pharmacy, Timis, oara, RO, Romania
Shruti Mathur Manipal University Jaipur, Jaipur, Rajasthan, India
Phayung Meesad King Mongkuts University Thailand NB, Bangkok, Thailand
Sushruta Mishra School of Computer Engineering, KIIT Deemed To Be Univer-
sity, Bhubaneswar, India
C. John Moses Sreyas Institute of Engineering and Technology, Hyderabad, India
Cristian Oancea Department of Pneumology, Center for Research and Innova-
tion in Precision Medicine of Respiratory Diseases (CRIPMRD), “Victor Babes, ”
University of Medicine and Pharmacy, Timis, oara, RO, Romania
A. Pandian School of Computing, SRM Institute of Science and Technology,
Kattankulathur, Chennai, Tamil Nadu, India
M. Safa School of Computing, SRM Institute of Science and Technology, Kattanku-
lathur, Chennai, Tamil Nadu, India
Dinesh Kumar Saini Manipal University Jaipur, Jaipur, India
xvi Editors and Contributors

Giovanna Sannino Institute for High-Performance Computing and Networking

(ICAR), National Research Council of Italy (CNR), Naples, Italy
G. Saranya School of Computing, SRM Institute of Science and Technology,
Kattankulathur, Chennai, Tamil Nadu, India
Mohd Khaled Yousef Shambour The Custodian of the Two Holy Mosques Insti-
tute for Hajj and Umrah Research, Umm Al-Qura University, Mecca, Saudi
Arabia
Qusai Shambour Department of Software Engineering, Faculty of Information-
Technology, Al-Ahliyya Amman University, Amman, Jordan
Akhilesh Kumar Sharma Manipal University Jaipur, Jaipur, Rajasthan, India
Moolchand Sharma Maharaja Agrasen Institute of Technology, MAIT, Delhi
(GGSIPU), India
Mohammad Shehab Department of Software Engineering, The World Islamic
Science and Education University, Amman, Jordan
Hrudaya Kumar Tripathy School of Computer Engineering, Kalinga Institute of
Industrial Technology (KIIT), Deemed To Be University, Bhubaneswar, Odisha,
India
Adriana Trus, culescu Department of Pneumology, Center for Research and Inno-
vation in Precision Medicine of Respiratory Diseases (CRIPMRD), “Victor Babes, ”
University of Medicine and Pharmacy, Timis, oara, RO, Romania
Emanuela Tudorache Department of Pneumology, Center for Research and Inno-
vation in Precision Medicine of Respiratory Diseases (CRIPMRD), “Victor Babes, ”
University of Medicine and Pharmacy, Timis, oara, RO, Romania
Kashif Zia Sohar University Oman, Sohar, Oman
Jafar Al Zubi School of Engineering, Al-Balqa Applied University, Jordan, India
Hybrid Cloud/Fog Environment
for Healthcare: An Exploratory Study,
Opportunities, Challenges, and Future
Prospects

Joseph Bamidele Awotunde , Akash Kumar Bhoi , and Paolo Barsocchi

Abstract The healthcare system has been on the frontline in recent years, and
new technologies have greatly benefited healthcare. Researchers have tried to find
solutions to different problems associated with the healthcare system by applied
various modern technologies approaches. Among the various technologies, are fog
and computing used in smart healthcare systems. These applications with the Internet
of things (IoT) recently have help in dispersed patient data globally and have advanced
healthcare systems. Hence, various applications and solutions using cloud computing
have been proposed by researchers to manage healthcare statistics. However, the
issues of latency, context-awareness, and a huge volume of data are remaining chal-
lenges in cloud computing. Hence, the possibility of transmission errors and the
probability of delay in data processing remain a problem as healthcare datasets
become more complex and larger. The most alternative solution to those challenges
is fog computing in reducing data management complexity in the healthcare system,
thus increasing reliability. But, before using fog computing, it is very essential to
look into its associated challenges in other to manage healthcare data effectively.
Therefore, this chapter discusses the areas of applicability in healthcare systems
of hybrid cloud/fog computing. The several extraordinary opportunities brought by
the technologies in the healthcare system with research challenges in deployment are
discussed. The applications of fog in IoT-based devices bring healthcare components
in a distant cloud operating nearer to data sources and the end-users, thus, resulting
in context-awareness and lower latency.

J. B. Awotunde (B)
Department of Computer Science, University of Ilorin, Ilorin, Nigeria
e-mail: [email protected]
A. K. Bhoi
Department of Computer Science and Engineering, Sikkim Manipal Institute of Technology
(SMIT), Sikkim Manipal University (SMU), Majitar, Sikkim 737136, India
A. K. Bhoi · P. Barsocchi
Institute of Information Science and Technologies, National Research Council, 56124 Pisa, Italy
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 1
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_1
2 J. B. Awotunde et al.

Keywords Hybrid cloud · Fog computing · Healthcare system · Internet of

things · Latency · Technologies

1 Introduction

The healthcare sector faces challenges including budget pressures, management of

various service delivery, and growth in aging populations [1]. The increased use of
information and communications technology (ICT) will help the healthcare sector
reduce if not overcome these challenges [2]. In connection to the need to make
service delivery more effective, ICT innovations have contributed to higher healthcare
technology adoption [3, 4]. To help healthcare, ICT has been used to provide greater
access to medical records [5] and the area of decision making in the healthcare system;
ICT has additional ability to help minimize costs and enhance service deliveries
[2, 5, 6].
One of the world’s most recent groundbreaking technology is cloud computing
(CC), and the applications are increasingly growing daily. The applications CC is
widely used in smart healthcare systems having storage devices capacity to stored
big data generated by the IoT-based devices daily, a significant part of the services
have been shifted to the cloud. This has help healthcare systems to provide cost-
effective and reliable services globally. The CC has brought an initiative to migrate a
widespread perception that such cloud boundaries and security problems will obstruct
the transition. To provide better services with an avoidable price to patients, the
healthcare system has been a move to the cloud globally with better healthcare
facilities.
The need for a greater low cost with effective healthcare system necessitates
the need for cloud computing in the medical system, thus explain why healthcare
providers’ services switching to the cloud. According to the report, various branched
of medical sciences have benefited from this paradigm. For example, cardiology has
benefited from efficient data storage and retrieval systems in accessing historical and
current patient data on the cloud database. The maintenance of CC software licensing
that is very low about other medical software contributed to the increased popularity.
In the delivering of an efficient information technology (IT), CC offers an innova-
tive system and play paramount roles. The technology is identified as one of the major
technology to enhance healthcare service deliveries using IoT-based devices in inpa-
tient data collections. When combines IoT and CC technologies, they benefited in an
equal manner with mutual understanding in handling medical information. Devel-
oped a monitoring system combining Cloud and Edge computing technologies has
been proved efficient in the areas of providing healthcare facilities in remote areas,
thus helps the caregivers and physicians to provide qualities healthcare services to
their patients. The CC is used as a supportive technology in an IoT-based system in
terms of computational capability, storage, resource utilization, and reduced energy
consumption. Also, the cloud has been a favorite of IoT-based system by enhancing
service deliveries globally and deliver unspeakable services in a distributed and
Hybrid Cloud/Fog Environment for Healthcare … 3

Table 1 Appraisal of cloud

Requirements Cloud computing Fog computing
and fog computing
Mark User Internet Users Mobile users
Location of servers Within Internet Edge nodes
Service type Global information Localized
information
services
Geographical Centralized Distributed
distribution
Distance between Multiple hops Single hop
client and server
Delay jitter High Low
Latency High Low
Type of connectivity Leased line Wireless
Location awareness No Yes
Server nodes Few Large
N/W bandwidth More Low
Reply time Minutes Milliseconds,
sub-seconds
Security Less secure Very secure

dynamic manner. The IoT-based cloud framework can still be extended in the smart
environment for the development and application of new service delivery. Table 1
compared the cloud and fog computing using IoT-based requests.
The huge amount of big data generated by IoT-based devices can no longer be
handle by the CC again, new powerful computing models are required. The security
concerns, low latency, speedy processing requirements need new powerful computing
techniques to best place processing, conserve network bandwidth, and making IoT-
based system operates in a reliable environment [7]. All these IoT-based system
requirements can never be met with the traditional CC architectures alone; therefore,
a better and powerful computing model is required. Latency creates dominant strategy
by transfers data for processing from the network edge to the data center. Bandwidth
is quickly outpaced by traffic from thousands of users. Also, the cloud servers neglect
other protocols the IoT devices use and interact only with IP. The best location for
most IoT data to be analyzed is close to the machines that generate and function on
that data, and this is called computing with fog.
To close the gap and bridge linking IoT-based devices, a powerful technology
called fog models has been employed to help the IoT-based devices to process the
big data generated from the patient. The processes at the edge of the computa-
tional resources become easy with the fog model to handle and outline the data
from IoT-based devices, thus, become easy and greatly improved. The idea of cloud
computing is very similar to fog computing since both are built with virtual systems
and offer many of the similar architectures and features that facilitate the versa-
tility and scalability of computing, storage, and networking resources on-demand
4 J. B. Awotunde et al.

supplies. Although with the emerging trend in networking in terms of demand, the
two technologies have a wide barrier. The businesses and the end-users are free to
use cloud computing from defining certain specifics, such as storage capacity, limits
on computing, and the cost of network connectivity. The problem of latency sensitive
applications in real-time still arises for nodes to meet their delay requirements is still
arising [8, 9].
The issue of security of these huge volumes of data should also be the main concern
for any business-minded experts because the problem causes to their reputation and
they are constrained by the law to keep all data safe. The cloud-based provides
the liberty of accessing data from the service providers anytime in any part of the
world, hence, exposure the IoT-based data to security and privacy threats. In medical
practice, the use of cloud/fog computing has increased tremendously.
To gain more accurate diagnosis results, cloud/fog has been widely used to
decrease the burden on the medical experts and help in decreasing the decision time
of traditional methods of the diagnosis process. There are significant and improve-
ments in the treatment, prediction, screening, drug/vaccine development processes,
and application of medication in healthcare sectors with continuing expansion in
cloud/fog computing. Their applications have reduced human intervention in medical
processes, and the cost of medical applications has reduced. Fog layer has moved
the cloud closer to end-users, and the data center can now be process close to the
network nodes. The computational at the edge of network nodes becomes easier and
the cloud storage has greatly expanded by the fog models.
This triggers the idea of a cloud-focused cutoff and embarks on how the sequence
of data generated by IoT-based devices can be stored and operated. CISCO introduced
another version of fog computing at the edge of the network that can used billions
of connected devices application strenuously [10, 11]. Fogging as it is called is
a distributed database that allow a remote data center manages its services at the
network edge, and the program runs on them. Fog is about the real world being dealt
with. The edge or fog paradigm addresses the problems with the basic concept of
finding small servers in the vicinity of users and devices called edge servers [12].
Therefore, this chapter presents both applications of CC and fog in smart healthcare
system; also, their applicability and challenges were also discussed. The deployment
of cloud and fog computing has greatly helped the smart healthcare system.

2 Applications of Cloud Computing in Smart Healthcare

System

Cloud computing will play an essential part in absorbing healthcare transformation

expenses, optimizing assets, and bringing the new age of technology to life. Emerging
policies are targeted at obtaining data at anytime, anywhere that can be achieved by
transferring health data to the cloud. This contemporary distribution model will make
healthcare more productive and operational and lowering the price of innovation
Hybrid Cloud/Fog Environment for Healthcare … 5

expenditures [13], but it also presents some obstacles due to issues regarding the
security of sensitive health information and compliance with a specific criterion
such as HIPAA. Healthcare providers, taking into account these security and privacy
concerns, can unquestionably reap the benefits of cloud computing technologies and
provide substantial benefits, such as assisting to expand the eminence of service for
patients and reducing healthcare spending [14].
Cloud computing’s critical features are (1) self-service on-demand, (2) huge
amount of database, (3) allow devices sharing, (4) swift elasticity, and (5) calculated
facilities. In complex resources, clouds offer benefits such as processing energy or
storage abilities, universal access to resources from anywhere at any time, and high
resource versatility and scalability. In several business fields, these advantages have
been the purpose for the growing acceptance of cloud computing. This principle has
also evidently been adopted in the area of healthcare in recent years.
The cloud with IoT-based online application works efficiently over the ordinary
cloud-based applications. The medical, bank, and military are instance of sectors that
have benefited greatly from emerging models. The cloud-based IoT-based system has
helped in providing medical sectors with effective services in the areas of monitoring
and access of medical records remotely. IoT-centric application has been used to
collect real time from patients using devices and sensors. Also, artificial intelligence
can be used to analyzed data capture using IoT-based devices for effective diagnosis
of disease at the correct time before the illness reach advanced stages.
The CC has the capacity to share data between different systems within IoT-based
system. This capacity is something that IT urgently needs for healthcare. Cloud
computing, for example, can enable healthcare professionals to share data such
as EHR, doctor’s references, medications, insurance data, research reports stored
via various information systems. In the radiological market, where many organiza-
tions have switched to the cloud to minimize their computing costs and promote the
sharing of pictures, this is already happening [15]. Cloud computing has provided
clinics, hospitals, insurance providers, pharmacies, and other healthcare companies
the ability to agree to cooperate and exchange healthcare data to provide improved
service quality and minimize costs. Looking at the developments in the industry, it
seems that once all the obstacles it presents are resolved, cloud-based schemes will
eventually turn out to be the standard in healthcare.
The healthcare system’s ecosystem, which comprises health insurance providers,
hospital and physician networks, laboratories, clinics, patients, and other institutions,
is vast, diverse and highly nuanced [16]. And all of these must operate under many
government regulations [17]. To function efficiently and rapidly in this ecosystem,
any sensitive details must be exchanged confidentially and in a safe way between
these agencies quickly and accurately. In the healthcare sector, protecting the patient’s
data is known to be very sensitive to privacy issues. Possibly, one of the reasons why
the development of healthcare moving into the cloud has negatively affected. Inno-
vative technology and resources must be managed when it comes to cloud sharing.
However, as they theoretically range between cities, states, and even nations, many
other records, knowledge, and resources can benefit from collaboration by cloud
usage.
6 J. B. Awotunde et al.

Private clouds tend to be deployed first because of security issues in the current
scenario and then shift into public networks [16]. It may be a good idea to first
set out the healthcare industry’s top priorities and then analyze which elements of
cloud computing can be efficiently implemented to support them. The efficiency of
services provided to patients and customers, privacy, data security and integrity, and
catastrophe recovery seems to be at the forefront of today’s rising global healthcare
costs [17, 18]. Some of the inherent features can be leveraged to meet some of these
objectives, such as flexible architecture, data centers for the provision of permanent
data, protection models, and rapid access to information, among others.
Cloud computing encourages IT facilities that are accessible from all locations and
at all times [19]. It is a new mechanism, not a new technology, to deliver computing
services [20]. Examples of non-medical cloud services are Microsoft Office 365 and
Google Docs, while examples of medical service apps are Microsoft HealthVault and
Google Health platforms [21]. Compared to traditional computing, there are three
major enhancements provided by the cloud computing model: (1) computationally
intensive solutions are accessible on request, (2) service delivery without charge.
Customer upfront commitment requirements, and (3) flexibility for short-term use
[22]. Many sectors have been influenced by the cloud model, and it is estimated that
approximately, 80% of today’s businesses will have embraced cloud computing by
2020 [15]. Besides, companies that lack capital and infrastructure should implement
cloud computing to set up on-site applications [23]. Cloud computing, especially
within the electronic health records (EHRs) field, is transforming healthcare IT [24].
Cost minimization in IT investments will contribute to improved healthcare facilities
[25] and estimated that drug costs can be decreased by 80% and payment can be
done within 2 h for patients and insurance providers as compared to up to seven days
with an implementation.
To dynamically compute patient records with sensors that are attached to medical
equipment to process data for collection, accessibility, and distribution, a cloud-based
framework has been suggested. During any disease outbreak, this device can reduce
typical errors or data collection errors manually [26, 27], not just simplification of
the procedure, but also increased access to high-quality data [28]. By combining the
ambulance services with patient records, the Greek National Health Service has built
an emergency care program in the cloud, ensuring direct access for doctors while
being willing to use all resources while maintaining low costs as much as possible
[29].
In Australia, as a partnership between Telstra and the Royal Australian College
of General Practitioners (RACGP) suggested an e-health cloud [15]. The goal of this
collaboration is to develop diagnostic and response to situations, medical software,
medications, and training and referral facilities. Cloud processing technologies have
provided successful support for bioinformatics research in the medical field [15,
30, 31]. Although cloud computing has several value-added ideas driven by a novel
paradigm of IT service distribution over the network, economic benefits appear to
be the most significant factor in its popularity and widespread acceptance.
Lowering the cost of healthcare delivery is a significant catalyst for the implemen-
tation of cloud technology in healthcare. This expense has risen to such immense
Hybrid Cloud/Fog Environment for Healthcare … 7

proportions that governments are facing severe problems with funding. The real-
ization that patient care can be enhanced by technology while lowering costs has
ensured that policymakers can drive the historically sluggish healthcare sector to
a faster rate of adoption. Big data development in healthcare is another signif-
icant factor [32]. When the quantity of digital information grows, the capacity
to manage this information is becoming an increasing challenge. This knowledge
embraces the keys to prospective clinical developments but is also inaccessible to
scientists. Cloud computing can be the supporting reason for large-scale knowledge
exchange and convergence [33]. The chapter by Merelli et al. [34] addresses high-
performance computing (HPC) bioinformatics solutions, big data analysis paradigms
for computational biology, and the challenges that are still accessible in the fields of
healthcare.
In particular, the authors pointed out that, thanks to virtualization that prevents
transferring too much big data, cloud computing solves big data management and
analysis problems in many fields of healthcare. Also, in the particular area of
telemedicine, it is critical to have an infrastructure to support high throughput, high
capacity of storage, and safe connectivity to allow effective management and auto-
matic analysis of broader patient populations. Horizontal scalability (i.e., the capacity
of a device to efficiently extend its resource pool for managing heavy loads) and
spatial usability (i.e., capacity to retain performance, usefulness, or usefulness inde-
pendent of local area concentration advancement to a more dispersed geographic
pattern) are two criteria that can be fulfilled by cloud computing [33]. In the medical
imaging region, the amount of data can exceed petabytes thanks to high-resolution
imaging instruments. It is also apparent that the Cloud Computing Paradigm will
render a significant benefit to addressing the computational needs relevant to medical
image reconstruction and processing and to facilitate the large exchange of digital
images and also advanced control processing.
Cloud computing is a new and progressively evolving field of healthcare improve-
ment. In combination with a pay-per-use model, universal, on-demand access to
nearly limitless resources allows for new ways of creating, providing, and utilizing
services. In an “OMICS setting,” cloud computing is also used in genomics,
proteomics, and molecular medicine computing. Medicine is a collaborative and
highly data-intensive endeavor [35]. Advances in the OMICS fields produce large
quantities of data to be processed and stored (genomics, proteomics, and the like).
The subordinate use of medical data with text or data mining techniques also
implies an increasing request for complex, accessible services. These tools are also
solitary temporarily used so that stable infrastructure projects are difficult to justify
and, alternatively, flexible on-demand services are pursued. To meet these demands,
cloud computing seems to be a feasible alternative. Commercial providers such as
Amazon and Microsoft pledge to make available at their fingertips hundreds of virtual
machines, almost instantly and only they are just wanted for the moment. The benefit
of such deals is that they only have to be paid for the setup, scale, and period they
are essentially used during these services.
Massive medical costs and the maintenance of big data during any disease outbreak
require technical advances so that at any time and everywhere, everybody has access
8 J. B. Awotunde et al.

to healthcare services. The development of technology has allowed telehealth to

provide online healthcare facilities. For patients that are permitted to travel, for
villages in rural zones, and for individuals that do not have access to medical care,
remote facilities are useful. The uses for telemedicine include the transmission and
storage of medical images, video conference patient counseling, continuing educa-
tion, and facilities in the electronic healthcare field. Sadly, the use of telemedicine
technology is hindered by technical and financial costs [36]. To this effect, studies
have given cloud computing that offers, among other things, remote support capa-
bility, accessible transparent resources, efficient large Internet connectivity, scalable
and resources pooling, robust medical data sharing and processing, and the sharing
of big data patient records.
Many studies have found that inadequate access to patient information is the
explanation for most medical errors especially during infectious diseases outbreak
[37]. The cloud-based medical system has been regarded as a possible system to
increase openness and reduce the extent of medical errors during disease diag-
nosis to correct health data [38, 39]. Many medical organizations have also chosen
cloud storage to obtain and store broad patient data and maintain their electronic
health records systems. Electronic health records have evolved rapidly over the last
decade, providing a gorgeous basis for data mining to recognize designs and styles in
the big data industry in healthcare. Another common point for exchanging medical
data is the interchange of electronic health records. By communicating a common
hub, these businesses facilitate healthcare sectors to transmit information rather than
maintaining ties with many peer businesses [40].
Cloud computing also offers secure storage and sharing resources that can reduce
the amount of local traffic to make organizations agile [41]. By reducing the cost
needed for starting up automated medical records, which is lacking in many health-
care segment facilities, will improve the efficiency of the healthcare sector [42].
During a disease outbreak, prescriptions and diagnoses, for instance, can be shared
through the cloud over different systems. Therefore, for service enhancement and
higher standards, hospitals and doctors exchange patient records. The primary advan-
tages of electronic health record cloud storage are the capacity to exchange patient
records with other specialists at home and overseas, the facility to pool data in one
location, and the capacity to access files anytime, anywhere. Electronic health record
cloud computing enables patients to view, replicate and transfer their secure health
records [43]. Regardless of the influences of cloud computing to capture and store
large health data, the prime problem is the failure of the network, protection, and
privacy of patient information that users, hackers, malware, and so on are exploiting
[44, 45].
Figure 1 displayed the applicability of cloud computing in the smart healthcare
system. There is an increasing influx of people to urban areas today. Healthcare facil-
ities are one of the most critical characteristics that have a major effect on people
arriving in city centers during infectious disease outbreaks globally. Metropolises
are therefore financing a digital transition to offer residents healthy environments
[46]. On the other hand, because of its huge number, high speed, and high variety,
conventional models and methods for full conservational performance assessment
Hybrid Cloud/Fog Environment for Healthcare … 9

Elderly and
Middle-aged Patients
Doctor
Family

Pharmacis

Healthcare
Cl d

Caregiver

Wearable Devices (Pressure

Monitor, Pulse, Oximeter, Blood
Glucose, Body Temperature)

Physical
Activities Service
Providers Digital
Authority
Telehealth

Fig. 1 Applications of cloud in smart healthcare system

are threatened by the advent of big data [5]. Also, because of their carbon emis-
sions, conventional ICT systems damage the atmosphere [47]. On the other hand,
cloud services are a cost-effective medium for accommodating large-scale infras-
tructure systems have gained considerable acceptance. The use of cloud computing
is, therefore, a significant phase in the green processing process that saves resources
and protects the atmosphere. The use of sufficient equipment and cloud space saves
the organization’s resources and eliminates the costs related to cooling systems,
computers, and central servers. Nevertheless, cloud computing supports renewable
computing with energy savings, rendering dangerous articles less harmful [13].
Through using intelligent mobile computers, cloud computing has inspired health-
care specialists to observe the wellbeing of patients at home remotely [27]. Besides,
IoT will build a network by leveraging integrated sensors to track the patient’s
real-time health status and control the treatment process. The IoT would also
play an important role in the development of healthcare for the next generation.
Although health monitoring systems for IoT-based patients are popular, observing
them outdoor hospital requirements increases the IoT’s cloud computing capabilities
for the handling and storing of health data [48].
10 J. B. Awotunde et al.

3 Applications of Fog Computing in Smart Healthcare

System

Fog computing allows a new range of applications and services by extending the
CC model. The fog has the following distinct features: (a) heterogeneity of appli-
cations, (b) knowledge location and low latency; (c) broad physical circulation; (d)
flexibility; (e) produce the huge number of nodes; (f) wireless access predominant;
(g) real-time applications and heavy presence of streaming; (h) heterogeneity of
applications. Since the edge of the network of fog is not specifically located, they
are highly scalable, hence network services between end devices and conventional
cloud data centers. Both fog and cloud building blocks are networking tools, storage,
and computation, edge network created various features that make fog a non-trivial
cloud extension.
The help endpoints come with rich network edge resources give birth to
fog computing with low latency applications requirement, with applications like
augmented reality, video streaming, and gaming. Fog computing targeted applica-
tions with a dispersed implementation that are more centralized on the cloud. For
instance, fog plays an active role through proxies and access points in providing
moving vehicles with high-quality streaming located along highways and tracks.
Large-scale sensor networks are other examples of inherently distributed systems
that require fog computing and storage resources to monitor the smart grid and the
environment. As evidenced in sensor networks, wide range geo-distribution, and
smart grid required a large number of nodes.
For applications, communicating directly with the mobile devices, thus, highly
supports mobility techniques like the LISP protocol, which decouples host iden-
tity from location identity. These mobile applications require a distributed directory
system. The applications involve real-time interactions rather than batch processing
in small cloud applications. The fog nodes will be deployed in a wide environ-
ment since the nodes come in various shapes. Streaming is a good example of fog
computing supporting certain services seamlessly and requires the cooperation of
various providers. Therefore, the applications must be federated across domains and
very necessary to be able to interoperate their components. The fog plays a prominent
role in the ingestion and processing of the data close to the source.
To prolong the life span of the battery or make it harvest electricity, wireless
sensor nodes (WSNs) are very useful and are made for this purpose. The WSNs need
low bandwidth, low processing power, small memory, unidirectional sink collector
sources to function very well. The tinyOS2 a de facto standard operating system is an
example of this class of sensor networks that has basic processing and transmission
of the static sink with environmental sensing. The motes have been used to collect
environmental data like temperature, amount of rainfall, light intensity, and humidity
among others [8].
Figure 2 displayed the fog computing ecosystem in a smart healthcare system.
The WSNs progressed in many directions like in the areas of wireless sinks, multiple
Hybrid Cloud/Fog Environment for Healthcare … 11

Fig. 2 Fog computing in smart healthcare system

sinks, distributed sinks, and to meet the requirements of modern applications, smart-
phone sensors have been proposed. The WSNs in successive incarnations have
provided energy-restricted in use. The actuators are requiring to perform physical
acts in applications beyond sensing and monitoring like in deploying sensors, even
carry, open, close, move, focus, and target. The actuators provide sensor networks
with new dimensions that can monitor either a device or the measurement process
itself. Since the controller node to actuators, sensors to sink, the flow of information
is not unidirectional, from the sensors to the sink. The issues of possible oscillatory
activity and stability are of importance; it then becomes a closed-loop device in a
subtler, but critical way. It has become a concern in a quest to provide real-time and
rapid response to works on jitter and latency in the smart healthcare system [8].
In Kashi and Sharifi [8], the contributions of actuator networks and wireless sensor
to the coordination of WSANs were discussed. The results of the survey shown that
WSAN has two networks in one architectural layout (i) the devices for capture patient
data and (ii) mobile ad hoc network (MANET). Banka et al. [27] embedded in their
work collaborative adaptive sensing of the atmosphere (CASA) with emphasis on
new technologies with higher bandwidth needs a collaborative sensing environment.
12 J. B. Awotunde et al.

Zink et al. [11] deployment information of fog computing to increase healthcare

services in remote areas which has been proved used by physician and caregivers.
The fogs become the appropriate environment for both WSNs and WSANs having
features like proximity and position sensitivity, geo-distribution, the hierarchical
organization in helping discharge their energy-restricted [8].
Fog computing takes resources to the edge of the network as an extension to
cloud computing. This brings CC closer to IoT-based devices collections center, thus
assisting end-users to be closer to the real-time services and speed up their services.
These decentralized models bring IoT-based devices closer to the edge where data
is generated, thus become the main objective of fog computing. The amount of data
transmitted to the cloud for processing and analysis is been reduced by the fog layer,
hence increase the security one of the main issues in the IoT-based industry [8].
The fog layer provides a meeting point for storage resources, network nodes, and
computing are available for proper management for local data capture to deliver fast
results and data are readily obtainable. The trade-off between processing performance
and power consumption is provided using low-power system-on-chip (SoC) systems.
Cloud servers however behalf like advanced analytics and performed the functions of
machine learning jobs combing time series generated by a variety of heterogeneous
or mixed kinds of items [48].
With the rapid development in wireless technology, IoT can generate a huge
amount of medical data, smart devices, and customized enhance services. Such big
medical data can be of countless types, which the cloud server needs to store, process,
and thereafter analyze [57]. High latency, security problems, and network traffic arise
as a result of the handling of big cloud medical data. Fog computing was introduced
to reduce the workload of the CC IoT-based system. The fog acts as a flyover among
terminal devices and cloud servers and helps in bringing the cloud service closer to
the network edge, hence provided secured and refined smart healthcare services.
The edge of the network is an ideal place for analyzing instantaneous gener-
ated data closer to where data is created. The feature of fog computing placed
it ahead of cloud computing like data pre-processing, local data analytics, data
security and privacy, temporary storage, data trimming, distributed, decentralized
storage. The combined computing paradigm is needed in an IoT-based applications
like health monitoring systems to efficiently perform big data analytics. The use of
fog computing has created a better way of handling cloud database for a smarter
healthcare system [57].

4 Challenges of Cloud and Fog Computing in Smart

Healthcare System

The big data 5Vs data importance has a result of the huge amount of patients’ data
receives from the medical device such as volume, veracity, variety, and velocity.
As a result, a fog node is needed to be connected to receive, store, process, and
Hybrid Cloud/Fog Environment for Healthcare … 13

communicates with IoT-based devices. The system administrative configuration must

be controlled to forestall the data fluctuation between fog and cloud computing. To
handle various types of data like text, videos, audios, and image files, fog layer
regular proprieties with data format are needed from various like smartphone, and a
smartwatch. For regular data transfer and urgent data requests, a gateway sufficient
router is needed.
The portable devices and sensors are used for patient data collection. There is a
need for adequate protection for medical facilities so that patient can use their smart-
phone for health status updates. The use of a smart healthcare system creates possible
ways of expanding the healthcare system to the whole population. The appointment
time used by patients to see physicians or waits for diagnosis outcomes can be
reduced with the use of the intelligent healthcare system. This also provides direct
access to real-time medical care and services. To maintain trust between patients and
medical experts, the scalability of a smart healthcare system must be taking with all
seriousness, and this will in turn save quality time.
It is the main concern and it is not appropriate to obtain information from end-
users through unauthorized entities, and this, also, poses threats to the personal safety
of medical data. The main problem in the introduction and deployment of the smart
healthcare system is security and privacy. However, with the integration of these
layers, security is needed in any layer, such as the system layer, fog layer, and cloud
layer [8]. Another problem facing fog computing is heterogeneity that refers to the
various communication capable devices. The devices like smartphones, autonomous
vehicles with other IoT smart devices are at the bottom of the layers within the
IoT-based system.
The heterogeneity within IoT-based systems arises during data processing, data
formatting, and data clearing, thus creates difficulty in processing medical infor-
mation. The enabling of a network to connect with various sensors is an example
challenge in a smart healthcare system for the monitoring of patients. For this to
take place, heterogeneity must be present when the data is transferred to another
system for processing or analysis. The fog layer while going up to the next layer
involves various nodes, clusters, switches, and other devices that are needed during
data processing and communication facilities [49]. In the designing of architecture
that enables multiple monitoring of devices, heterogeneity is, therefore, an important
factor to communicate with end-users using IoT-based devices and sensors [11].
There is another research issue in fog computing due to the location of the network,
their protection can create a concern. The threats that are not present in an organized
cloud architecture arise in fog computing, and this poses a threat in IoT-based system.
The major threat from this problem is when an attacker transmits and changes contact
between two parties, patients are in the middle of attack [21]. There may be compro-
mise within the gateway between the sensors and fog node in the smart healthcare
system. A serious problem may arise if the intruder changed the data being handled
by an IoT-based system, this may create serious implications for the welfare of these
patients [50].
There is no standard guide and regulations for computing the protocols and inter-
faces for various products and services in a smart healthcare system. There should
14 J. B. Awotunde et al.

be a standardization put in place to standardize the healthcare system like a dedi-

cated agency is needed to solve this problem. This will help in data dissimilarity
and helps to accomplish the real-time response. For good standardization issues like
communication protocol, data aggregation interfaces, system interfaces, and gateway
interfaces should seriously be considered [51].
There is a broad range of communication protocols, such as WiFi, similar to
the application diversity issue. To be interoperable, the fog node should therefore
perform the required protocol translation on various core layers. Also, there should
be complex regulatory systems before healthcare tools and equipment are available on
the market for consumers used. The stakeholders and end-users of e-health products
should be contacted by developers for their input on their likes, dislikes, and comforts.
This will help to build an acceptable system, interfaces that integrate patient-centric
intelligent medical devices and user-friendly interfaces [51].
Authentication techniques that are also needed in cloud–fog computing are one of
the security services issues, and it is needed in their architectural design. There must
be proper authentication from the end-users or devices before accessing any services
to the receiver end like the server. Since fog devices provided services after the
end-user authentication, fog computing plays an important role; similarly, both end-
user and fog devices must also authenticate by the cloud server before approving
access to resources [52]. The one-way authentication approach is too simple for
a smart healthcare system. However, due to the vast amount of networks, shared
authentication is also equally necessary to avoid cloud masquerade attacks.
The security issues were extremely important to be taken into consideration to
prevent an attacker that wants to impersonate or breaking the sharing authentication
protocol. The fog server will suffer if an attacker gains access to the information
of the legitimate end-user, thus, stop end-users from receiving an original message.
Creating a safe contact within the fog node, end-users, and cloud servers is another
important feature in cloud and fog models. There can be several security issues
when an adversary launches an attack. Therefore, this becomes a difficult research
challenge in the cloud–fog model to develop a robust authentication protocol [52].
One of the security issues in the three-layer architecture cloud–fog computing
paradigm is the man-in-the-middle attack. The attackers trace from the end of the
source and modify it before sending another message to the recipient by crack the
protection framework. This problem was common to various applications like a wire-
less sensor network, cloud computing, smart grid among others. Since an attack can
be launch within fog and cloud servers, protecting the system became an important
security issue [52].
The replay retrieves a message from the sender by the attacker is forwarded to the
recipient to initiate an attack. No useful cryptographic contrivance at a standstill that
can secure the reply attack. The timestamp technique is one of the current ways to
preserve it. However, due to time synchronization issues, it is not ideal for distributed
environments. To protect against reply attack, security protection is therefore neces-
sary. In all cryptographic protocols, the perfect forward secrecy property is extremely
significant. What if the entity’s private key is somehow revealed, the encrypted key
previously identified must not be disclosed to the attacker.
Hybrid Cloud/Fog Environment for Healthcare … 15

When the receiver end receives the message the same as it is sent by the sender,
then the principle of data integrity been confirmed and established. It is also one of the
significant characteristics required in cloud–fog models. The credibility of the data
needs to be provided whenever data needed is been sent from the fog or cloud server.
In the same way, integrity should be maintained either to the fog or the cloud server
when end-users want to send data. The current cryptography techniques called the
hash function like SHA-1, MD5, SHA-2 can provide security and integrity features
when used for safety purposes. Numerous investigators have been working on it and
are seeking to gain solid assets of honesty. The issues of credibility and effective
handling of complexity have remained open research problems in the cloud–fog
computing model.

5 The Future Prospects of Cloud and Fog Computing

Fog computing is predicted to play a prominent role in Tactile Internet, an emerging

technology. The current Internet allows the delivery of content like text, voice, and
video, and the skill sets are distributed over the networks through the Tactile Internet.
The distribution of the skill set will be achieved by haptic communication, implying
the remote control of physical tactile sensations in real time. Telesurgery, telerehabil-
itation, platoons of cars, and virtual reality are some examples of future applications
[11].
The Tactile Internet operates over tactile-human interfaces with a core and edge
domain together with a functional architecture that comprises a likely distributed
master domain hosted by intelligent Tactile Support Engines. The fog application
managed edge works like remote-controlled robots [53]. The major pre-requisite for
effective connection in fog computing is ultra-responsive and ultra-reliable connec-
tivity. An end-to-end latency of 1 ms or less is needed and a maximum of one second
of outrage yearly. The 5G technology has been identified as an enabler of the Tactile
Internet for offering ultra-responsive and ultra-reliable connectivity. The edge and
cloud computing are also an enabler of the said technology [53, 54]. The holistic
approach of fog computing incorporates end-users and IoT-based devices with fog
layer and cloud layer with associated interactions make fog an ideal enabler.
The fog systems can naturally be mapped into the Tactile Internet functional
architecture. The IoT and end-user layer are part of the managed domain like remotely
controlled robots and with a touch, a human–systems interface called the master
domains. The managed domains belong to the layer of end-user devices domains
while the master domains belong to the IoT-based device’s domain. The edges are
the fog layer with the intelligent Tactile Support engines. The cloud can be very
useful when there is a need for large storage and powerful processing. Bring the
architectural and algorithmic together in fog computer are the prospect of Tactile
Internet but their research directions and challenges are beyond the scope of this
chapter.
16 J. B. Awotunde et al.

From the architectural perspective, the design of an ultra-responsive and ultra-

reliable higher layer APIs and protocols for fog device both inter- and intra-layers
communications is an instance of the challenge of fog computing in Tactile Internet.
Both the transport and application protocols operate on top of the physical and MAC
protocols layer required 5G ultra-responsive and ultra-reliable. It will be important
to take into account recent attempts to design novel ultra-high data rates [21]. The
functionality split between the cloud layer and the fog layer is just another example of
a challenge. From the algorithm repositories, the intelligent Tactile Support Engines
in the fog layer can be fed into the cloud layer directly. Their components are located
in both technologies for efficient storage and processing in the cloud layer and harness
mutually proximity in the fog layer.
Also, the combination of cloud and fog computing in building a robust IoT-based
system will help in edifice a strong and fast processing system in a smart healthcare
system. Also, the ultra-responsiveness and ultra-reliability need algorithms running
in the device to guarantee touch applications. The fog system provided an enabling
platform for running various tasks related to Tactile Internet applications in fogs or
cloud layers. There is a need for novel algorithms for task planning due to network
traffic that may occur on the computing nodes with other variables where tasks are
performed and delay sometimes will far surpass the 1 sm threshold. This is necessary
to make sure that activities are carried out within the cumulative threshold and is
not surpassed 1 ms. To have optimal task scheduling, novel machine learning and
artificial intelligence algorithms are required. To predict behavior and reactions, they
may run exclusively in the fog stratum or even in a distributed manner through the
cloud and/or fog strata. The fog layer helps to reduce the total network load during
network connection in an IoT-based platform, and this will help meet the latency
requirement of 1 ms. A various sophisticated algorithms like neural network-based
techniques and simple regression models can be considered in modeling the system
[55, 56].

6 Conclusion and Future Research Directions

The smart healthcare system has the capacity of producing and communicating huge
quantities of data produced from IoT-based devices and sensors. The sizes and quan-
tities of networking equipment are more than ever and the volume of data grows
even more than ever before. For example, IoT-based devices and sensors can be used
to capture and collect patient physiological features and check metrics for patient
diagnosis and monitoring. The emergence of cloud computing has touched almost
all the human life domains; especially, smart healthcare system has greatly benefited
from the paradigm. Cloud computing has brought a technological revolution needed
by IoT-based services like high processing, storage capabilities, heterogeneity, and
computation resources among others. Nevertheless, in IoT-based systems, cloud
computing has weaknesses when it comes to high delays that require real-time
response, and it has a low delay response time, thus not match the industrial control
Hybrid Cloud/Fog Environment for Healthcare … 17

system of the smart healthcare system. The fog computing architecture is heteroge-
neous devices, and geographically distributed ubiquitously connected at the end of
a network to provide collaboratively variable and flexible communication, storage
services, and computation. The new computing paradigm has various advantages in
applications where real time, low latency, and high response time are of utmost impor-
tance like in the case of a smart healthcare system. Therefore, the chapter presents
a general review of technologies of cloud and fog computing in the smart health-
care system. The applicability of both on smart healthcare systems is discussed; the
challenges and the prospects of cloud and fog computing are elaborately discussed.
Fog computing presents better infrastructure by providing low latency, distributed
processing, better security, fault tolerance, and good privacy when compared with the
cloud computing infrastructure. The fog infrastructure sufficiently produces various
fog nodes, virtualized data centers, and edge device networks to connect the IoT-
based devices to implement large storage and rich cloud computing. Fog computing
offered millisecond to sub-second latency faster than real-time interaction, perform
better than cloud computing in low-latency applications, and supports multitenancy
where cloud computing cannot have provided. In future work, the prospects of both
cloud and fog models will be extended in the smart healthcare system to provide a
quick and real-time health diagnosis and monitoring for patients suffering from any
diseases. Fog computing is very necessary majorly because cloud’s recorded health-
care data may be subject to different types of security risks. Finally, the challenges
of cloud and fog computing needed to be looked into and find a lasting solution to
ensure reliable and flexible deployment of cloud and fog computing models in smart
healthcare systems.

References

1. Kaur, P. D., & Chana, I. (2014). Cloud-based intelligent system for delivering health care as a
service. Computer Methods and Programs in Biomedicine, 113(1), 346–359.
2. Fichman, R. G., Kohli, R., & Krishnan, R. (Eds.). (2011). Editorial overview—the role of
information systems in healthcare: Current research and future trends. Information Systems
Research, 22(3), 419–428.
3. Usak, M., Kubiatko, M., Shabbir, M. S., Viktorovna Dudnik, O., Jermsittiparsert, K., &
Rajabion, L. (2020). Health care service delivery based on the Internet of Things: A systematic
and comprehensive study. International Journal of Communication Systems, 33(2), e4179.
4. Abimbola, S., Keelan, S., Everett, M., Casburn, K., Mitchell, M., Burchfield, K., & Martiniuk,
A. (2019). The medium, the message, and the measure: A theory-driven review on the value
of telehealth as a patient-facing digital health innovation. Health Economics Review, 9(1), 21.
5. Adeniyi, E. A., Ogundokun, R. O., & Awotunde, J. B. (2021). IoMT-based wearable body
sensors network healthcare monitoring system. In IoT in healthcare and ambient assisted
living (pp. 103–121). Springer.
6. Romanow, D., Cho, S., & Straub, D. (2012). Editor’s comments: Riding the wave: Past trends
and future directions for health IT research. MIS quarterly, iii–x.
7. Nandyala, C. S., & Kim, H. K. (2016). From cloud to fog and IoT-based real-time U-healthcare
monitoring for smart homes and hospitals. International Journal of Smart Home, 10(2), 187–
196.
18 J. B. Awotunde et al.

8. Palumbo, F., Barsocchi, P., Furfari, F., & Ferro, E. (2013). AAL middleware infrastructure for
green bed activity monitoring. Journal of Sensors.
9. Saad, M. (2018). Fog computing and its role in the Internet of Things: Concept, security and
privacy issues. International Journal of Computers and Applications, 975, 8887.
10. Bonomi, F. (2011, September). Connected vehicles, the internet of things, and fog computing.
In The Eighth ACM International Workshop on Vehicular Inter-Networking (VANET), Las
Vegas, USA (pp. 13–15).
11. Yadav, A., Kumar Singh, V., Kumar Bhoi, A., Marques, G., Garcia-Zapirain, B., & de la Torre
Díez, I. (2020). Wireless body area networks: UWB wearable textile antenna for telemedicine
and mobile health systems. Micromachines, 11(6), 558.
12. Awotunde, J. B., Adeniyi, A. E., Ogundokun, R. O., Ajamu, G. J., & Adebayo, P. O. (2021).
MIoT-Based Big Data Analytics Architecture, Opportunities and Challenges for Enhanced
Telemedicine Systems. Studies in Fuzziness and Soft Computing, 2021, 410, pp. 199–220.
13. Marques, G., Miranda, N., Kumar Bhoi, A., Garcia-Zapirain, B., Hamrioui, S., & de la Torre
Díez, I. (2020). Internet of Things and enhanced living environments: Measuring and mapping
air quality using cyber-physical systems and mobile computing technologies. Sensors, 20(3),
720.
14. Rajabion, L., Shaltooki, A. A., Taghikhah, M., Ghasemi, A., & Badfar, A. (2019). Health-
care broad data analysis techniques: The role of cloud services. International Journal of
Management of Intelligence, 49, 271–289.
15. Ali, O., Shrestha, A., Soar, J., & Wamba, S. F. (2018). Cloud computing-enabled health-
care opportunities, issues, and applications: A systematic review. International Journal of
Information Management, 43, 146–158.
16. Palanisamy, V., & Thirunavukarasu, R. (2019). Implications of big data analytics in developing
healthcare frameworks—A review. Journal of King Saud University-Computer and Information
Sciences, 31(4), 415–425.
17. Aceto, G., Persico, V., & Pescapé, A. (2018). The role of information and communication
technologies in healthcare: Taxonomies, perspectives, and challenges. Journal of Network and
Computer Applications, 107, 125–154.
18. Dash, S., Shakyawar, S. K., Sharma, M., & Kaushik, S. (2019). Big data in healthcare:
Management, analysis, and prospects. Journal of Big Data, 6(1), 54.
19. Ali, O., Soar, J., & Shrestha, A. (2018). Perceived potential for value creation from cloud
computing: A study of the Australian regional government sector. Behaviour and Information
Technology, 37(12), 1157–1176.
20. Senyo, P. K., Addae, E., & Boateng, R. (2018). Cloud computing research: A review of
research themes, frameworks, methods, and future research directions. International Journal
of Information Management, 38(1), 128–139.
21. Oniani, S., Marques, G., Barnovi, S., Pires, I. M., & Bhoi, A. K. (2021). Artificial intelligence for
Internet of Things and enhanced medical systems. In Bio-inspired neurocomputing (pp. 43–59).
Springer.
22. Bayramusta, M., & Nasir, V. A. (2016). A fad or future of IT? A comprehensive literature
review on cloud computing research. International Journal of Information Management, 36(4),
635–644.
23. Ahuja, S. P., Mani, S., & Zambrano, J. (2012). A survey of the state of cloud computing in
healthcare. Network and Communication Technologies, 1(2), 12.
24. Abdulraheem, M., Awotunde, J. B., Jimoh, R. G., & Oladipo, I. D. (2021). An Efficient
Lightweight Cryptographic Algorithm for IoT Security. Communications in Computer and
Information Science, 2021, 1350, pp. 41–53.
25. Culyer, A. J., & Chalkidou, K. (2019). Economic evaluation for health investments En route to
universal health coverage: Cost-benefit analysis or cost-effectiveness analysis? Value in Health,
22(1), 99–103.
26. Elavarasan, R. M., & Pugazhendhi, R. (2020). Restructured society and environment: A review
on potential technological strategies to control the COVID-19 pandemic. Science of the Total
Environment, 138858.
Hybrid Cloud/Fog Environment for Healthcare … 19

27. Mora, N., Grossi, F., Russo, D., Barsocchi, P., Hu, R., Brunschwiler, T., Ciampolini, P.,
et al. (2019). IoT-based home monitoring: supporting practitioners’ assessment by behavioral
analysis. Sensors, 19(14), 3238.
28. Venkatesh, V., Rai, A., Sykes, T. A., & Aljafari, R. (2016). Combating infant mortality in rural
India: Evidence from a field study of eHealth kiosk implementations. MIS Quarterly, 40(2),
353–380.
29. Koufi, V., Malamateniou, F., & Vassilacopoulos, G. (2010, November). Ubiquitous access to
cloud emergency medical services. In Proceedings of the 10th IEEE International Conference
on Information Technology and Applications in Biomedicine (pp. 1–4). IEEE.
30. Sharma, M., & Sehrawat, R. (2020). A hybrid multi-criteria decision-making method for cloud
adoption: Evidence from the healthcare sector. Technology in Society, 101258.
31. Seth, B., Dalal, S., & Kumar, R. (2019). Securing bioinformatics cloud for big data: budding
buzzword or a glance of the future. In Recent advances in computational intelligence (pp. 121–
147). Springer.
32. Aceto, G., Persico, V., & Pescapé, A. (2020). Industry 4.0 and health: Internet of Things, big
data, and cloud computing for healthcare 4.0. Journal of Industrial Information Integration,
18, 100129.
33. Calabrese, B., & Cannataro, M. (2015). Cloud computing in healthcare and biomedicine.
Scalable Computing: Practice and Experience, 16(1), 1–18.
34. Awotunde, J. B., Jimoh, R. G., Oladipo, I. D., & Abdulraheem, M. (2021). Prediction of
Malaria Fever Using Long-Short-Term Memory and Big Data. Communications in Computer
and Information Science, 2021, 1350, pp. 41–53.
35. Navale, V., & Bourne, P. E. (2018). Cloud computing applications for biomedical science: A
perspective. PLoS Computational Biology, 14(6), e1006144.
36. Jin, Z., & Chen, Y. (2015). Telemedicine in the cloud era: Prospects and challenges. IEEE
Pervasive Computing, 14(1), 54–61.
37. Sundwall, D. N., Munger, M. A., Tak, C. R., Walsh, M., & Feehan, M. (2020). Lifetime
prevalence and correlates of patient-perceived medical errors experienced in the US ambulatory
setting: A population-based study. Health Equity, 4(1), 430–437.
38. Maduravoyal, C. (2018). Patient-controlled personal health record enforcing patient privacy in
the cloud-based healthcare system. International Journal of Pure and Applied Mathematics,
119(10), 375–392.
39. Sahoo, P. K., Mohapatra, S. K., & Wu, S. L. (2018). SLA based healthcare big data analysis and
computing in a cloud network. Journal of Parallel and Distributed Computing, 119, 121–135.
40. Regola, N., & Chawla, N. V. (2013). Storing and using health data in a virtual private cloud.
Journal of Medical Internet Research, 15(3), e63.
41. Abatal, A., Khallouki, H., & Bahaj, M. (2018, April). A semantic smart interconnected health-
care system using ontology and cloud computing. In 2018 4th International Conference on
Optimization and Applications (ICOA) (pp. 1–5). IEEE.
42. Schweitzer, E. J. (2012). Reconciliation of the cloud computing model with US federal elec-
tronic health record regulations. Journal of the American Medical Informatics Association,
19(2), 161–165.
43. Chen, S. W., Chiang, D. L., Liu, C. H., Chen, T. S., Lai, F., Wang, H., & Wei, W. (2016).
Confidentiality protection of digital health records in cloud computing. Journal of Medical
Systems, 40(5), 124.
44. Hayes, A. N. (2016). Is cloud computing in healthcare providing a safe environment for storing
protected health information? A systematic review and meta-analysis.
45. Mizani, M. A. (2017). Cloud-based computing. In Key advances in clinical informatics
(pp. 239–255). Academic Press.
46. Yassine, A., Singh, S., & Alamri, A. (2017). Mining human activity patterns from smart home
big data for health care applications. IEEE Access, 5, 13131–13141.
47. Ayo, F. E., Awotunde, J. B., Ogundokun, R. O., Folorunso, S. O., & Adekunle, A. O. (2020). A
decision support system for multi-target disease diagnosis: A bioinformatics approach. Heliyon,
6(3), e03657.
20 J. B. Awotunde et al.

48. Ghanavati, S., Abawajy, J. H., Izadi, D., & Alelaiwi, A. A. (2017). Cloud-assisted IoT-based
health status monitoring framework. Cluster Computing, 20(2), 1843–1853.
49. Oladipo, I.D., Babatunde, A.O., Awotunde, J.B., Abdulraheem, M. (2021). An Improved
Hybridization in the Diagnosis of Diabetes Mellitus Using Selected Computational Intelligence.
Communications in Computer and Information Science, 2021, 1350, pp. 272–285.
50. Rehman, H. U., Khan, A., & Habib, U. (2020). Fog computing for bioinformatics applica-
tions. In Fog computing: Theory and practice (pp. 529–546).
51. Devarajan, M., Subramaniyaswamy, V., Vijayakumar, V., & Ravi, L. (2019). Fog-assisted
personalized healthcare-support system for remote patients with diabetes. Journal of Ambient
Intelligence and Humanized Computing, 10(10), 3747–3760.
52. Kunal, S., Saha, A., & Amin, R. (2019). An overview of cloud-fog computing: Architectures,
applications with security challenges. Security and Privacy, 2(4), e72.
53. Yi, S., Li, C., & Li, Q. (2015, June). A survey of fog computing: Concepts, applications, and
issues. In Proceedings of the 2015 Workshop on Mobile Big Data (pp. 37–42).
54. Mouradian, C., Naboulsi, D., Yangui, S., Glitho, R. H., Morrow, M. J., & Polakos, P. A.
(2017). A comprehensive survey on fog computing: State-of-the-art and research challenges.
IEEE Communications Surveys and Tutorials, 20(1), 416–464.
55. Al-Khafajiy, M., Webster, L., Baker, T., & Waraich, A. (2018, June). Towards fog driven IoT
healthcare: Challenges and framework of fog computing in healthcare. In Proceedings of the
2nd International Conference on Future Networks and Distributed Systems (pp. 1–7).
56. Kumari, A., Tanwar, S., Tyagi, S., & Kumar, N. (2018). Fog computing for Healthcare 4.0
environment: Opportunities and challenges. Computers and Electrical Engineering, 72, 1–13.
57. Ayo, F. E., Ogundokun, R. O., Awotunde, J. B., Adebiyi, M. O., & Adeniyi, A. E. (2020,
July). Severe acne skin disease: A fuzzy-based method for diagnosis. In Lecture notes in
computer science (including subseries Lecture notes in artificial intelligence and lecture notes
in bioinformatics) (Vol. 12254, pp. 320–334). LNCS.
58. Puthal, D., Obaidat, M. S., Nanda, P., Prasad, M., Mohanty, S. P., & Zomaya, A. Y.
(2018). Secure and sustainable load balancing of edge data centers in fog computing. IEEE
Communications Magazine, 56(5), 60–65.
59. Simsek, M., Aijaz, A., Dohler, M., Sachs, J., & Fettweis, G. (2016). 5G-enabled tactile internet.
IEEE Journal on Selected Areas in Communications, 34(3), 460–473.
60. Maier, M., Chowdhury, M., Rimal, B. P., & Van, D. P. (2016). The tactile internet: Vision,
recent progress, and open challenges. IEEE Communications Magazine, 54(5), 138–145.
61. Petrov, I., & Janevski, T. (2016, October). Design of novel 5G transport protocol. In 2016
International Conference on Wireless Networks and Mobile Communications (WINCOM)
(pp. 29–33). IEEE.
62. Sakr, N., Georganas, N. D., Zhao, J., & Shen, X. (2007, July). Motion and force prediction
in haptic media. In 2007 IEEE International Conference on Multimedia and Expo (pp. 2242–
2245). IEEE.
63. Marques, G., Bhoi, A. K., de Albuquerque, V. H. C., K. S., H. (Eds.). (2021). IoT in healthcare
and ambient assisted living. Springer.
Hybrid Intelligent System for Medical
Diagnosis in Health Care

Moolchand Sharma, Akanksha Kochhar, Deepak Gupta, and Jafar Al Zubi

Abstract The autonomous city implies a global vision that incorporates artificial
intelligence, deep learning, big data, decision-making, ICT and the Internet of Things
(IoT) to promote sustainable development. The ageing issue is that researchers,
companies and the government should devote efforts to developing smart health care,
innovative technology and applications. For an extended period, conventional intelli-
gence systems have played a critical role in health care. However, with the increased
popularity and widespread use of these hybrid intelligent computer systems, there
is a significant shift in the healthcare sector. Diagnosis and detection of various
diseases using several techniques can be resolved by using these techniques. Different
novel methods are applied to biomedical engineering to diagnose diseases, and new
models are being studied and compared with the existing technologies. Hybrid intel-
ligence systems can be implied in decision-making, remote monitoring, healthcare
logistics, medical diagnosis, and modern information system. This success’s funda-
mental cause seems to be derived by different intelligent computational mechanisms,
such as genetic algorithms, evolutionary computation, convolutional neural network
(CNN), long short-term memory (LSTM), autoencoders, deep generative models
and deep belief networks. To solve complex problems, we need domain knowledge
that comprises the methodologies that provide hybrid systems with complemen-
tary reasoning and empirical data. This chapter will focus on the need for a hybrid
intelligent system in the healthcare industry and their medical diagnosis applicability.

M. Sharma (B) · A. Kochhar · D. Gupta

Maharaja Agrasen Institute of Technology, MAIT, Delhi (GGSIPU), India
e-mail: [email protected]
A. Kochhar
e-mail: [email protected]
D. Gupta
e-mail: [email protected]
J. A. Zubi
School of Engineering, Al-Balqa Applied University, Jordan, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 21
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_2
22 M. Sharma et al.

Keywords Intelligent system · Deep learning · Genetic algorithms · Neural

networks · Hybrid intelligence system · Computational intelligence

1 Introduction

For medical applications and intelligent computerized system, research is a signifi-

cant and thrilling area. Based on the symptoms, a general doctor accumulates typi-
cally his/her insight dependent on patients’ history and affirms by diagnosing them.
Thus, a physician’s experience is vital for diagnostic accuracy and predictive impor-
tance of symptoms towards certain diseases. As rapid strides have been made in the
medical field and therapy of that particular cause, e.g. the evolution of new diseases
and drug accessibility, it has become stimulating/demanding for a doctor to stay
updated with all the current knowledge developments clinical practice [1]. Addi-
tionally, with the development of computer technologies, it is easy to gather and
store a massive amount of information in the digital form, e.g. an exclusive database
of patient records in electronic form [2]. In such a scenario, applying an electronic
medical decision support system is a feasible approach that would assist physicians
so that patients’ diagnosis can be made quickly and efficiently [3]. However, several
problems have to be solved before an efficient/effective medical support system, i.e.
decision support system (DSS), can be formulated and applied. It includes result-
making in the presence of multiple factors, like inaccuracy and uncertainty [4].
Though it is important in terms of expert’s knowledge and experience, which ranges
from the assessment of a patient’s condition so that a diagnosis can be made, develop-
ments in machine learning techniques have resulted in paving the way for medicinal
physicians to use computerized hybrid systems in their field, e.g. medical imagery
and X-ray photography [5]. A physician must conclude the expected disease to the
root cause among the list of possible causes with similar symptoms by applying
his/her experience and skill and then confirming the disease using multiple tests [6].
Simultaneously, hi-tech intelligent systems can help physicians make rational and
well-informed inferences quickly, e.g. by imbibing past experiences from electronic
patient records and making an inference regarding the diagnosis of a present patient
proper rationale. The advantage of using intelligent systems is accurate in detecting,
diagnosing, and diminishing time and expenses related to patient treatment [7].
The primary objective of developing/formulating machine learning models is to
support various decision-making tasks in medicine. For example, intelligent clas-
sifiers are used for diagnosis, prognosis and airing of diabetes, breast cancer and
Parkinson’s disease [8]. The capability of fuzzy neural models to learn from the facts
(i.e. patient records) and generalizing the process beyond the training samples makes
them perfect for classifiers’ task to detect any disease [9]. It includes fuzzy learning
vector quantization networks, vague neural networks and vague probabilistic neural
networks [10]. However, they are hit by a limitation, i.e. lack of these models’ ability
to explain their predictions [11]. This is the inspiration for this study. We try to invent
Hybrid Intelligent System for Medical Diagnosis in Health Care 23

the machine knowledge-based system that can reflect its reasoning for handling an
input case and providing the rationale behind its forecasts.
This chapter has discussed the hybrid intelligent model, which has two real-world
influences in medicinal decision support. First, the capability to provide rationale and
validation for prediction is the most important to convince medical practitioners that
computerized decision support can be efficient. This skill is of prime importance in
security applications, such as medical diagnosis and prognosis, whereby practitioners
need to understand and form a conclusion regarding the conclusion-forming process
used by a computerized system to make accurate predictions [12]. It can become
a source of second opinions in situations involving medical diagnosis. The rules in
the form of a decision tree from the hybrid model are of paramount importance in
practice [13]. Secondly, it is critical for a decision support system to have a high
accuracy in medical applications. As stated before, an elevated false-negative rate
would result in increased risk for patients as they would be deprived of obtaining
essential facilities medical.
In contrast, a highly inaccurate alarm rate would result in unnecessary stress and
anxiety in patients and increased medical resources pressure due to huge demand.
Nevertheless, a DSS with the right positive rate and accurate negative rate results in
lower therapeutic costs and vestibular patients’ development in an otolaryngology
clinic [14]. Also, it is a fact that machine learning models are useful in minimizing
cost and time for medical diagnosis [15].

1.1 Intelligent Systems

What is Intelligence?
A machine can learn from training data, compute, reason, perceive relationships
and analogies, store and retrieve information from memory, solve problems, compre-
hend complex ideas, use the natural language fluently, classify, generalize, and adapt
to new situations.
Types of Intelligence
A. Linguistic Intelligence:

The capability to talk, recognize and use mechanisms of descriptive linguistics

(speech sounds), syntax (grammar) and linguistics (meaning), e.g. Narrators, Orators.
B. Musical Intelligence:

The capability to make, communicate with and perceive meanings made from sound,
understanding of pitch, rhythm, e.g. Musicians, Singers, Composers.
C. Logical-mathematical Intelligence:
24 M. Sharma et al.

The capability to use and perceive relationships within the absence of action
or objects. Understanding advanced and abstract concepts, e.g. Geometricians,
Researchers.
D. Spatial Intelligence:

The capability to understand visual or spacial info, modify it, recreate visual pictures,
construct 3D pictures, and manoeuvre and rotate them, e.g. Map readers, Astronauts,
Physicists.
E. Bodily Kinesthetic Intelligence:

The willingness to use a whole or part of the body to solve problems or fashion items,
handle fine and coarse motor skills, and control objects, e.g. Singers, Players.
F. Intra-personal Intelligence:

The capability to distinguish among one’s feelings, intentions and motivations, e.g.
Gautam Buddha.
G. Interpersonal Intelligence:

The capability to identify and make distinctions among other people’s feelings,
beliefs and intentions, e.g. Mass Communicators, Interviewers.
We should know at least what does intelligence composed of?
The intelligence is impalpable. It is composed of different attributes as mentioned
below and also shown in Fig. 1.
i. Intellectual, i.e. reasoning
ii. Knowledge
iii. Problem-solving
iv. Observation, i.e. perception

Fig. 1 Classification of intelligence

Hybrid Intelligent System for Medical Diagnosis in Health Care 25

v. Linguistic intelligence.
Let us quickly go through all the components.
A. Intellectual—Two types permit us to provide the basis for judgement, making
decision and prediction.
B. Knowledge—By learning, practising, being taught or experiencing something,
we can acquire knowledge and skills.

Knowledge improves the consciousness of the themes of the study.

The capability of knowledge is possessed by different categories like humans, some
animals and AI-enabled systems. Learning/knowledge is categorized as—
i. Sensory System Knowledge—Students are listening to recorded audio
lectures using listening and hearing.
ii. Episodic Knowledge—To learn by remembering the order of events that you
have heard or seen. It is ordered and linear.
iii. Motor Education—It is the method of muscles’ precise movement, for
example, object picking, writing, etc.
iv. Data-Based Knowledge—It is the process of watching others and receiving
information. The infant, for instance, tries to learn by imitating its parent.
v. Sensory Activity Knowledge—It is the mechanism by which stimuli that one
has seen before can be recognized. They are identifying, for example, and
classifying objects and circumstances.
vi. Relational learning requires learning, rather than simple properties, to distin-
guish between various stimuli based on relational properties. For Example,
adding a very small amount of salt while cooking potatoes that came up salty
last time, when cooked with adding, say, a tablespoon of salt.
vii. Abstraction Learning—Via visual stimuli like pictures, colours, maps, etc., it
is learning. An individual may build a roadmap in mind, for instance, before
actually following the route.
viii. Stimulus–Response Learning—When a particular stimulus is present, it is the
process of learning to conduct specific behaviour. For example, a dog becomes
cautious by raising its ear when it hears the doorbell.
C. Problem-Solving—Based on past movements, it is the phase in which one
sees and observes can attempt to overcome the present situation, blocked by
known or unknown barriers. Problem-solving often requires decision-making,
choosing the most appropriate option from various solutions to achieve the
ultimate goal.
E. Observation—It is the method of sensory knowledge processing, percep-
tion, collection and organization. Perception presumes sensitivity—sensory
organs of humans aid vision. The sensor method incorporates the data obtained
by the sensors in a practical way within the AI domain. It is the method
of sensory knowledge processing, perception, collection and organization.
Perception presumes sensitivity—sensory organs of humans aid vision. The
26 M. Sharma et al.

sensor method incorporates the data obtained by the sensors in a practical way
within the AI domain.
F. Philological Intelligence—The study of oral language and the spoken and
written language of speech and writing. In interpersonal communication, it is
important.

1.2 Hybrid Intelligent System

Intelligent systems are hybrid and are considered the most important research field for
computing intelligence in modern ways and only concerned with developing intel-
ligent systems for the next generation. To investigate the hybrid systems, a stimulus
must be the spread of awareness in the communities leading with academics which
combines different methods to solve the problems related to artificial intelligence
(AI). The combination of different learning methods and adaption is required to over-
come the drawbacks to achieve synergetic effects with a combination of two or more
than two computational methods or inclusion of techniques that have contributed to
the intelligent systems that are newly designed. These approaches have followed an
ad hoc method which may not certainly be used. Different frameworks are made for
modelling expertise in the current ages: one is soft computing approaches, support
of decisions, segmentation techniques used for images and videos, process control,
robotics, automation, etc [16]. Most of the techniques use various information acqui-
sition schemes, models for decision-making, and learning strategies to solve the
computation problem. This helps to overcome the drawbacks of one is through more
than one methods or a combination of different methods. These notions lead to
different emerging architectures of intelligent systems [17]. We are familiar with
the fact that intellectual systems provide human-like information such as domain
data, indeterminate reasoning, and a noisy and time-variable environment to handle
computing problems. The most important phase is designing the hybrid systems as
the main aim is to combine and interact with various techniques and emerge with a
new working technique. The properly known methods can be applied to a specific
problem, and that problem can be solved within the system. If a system is faced with
some problem, then the drawbacks must be addressed with traditional systems. So,
to develop a hybrid system will lead to an evolution in the traditional systems.
Hybrid intelligent architecture classification is carried out in four separate groups
depending on the system’s overall success, namely stand-alone, transformative, hier-
archical hybrid and integrated hybrid system [18]. Fused architectures are the first
proper type of interconnected intelligent systems. They have structures that combine
a single machine model with various techniques. They swap systems with data and
information representations. Another method is to place the different strategies side-
by-side in a problem-solving task and rely on their interaction. Robustness, higher
performance and increased problem-solving skills gain from integrated models.
Finally, a broad range of features, including adaptation, widespread use, noise toler-
ance and rationale, can be supported by fully integrated models. The two architectures
Hybrid Intelligent System for Medical Diagnosis in Health Care 27

belong to the integrated model in this chapter. Combining neural network learning
with evolutionary computation, Sect. 1.2.1 presents the hybrid mechanism for the
inference system’s refinement. In this section, an example of an application is also
given. We also addressed a hybrid compound of the fuzzy clustering algorithm in
Sect. 1.2.2 and a fuzzy inference method for a web mining task.

1.2.1 Adaption of Fuzzy Inference Systems

In traditional times, the fuzzy inference method uses the expert model, which defines
the process’s essential properties. Expert expertise is the key source of the design of
the fuzzy inference system. If the performance indicator is taken into account, it is
appropriate to adopt the following, such as the membership function, the knowledge
base, the inference process. It takes a lot of research work to adapt to the fuzzy
inference method [19]. It involves the adaptation of the functions of membership,
rule bases and aggregation operators. They contain but are not restricted to:
• It is well known that the self-organizing fuzzy controller proposed by Procyk
considers the problem of rule generation and adaptation [20].
• The gradient descent and the variants applied to alter the member functions’ input
and output parameters [21].
• Cutting the amount and adjusting the state of input/output participation capacities
[22].
• Apparatuses to distinguish the models of fuzzy models.
• The fuzzy principles’ hypothesis generally applies to the min and max opera-
tors for fuzzy intersections and union. If T-norm and T-conorm operators are
parameters, the gradient descent technique can fine-tune the fuzzy operators in a
supervised learning setting.
The local fuzzy region defines the antecedent of fuzzy rule, while a consequent
shows the region’s behaviour through various constituents. It comprises a linear equa-
tion or a membership function [23]. The fuzzy inference systems are adapted using
techniques that are evolutionarily computed and have been explored widely. This
adaption of relationship function is known as self-tuning. Adaptation of fuzzy deriva-
tion frameworks utilizing transformative calculation strategies has been broadly
investigated [24]. The programmed adjustment of participation capacities is preva-
lently called self-tuning. The genome encodes parameters of the trapezoidal, triangle,
calculated, hyperbolic-digression, Gaussian enrollment works, etc [25, 26]. Fuzzy
rules by using the evolutionary search can be performed using three approaches.
• The foremost approach is the Michigan approach, where the fuzzy knowledge
base is considered, and it results in the antagonistic rules to lead competition and
cooperation with the fuzzy rules. To represent a fuzzy rule, the use of genotype
is done, and the solution is considered out of the entire population.
28 M. Sharma et al.

• Another method could be the Pittsburgh approach which evolves knowledge from
the population instead of applying fuzzy rules to the individual. The new collab-
oration of rules is done in order to serve the genetic operators. It may suffer from
a drawback: the complexity is increased of search space, and the computation
burden is also overhead, mostly when online learning is done.
• The third method is the iterative knowledge approach, similar to the primary
method, where every chromosome represents a single rule. However, in diver-
gence, to the first approach, only the chromosome which is superlative will form
the solution, and the left ones are discarded [27].
The developmental learning process has developed to finish the standard base
using an iterative learning process. Utilizing the neuro-fuzzy model, no affirmation
is given that the neural system calculation will join, and for the fuzzy framework to
be fruitful, the tuning is to be done. It incorporates the assurance of ideal parameter
estimations of the participation capacities, procedure on fuzzy, etc. One of the devel-
opmental frameworks’ kind elements is that they are versatile to any condition [28].
The test confirmations demonstrate that developmental calculations are wasteful for
tuning arrangements. However, they are viewed as proficient when finding global
basins of attraction [27, 29]. The efficacy in evolutionary is enhanced if a limited
search process is considered. These calculations first discover an area in the space
and afterwards take care of the nearby looking through a technique to locate an ideal
arrangement. To locate a decent beginning parameter esteem by finding the great
locale in space is truly fascinating. To characterize the bowl of fascination for the
neighbourhood least worth includes all the focuses, parameter esteems that may join
to a nearby least using a calculation lastly finding a worldwide least incentive with
the assistance of a neighbourhood search calculation.
Referring to Fig. 2, tt1 and tt2 are considered the original parameter values cited by
the evolutionary algorithms, and WA and WB correspond to the concluding parameter
that tweaks the technique of meta-learning. Now the model which we present is
the evolving neural fuzzy model. The function of this model is that it optimizes the
inference system of fuzzy by using a metaheuristic approach that combines the neural

Fig. 2 Fine-tuning using

hybrid learning parameters
Hybrid Intelligent System for Medical Diagnosis in Health Care 29

Fig. 3 EVoNF general computational framework

networks with evolutionary computation. The proposed technique is a methodology

that can integrate various neural networks and fuzzy inference systems, along with
the evolutionary search process [24, 30]. Figure 2 shows the connection component
with the evolutionary search, which advances at the best level with the time scale
at the slowest pace. For each developmental search calculation that supports fuzzy
operation, it requires the quest for the fuzzy rule. It requires a short timescale to
choose the induction framework and the issue proclamation.
Therefore, the fuzzy interpretation systems have evolved at a slow rate, whereas
the evolution of the quantity and the membership type is evolving faster. Other layers
also provide some functions. The adaptation layers form a hierarchy that depends on
previous knowledge. If there is complete knowledge about the knowledge base, then
the inference mechanism will be better implemented and acceptable. If some fuzzy
inference system is considered best to solve a problem, then the computational task
will be minimized as the search space will be reduced. The chromosome architecture
is shown in Fig. 3. Figure 4 shows the engineering architecture, and the component
that evolves can be considered by way of a basis for the adaptive fuzzy systems.
Every layer of the hierarchical evolutionary search procedure is characterized by
a chromosome used for efficient modelling of EVoNF.
The comprehensive working and demonstrating procedure is considered by way
of:
Layer 1: It is the easiest way to encode each input variable’s membership functions
and parameters for every membership function. Figure 5 shown represents the
chromosome of a membership function represented by the parameters represented
by the variables. The minimal constraints of membership functions are located
in the evolutionary algorithm, which is tuned by the network’s flow process. To
represent a Mamdani fuzzy inference system, a similar approach is used to output
the membership function. The master’s recommendation is to evaluate the MF
shape and its structures parameters to gauge the hunt space. The precise coding
30 M. Sharma et al.

Fig. 4 Chromosome structure of EVoNF model

Fig. 5 Representation of n membership functions of a bell-shaped MF

strategy proposed by Cord’on is considered to speak to the resulting parameters

of the Takagi–Sugeno derivation framework [19].
Layer 2: In this layer, the optimization of the rule base is performed, which
comprises the overall sum of rules, the representations of the parts that is the
antecedent and the consequent. The rules increase drastically, increasing the
number of variables and the fuzzy sets based on the representation. The easiest
method is representing a gene that characterizes a single rule where ‘1’ is used
for selected, and ‘0’ is used for rejection. Figure 6, shown below, represents the
structure of a chromosome. To show the single rule, a position-dependent code
is used with all the elements by way of the system’s number of variables. Each
component is a parallel string that is a piece in the fuzzy set, which shows the

Fig. 6 Representation of the entire rule base consisting of m fuzzy rules

Hybrid Intelligent System for Medical Diagnosis in Health Care 31

Fig. 7 Depiction of an individual fuzzy rule

nonappearance or nearness of the relating semantic name utilized in the standard.

On the off chance that we think about three information and one yield variable,
with fluffy allotments, contains 3, 2, 2 fluffy sets for input factors and three fuzzy
sets for yield variable, the fuzzy guideline will have an illustration as presented
in Fig. 7.
Layer 3: This layer represents a chromosome with a number of the same T-norm
and T-conorm operator parameters. Representation in real numbers is considered
appropriate to characterize the fuzzy operator parameters. These parameters are
refined by techniques called gradient descent.
Layer 4: The responsibility of this layer is to select the minimum learning param-
eters. The gradient descent strategies are known for their exhibition that relies
legitimately upon the learning rate whenever contrasted with the error surface.
Representation in real numbers is used for the representation of learning parame-
ters. The evolutionary algorithms decide the learning parameters, which are further
used to refine the membership functions and the method used for inference rules.
Layer 5: Interaction with the environment is done in this layer which further
decides the inference system and is considered minimal according to the environ-
ment. When the representation of the chromosome is completed in the EVoNF
model, the evolutionary search procedure can be started as follows:

Step 1: Initially, the initial population’s generation is done in N no’s of C chro-

mosomes. The fitness function is calculated for every chromosome, but it entirely
depends on the problem.
Step 2: Once the fitness function is calculated, the selection method is used to produce
the children for every individual from the current population.
Step 3: The genetic operators are applied to every individual from the populace to
generate the next generation.
Step 4: If the present model is completed, then the error rate is calculated.
Step 5: End.

1.3 Health Care

Through prevention, diagnosis and treatment of illness, sickness, impairment and

other physical and mental conditions, health care means maintaining or improving
human health. Allied health professionals provide health treatment. Physicians
32 M. Sharma et al.

and physician assistants comprise these health professionals. Dentistry, midwifery,

nursing, medicine, optometry, audiology, pharmacy, counselling, occupational
therapy, physical therapy and other healthcare professions are included in all health-
care systems. This includes jobs in primary care, secondary care, tertiary care and
public health sectors. Access to health care, mainly influenced by social, economic
and health policies, can vary across nations, communities and individuals. ‘The use of
personal health care on time for the best health outcomes’ [31] means the availability
of health services. There are hardly any elements that should be considered to enter
the human services offices, including monetary restrictions such as security inclusion
and geographical barriers (travel costs). The possibility of taking care of the time of
work to use administrations and individual impediments such as the inability to speak
to social insurance companies, unforeseen health treatment and treatments’ efficacy.
The overall performance is adversely affected by healthcare constraints (well-being,
mortality rates). Health facilities are businesses that are structured to meet individual
communities’ health needs. According to the WHO, a well-functioning healthcare
system requires a financing facility, well-trained and adequately paid personnel,
sound policy and decision-making expertise, and well-maintained health facilities to
provide quality medicines and technology.

2 Need for Health Care’s Intelligent Infrastructure

for Medical Diagnosis

The development in the field of intelligent systems has been improved. It has shown
excellent results in ability and reliability by using different techniques for computer-
supported medical diagnosis. The fundamental justification for using the machine
is this. In massive disease prevention, including cancers: breast cancer, melanoma,
etc., medical treatment is only successful if the disease is diagnosed in the early
stages [32]. Many different technologies are now available, and the question is
how many laboratory examinations will diagnose them. New steps can be taken
towards a computer-supported medical diagnosis to implement learning methods
based on various mathematical theories. This can be beneficial, particularly in situa-
tions where there are inadequate experts, and the critical element is rapid diagnosis.
This chapter describes an innovative method of classification capable of handling
cases with limited data sets and nonlinear reciprocal relationships.

2.1 Basic Algorithm

A small portion of training data called support vectors (SV) is the algorithm’s concept,
which is similar to an optimal separation of the entire data set. Suppose optimum
separation means that the minimum distance between the two groups for the closest
Hybrid Intelligent System for Medical Diagnosis in Health Care 33

point is maximum from the separating hyperplane. If the hyperplane’s function is

y = m Xi + c, then finding the optimal hyperplane means minimizing ||m||2 . For
nonlinear tasks such as regression, recognition, regulation, fluid system modelling,
the nonlinear support vector approach is used. To map the data to a higher space
where linear regression is feasible when nonlinear mapping is used. When the kernel
functions are introduced, this change is possible. They come from the Hilbert Spaces
kernel replication theory [33, 34].

2.2 Applications in Diagnosis

The wrong diagnosis can lead to a significant threat to the quality and safety of
health care. The rates of ambulatory diagnostic errors are projected to be 5.08%
in the USA, equivalent to 12 million adults per year. Approximately, fifty percent
of these errors may be detrimental. Because of the high volume of medical image
data, hybrid technology was used to improve medical diagnosis quality, particularly
in radiology [35]. Keith Dreyer, a radiologist at the Harvard Medical School, said
that “Meaningful technology would enhance consistency, productivity, and perfor-
mance”. Esteva et al. trained deep convolutionary neural networks (CNNs) based
on 129,450 clinical images for skin cancer diagnostics. The findings show that this
method can classify skin cancer at a level comparable to dermatologists. You thought
that smartphones could be a cheap way to help dermatologists broaden their scope
to improve their access to diagnostic treatment [36].
Liu of Google, Inc. has announced the CNN framework for the pathological
detection of lymph node breast cancer metastasis. The findings showed that this
technology would improve the tempo, precision and uniformity of diagnosis and
decrease the false-negative rate to one-fourth of that for human pathologists. AI tech-
nology has been extensively adopted into numerous areas of diagnosis. For instance,
recently published a cardiac motion magnetic resonance imaging algorithm to fore-
cast patients’ pulmonary hypertension outcomes accurately. In the study of electro-
cardiograms in chronically ill patients, Moss et al. employed an automatic rhythm
classification approach. They concluded that AI technology provided new informa-
tion and data observations that physicists could have ignored. Lee et al. have also
addressed promising outcomes from recent studies using AI in stroke imaging and
suggested that AI technology may play a key role in stroke patient treated with an
individualized strategy.
Current health systems focus on treatment-based medicine and do not provide
safe patients at high risk of reliable, low-cost treatments. At the same time, a vast
global economic burden has been created by the pandemic. In the USA, dissatisfac-
tion with the healthcare system has prompted the government to consider incorpo-
rating AI personal health monitoring platforms in healthcare management systems,
decreasing healthcare costs and strengthening health quality. AiCure, a mobile appli-
cation funded by the United States National Institutes of Health, was created to
34 M. Sharma et al.

monitor patients’ medications and conditions. Because of advancements in insen-

sitivity, over-use in health care is a rising issue attributable to medical detection
methods. It is possible to diagnose up to one-third of cancer diagnoses identified
through screening, and 30% of individuals diagnosed with asthma do not have the
condition. Testing and overtreatment are also standard; more than 50% of tests were
needless before cataract surgery, for example, and were likely to worsen their compli-
cations. In particular, in developed nations, antibiotics prescription is still a prevalent
issue due to insufficient health staff training. In health management systems, medical
AI technologies can help to detect needless diagnostics and procedures. Therefore for
large-scale enterprises, we need some hybrid technology that works on controlling
health services.
Finally, we concluded that the new advances in the intelligence sector had taken
place. The devices also dramatically increased the capability and reliability of
numerous medical diagnostic techniques. There may be occasions where the issue has
been linearized, adding to the reliability of the diagnosis. For nonlinear modelling,
different methodologies are being developed. CNN and ANN have been used in
numerous applications and have demonstrated that the approaches are considered
adaptive to the initialization of the number and forms of neurons and weights. So for
excellent precision, the data set is mandatory. This is also not available for well-being
since it takes a long time to create such an extensive database.

3 Hybrid Intelligent Medical Diagnosis System

Biomedical engineering is a fast-growing field because of the needs and increases in

automation. Collaboration between medical professionals and engineers is required
to build smart systems for various bio-medicine tasks. In order to detect various
diseases, these instruments are used. To help doctors understand diseases’ presence,
these tasks serve as clinical decision support systems (CDSS). They, therefore, act as
valuable instruments in disease research for physicians. This is especially important
given the workload on doctors and the massive prevalence of diseases. The increasing
awareness of health has further emphasized the production and use of such systems
among people. We use soft computing techniques for disease detection here. This is
a classification issue where the aim is to classify the parameters in one of the two
groups that indicate whether the disease is present or absent.
In order to execute this classification, the device is given several inputs or
attributes. Suppose these characteristics are displayed graphically in each class on
a graph with a different symbol. There are as many attributes to this graph as the
sum of input space. The classification issue’s primary task is to fragment the input
space’s distinctive areas so that each class is part of a particular section. The borders
of the class are known as judgement boundaries. Based on data or information, every
system tries to calculate or predict these constraints. As they have their problems and
complications, the classification problems have always merited special mention from
the scientific community. One of the curious things about these topics is that while
Hybrid Intelligent System for Medical Diagnosis in Health Care 35

the mechanism and means of classification have been revolutionized, compared to

accuracy, the artificial systems seem to be well behind the human classification. A
corresponding vocabulary is a sequence corresponding to some particular pattern
or attribute’s capacity to distinguish. Machine learning is an exciting environment
for researching historical records. In most of the issues, a lot of background data is
available. The systems are developed using system-specific training algorithms to
learn this data. Learning means removing laws or patterns from historical records.
Well-trained programmes can deliver the correct outcomes for the concerns they have
been trained for.
In comparison, the time and memory constraints would be slightly smaller since
the device already has historical data summarized by specific patterns or regula-
tions. Generalization is the tendency of the method to generate the correct outcomes
for unknown problems that are identical. This happens by imposing the system’s
undefined inputs with extracted patterns or rules. If it is somewhat generalized, a
process is considered efficient. In comparison, the time and memory constraints
would be slightly smaller since the device already has historical data summarized
by specific patterns or regulations. Generalization is the tendency of the method to
generate the correct outcomes for unknown problems that are identical. This happens
by imposing the system’s undefined inputs with extracted patterns or rules. If it is
somewhat generalized, a process is considered efficient.
The field that works here is soft computing that comprises the ANN, evolutionary
systems and fuzzy inference systems. One of the networks considered to be moti-
vation from the human brain is an artificial neural network [37]. The brain includes
an enormous number of corresponding processing components known as neurons.
The data is received by each neuron by way of electrical signs, forms the data and
further transmits the information to prepare each other neuron. The neural system
includes various counterfeit structured neurons that work in an incrusted style. Each
neuron gets the total sum of the approaching data and then moves it by an initia-
tion work before contributing it to the following neuron. By and large, we utilize
three-layered engineering, which involves a passive input layer, a hidden layer and
an output layer. The ANNs structure a decent taking in implies from the critical infor-
mation (or AI) and sum up the patterns which are found out into the obscure data
sources. These systems additionally experience two different degrees of preparing
and testing. One of the usually utilized methods for preparing the ANNs is the back-
propagation algorithm (BPA). It is a managed learning model where the outputs are
known for each information when the training is performed. BPA consists of two
different passes, which are the forward pass and the reverse pass. The forward pass
is a customary neural network where the data sources are used before assessing the
comparing yields for the accompanying. The purposeful publicity of any learning
calculation is to locate the worldwide optima in the inquiry space.
Another method for adequately addressing the issue incorporates fuzzy surmising
frameworks (FIS) [38]. To demonstrate the issue, we need to utilize a fuzzy set
hypothesis. According to the fuzzy set hypothesis, each component has a place with
some class by a specific degree known as the participation esteem. The information
included is taken from the class, which is related somewhat, and there must be a
36 M. Sharma et al.

class of degree that does not have a place with the class. The framework can map the
information sources powerfully to deliver yield by the fuzzy rules applications. Fuzzy
rules are considered a stimulus compared to the ancient production systems that use
the rules to be used on the state to adopt the final state or output. The serious issue that
slacks with fuzzy rules are that it works on people’s parameters and rules. The fixed
human characteristics are slanted not to be right or dangerous. These frameworks
should be physically prepared and tuned, which is not frequently practical for all the
issues. This spot a major restriction over the viable utilization of these frameworks
in down to earth applications. The constant difference in rules, fixing of framework
parameters, and so forth to get the planned framework to proceed according to the
excellent yields might be exceptionally troublesome physically.
The algorithms that are considered the best for optimizing and searching for the
problems are the evolutionary algorithms (EA) [39]. These algorithms are considered
nature-inspired. The population is considered the basis on which the evolution of life
surrounds, so there is a need to generate the next generations. The fittest one will
survive according to Darvin’s theory, and the fitter one will evolve and move to the
next generation. The reproduction of the individuals will be more robust who are fitter
than the one which is the weaker. The various solutions to the individual’s problem
are modelled in the evolutionary algorithms. The new population of individuals
are better and optimal compared to the population of the old one. So, the optimal
solution to the problem is finding the individual who is considered the best. Different
evolutionary operators are motivating by nature and are considered better to create the
youngster populace’s ideal populace from the parent’s populace. These calculations
are considered powerful to scan for the worldwide minima, which are limited and
constrained. This issue can be measured by using these frameworks.
If the knowledge size is thought to be immense and the system is considered very
complex, the problems are significant. These systems are often considered fragile,
and a single mechanism is preferably unable to solve the problems. The method is
then implemented to combine the system’s aspects positively, and one of the solutions
may be called to eliminate the negative aspects. Harmful elements may be obsolete
by the positive component of the scheme. We are concerned with the first form of
implementation in Sect. 3.1, i.e. ANFIS. In Sect. 3.2, we address numerous modern
methods and new assembly techniques. In evolutionary neural networks, Sect. 3.3
is devoted to using a connectionist approach. We will define the problem-solving
technique along with the database in Sect. 3.4. In this chapter, with the assistance of
an intellectual device, we explain the different strategies used to diagnose diseases.

3.1 Adaptive Neuro-Fuzzy Inference Systems (ANFIS)

Adaptive neuro-fuzzy inference systems (ANFIS), a hybrid framework considered

the method of problem-solving, is believed to be the first approach [40]. It is possible
to view this method as the integration of ANN with FIS. The fuzzy inference method
(FIS), which will be constructed over the neural network design, is also called ANFIS.
Hybrid Intelligent System for Medical Diagnosis in Health Care 37

Fig. 8 ANFIS layered

architecture

This will allow us to use the algorithms used in FIS training for a historical database.
It will tune FIS and sees it as capable of behaving according to the problem’s demands
and providing planned outcomes to issues if it happens to be exposed to them. Mainly,
it leads to minimizing errors and achieving excellent problem-solving results. In
traditional FIS, automatic tuning of the fuzzy parameters is also an issue that is
performed manually. The structure overall is shown in Fig. 8. It is a neural system
model with several layers. In both ways, for example, the whole architecture can
now be seen as a neural system that has nodes organized in different layers and each
layer is relegated to a specific errand to execute. Another FIS will carry out fuzzy
activities to measure the yields according to the sources of knowledge.
To solve the problem in the above system, as in figure shown (i.e. Fig. 8), it consists
of steps. The underlying advance is making an underlying fuzzy model according to
the prerequisites of the issue. This may be done physically by the client according
to his comprehension of the framework, or consequently, where all the mix of rules
might be produced. The main fuzzy model would comprise many participation capac-
ities for each info variable. The quantity of enrolment work assumes an indispensable
job in choosing the summing up capacity of the framework. The littler the number
of membership functions, the bigger would be the summing up ability. The higher
number of part transport capacities makes the issue increasingly confined in nature.
It might be noted here that summed up frameworks attempt to discover exceptionally
broad principles that fit in the whole arrangement of preparing information. This is
opposed to the limited frameworks attempt to outline the standards for some piece
of the database. The entire preparing data set might be said as made out of numerous
such sets. Speculation is viewed as explicit to the problem as well. All issues cannot
be tackled from the most extraordinary level of speculation.
38 M. Sharma et al.

We have to restrict it by using a planning estimate after the model has been
constrained. For this, we will use the authentic database of known sources of knowl-
edge and yields. This is a type of supervised method of learning used by ANFIS.
Two frequently applied learning algorithms are the back-propagation algorithm and
the mixed learning algorithm. The back-propagation algorithm is the same as that
described in Sect. 3.1. This attempts to train the algorithm through the gradient
descent of the errors in the error space. The test requires providing unknown inputs
to the algorithm and separating the data according to the training performed. This
stage determines the algorithm’s efficiency.

3.2 Ensemble Approaches

The ensemble is carried out as the second hybrid approach or process that can be
used for disease detection and diagnosis. A sort of modular neural network (MNN)
[40] is the ensemble. The MNN expects to leverage the modularity of the problem by
categorizing it into another group of modules. These modules function independently
and lead to a complex problem being solved. The old ANN suffers from issues that
result in execution dissatisfaction when the planning database is extremely large or
the issue is particularly muddled. It hampers their success and also contributes to
very long training cycles. MNNs can solve the problem by modularity in the MNNs.
The first step to solve an issue is to divide the problem via MNNs into different
modules. This splitting takes the same time as the system takes itself while being
created. When the issue is given to the unit, it is divided first into the various modules.
Ensembles, particularly for such classifying problems, are considered a widely used
type of MNNs [41]. In this section, we will only discuss the basic model that is
being used for problem-solving. There is also no mention of other approaches and
techniques. The first step is to train the machine. All the training data in the system is
trained by each module or the ANN independently. Both ANNs must have planned in
parallel. Here, each ANN’s output is regarded as the probability vector that denotes
each class’s presence or absence. The system states that there are N categories.
These modules are researching the issue and working on a response. At that point,
they return a probability vector containing probability events for each class. Each
party restores the probability vector where each pi indicates the likelihood of class
I being present if the classification problem had n classes for which the data might
have a location. The case may not be the same because of the system’s weakness but
differs if ideally only one out of the considerable number of numbers right now be
one, and all others must have −1.
According to the principles of the MNNs, the various types of probability vectors
are given to the focal integrator. The option of the system’s last yield begins when
the integrator gets all the likelihood vectors. The first assignment is to combine the
probabilities of the different likely vectors. Henceforth, this represents the last vector
of probability that consolidates a large number of modules’ aftereffects. Because of
a portion in which the integrator collects the modules’ reactions to the problem to get
Hybrid Intelligent System for Medical Diagnosis in Health Care 39

Fig. 9 Probability-based approach using ensembling

the expected reaction. The integrator selects the class at that stage that gets the most
important reaction or has the most serious possibility. The class will be returned to
the last class. The notion appeared in Fig. 9.
Parallel use of various models or ANNs is the principal aspect of the ensembles.
We know that any ANN will not always be able to train itself as per the issue. It can
be due to the magnitude of the problem or the data’s size and noise. Also, it might
not be prudent to add neurons to the system because they have a detrimental effect
on the system’s generalization ability. We, therefore, use several ANNs. Each ANN
can now have a different configuration. Differences could be in weight, bias, design,
or even the ANN model used in its entirety.
As a consequence, various skills and performances are seen. All can be said to
specialize in various ways. We must build an integrator to make the most of each
module or ANN’s power. Other ANNs eliminate the sub-optimality in the preparation
of the ANNs. The entire method, therefore, provides a better outcome for the problem.
This makes valuable ensembles tools to solve issues that work badly when single
ANNs are used. This notion is shown in Fig. 10.

3.3 Evolutionary Artificial Neural Network

Evolutionary neural networks are the last hybrid approach that we use to solve the
problem. We train the artificial neural network in this technique using the genetic
algorithm. ANN’s training uses BPA, but many limitations are overcome by using
40 M. Sharma et al.

Fig. 10 Ensemble using a polling approach

GA. The ANN might suffer from the local minima problem. The algorithm design
was defined initially, which could not be optimized according to the problem [42, 43].
The methods used by evolutionary ANN develop and train the ANN using the genetic
algorithm’s optimization property. The ANN approach optimizes the methodology
in which mistake or output can play an important role. The weights, prejudices and
the framework are the modifiable parameters.
A fully associated model in which each neuron of each layer has a complete
relationship to the neurons of the associated layers is the ANN we consider to solve
the problems shown in Fig. 11. This arrangement results in an excessive demand for
calculation that increases the overall time for preparation. Neuron subtraction may
not be that important as the network may not train fewer neurons. If the scheme
fails to plan, the general methodology at that point is the expansion of a neuron. Be
that as it may, the enormous increase in size may expand a solitary neuron that will
increase the dimension by an enormous amount. This will make it difficult to prepare
the device for the training calculation because of the equivalent. Henceforth, we use
the idea of a connectionist approach as a general multifaceted nature of the problem
that will try to hinder or prevent such connections by limiting the calculation time.
This makes the forecasts in the blink of an eye and much more; since the similarities
are confined, preparation will eventually show itself. We will usually use GA to
establish the connectionist framework during this downside, despite settling ANN
loads. Nevertheless, the most intense range of neurons must be highlighted at the
start.
The first portion of the body consists of a zero or one that depends on an interaction
between the input layer and the hidden layer’s neurons. One reflects the existence
Hybrid Intelligent System for Medical Diagnosis in Health Care 41

Fig. 11 ANN architecture

of the association, and a zero suggests the lack of association. Amid a hidden nerve
layer cell and an output nerve layer cell, the following section also shows weight.
The bits have an indicative constant. The opposite half requires the basic weight
values between the various relations of the body. The last half consists of the weights
between the input and hidden layer neurons, hence the second segment between the
hidden and output layer neurons. These squares test the basic values of associations
that hold whether or not the relationship exists. The corresponding weight value is
also ignored if the physical relationship is not there. It includes the values of the
biases in the last half.
The most important factor to use the GA algorithm is the genetic operators. This
algorithm helps in the production of the individuals to achieve them to greater heights.
The selection is considered the genetic operator, and it deals with selecting the indi-
viduals to go to a higher generation. The theory of selecting the fittest individual
and moving to the next generation is considered Darwin’s theory, which says only
the fittest individual will survive and move to the next generation. From the popula-
tion pool, only the individual who is the fittest will survive. The operator used is a
crossover, and the function of this operator is to mix the parent’s solutions and form a
child chromosome. The modification of this operator is required to fulfil the require-
ment of the problem. With the help of mutation, we can add new characteristics to
the individual. If the characteristics added are good, then the algorithm will work,
and the new off springs will be good. However, if the added characteristics are not
good, then the individual will be removed by the concept of survival of the fittest.
The fitness function is a function that can measure how good or optimal the solu-
tion is. The inhabitants are improved by the GA in its run when the fitness function
42 M. Sharma et al.

calculates the optimality. Every individual is considered as an ANN which predicts

the presence of disease.

3.4 Application of Hybrid Intelligent System in Health Care

The hybrid intelligent system must solve problems like diagnosing and detecting
diseases in the preliminary stages. Many diseases can be detected using these systems,
which have various applications in various diseases. We need a data set from the UCI
machine learning repository [44] for implementation. The UCI machine learning data
is obtained from the repository of machine learning. Depending on the device, it is
sent for training after the elicitation of data using the various training algorithms.
The system is fed with the new data selected from the sample data set. The output of
the system is collected, and the comparison is made with the standard output. After
the data is trained, it is sent for testing. The systems are tested for various parameters
like performance, efficiency, ability and evaluation [45]. The general methodology
is summarized by Fig. 12.

Fig. 12 General working

methodology
Hybrid Intelligent System for Medical Diagnosis in Health Care 43

Hence, it is based on the optimization and automation of IT processes and the

technical environment of interrelated objects. This environment aims to develop
the current procedures for providing advanced medical care and open up new
medical possibilities. The automated system’s fundamental principle includes effi-
ciency, quality improvement, expansion of the patient’s capabilities, accessibility for
everyone, etc.

4 Need of Hybrid Intelligent System in Medical Diagnosis

The hybrid intelligent machine becomes increasingly difficult to do what people do,
but more effectively, more efficiently and at less cost. In several real-world appli-
cations, many intelligent technologies need to be implemented. Each technology
compensates for the other technology’s weak areas. A smart hybrid system combines
two smart technologies. The combination of a neural network with a fugitive tech-
nique, for example, leads to a hybrid neurofugal technique. The combination of
probability reasoning and neural networks thus forms the foundations or essential
components of soft computing. This soft machine is considered a developmental
viewpoint to construct hybrid intelligent systems in an unresolved and unclear sense,
reason and understand. The scope of social insurance for simulated intelligence and
mechanical autonomy is immense. Like our consistent lives, computer-based infor-
mation and mechanical technology are increasingly a part of our eco-framework for
medical services. We have featured eight different ways that exhibit how this change
is right now in progress, as shown in Fig. 13.
(i) Keeping Well
One of the biggest possible benefits of the hybrid intelligent system is to assist
citizens with staying solid because they do not need to bother with a profes-
sional, or if nothing else, not as often. As of now, the use of a half-breed-wise
system in shopper well-being applications benefits individuals. Applications
and applications for creativity promote more positive action in individuals
and help proactively manage a sound way of life. This puts consumers in
control of health and prosperity. It also improves human services profes-
sionals’ willingness to consider the common examples and necessities of
the people they care about. They can provide better feedback, direction and
backing for remaining sound with that understanding.
(ii) Early Detection
To recognize disease all the more precisely and in the beginning times, half
and half canny framework is probably the best framework to distinguish
it in the beginning periods. As indicated by American Malignant growth
Study (ACS), a great extent of X-Beams yields fake results encouraging
1 out of 2 sound ladies to be told that they have the disease. These frame-
works’ proper use gives quicker outcomes, best review and exactness esteems
planning to accomplish around 99% precision, lessening the requirements’
44 M. Sharma et al.

Fig. 13 Hybrid intelligent system

superfluous assessments. The spread of shopper wearables and other clinical

gadgets joined with the mixture smart framework is likewise applied to regu-
late beginning period coronary illness, empowering specialists and different
parental figures to more readily screen and distinguish conceivably perilous
scenes at prior, increasingly treatable stages.
(iii) Diagnosis
To unlock vast volumes of health information and capacity diagnosis, IBM’s
Watson for health helps service institutions implement psychological feature
technologies. Watson can scan and archive more medical evidence infinitely
faster than every human—each medical journal, symptom, and case study of
care and reaction worldwide. Google’s DeepMind well-being aims to unravel
real-world concentration challenges in collaboration with doctors, academics
and patients. The technology comprises machine learning and systems neuro-
biology to transform the general learning algorithms into neural networks
that imitate the human brain.
(iv) Decision-Making
Upgrading treatment involves aligning vast patient information with appro-
priate and timely decisions, and therapeutic decision-making and behaviour
and prioritizing body activities can be assisted by predictive analytics.
Another area where the hybrid intelligent system starts to take shape in health
care is identifying victimization trends to spot patients at risk of acquiring
a disease or watching it deteriorate due to modus vivendi, occupational,
genomic or alternative causes.
Hybrid Intelligent System for Medical Diagnosis in Health Care 45

(v) Treatment
In addition to scanning health records to aid suppliers in establishing invet-
erately unwell persons, the United Nations organization is now at risk of
AN adverse episode. A hybrid intelligent device would make it possible
for physicians to take a holistic approach to illness treatment, help organize
clinical arrangements and make it easier for patients to handle their long-run
therapy services more efficiently. Robots have been used in medicine for
over 30 years. They range from basic laboratory robots to overly sophisti-
cated surgical robots that can support the operative surgeon with a personality
or conduct procedures independently. In hospitals and laboratories, they are
used for routine activities, recovery, physiotherapy and long-term care for
surgery.
(vi) End of Life Care
Firstly, lives have become longer in comparison with previous generations.
Second, death is different and slower as the end of life is reached due to
varying heart failure, dementia and osteoporosis. Also, isolation has become
a part of life. Robots will theoretically change the area of life care, enabling
patients to remain free for a prolonged period, thus reducing hospitaliza-
tion and care home requirements. In conjunction with advances in humanoid
design, the hybrid intelligent system has allowed robots to hold ‘conver-
sations’ and numerous social interactions so that ageing minds remain
sharp.
(vii) Research
The distance between an exploration laboratory and an actual patient might
be long and demanding. In the context of the American state clinical strength
investigation affiliation, it takes a mean of twelve years for a medication to
reach the patient from the examination research facility. Exclusively 5 out
of 5000 special test of the medication are assembled for human testing,
and only one of those five is ever endorsed for human utilization. By and
large, it will esteem an enterprise US $359 million to build up a substitution
tranquillize from the exploration research facility to the patient. Medication
investigation and revelation are among the more current applications for the
cross-breed astute framework in medicinal services. By managing the most
up to date propels inside the half and half insightful frameworks to shape
the medication disclosure and medication repurposing forms, there is the
possibility to impressively cut each the time and incentive to plug for fresh
out of the plastic new medication.
(viii) Training

The training is done via realistic simulations using hybrid intelligent systems that
simple computer algorithms cannot do. The natural language is the advancement
and is considered the ability to draw on a big database so that the decisions can be
taken as a challenge that humans cannot. The previous responses can be used as a
training programme to make them learn. The training can also be done anywhere via
the hybrid intelligent systems that can be used on smartphones.
46 M. Sharma et al.

Table 1 Intelligent technologies for the solution to healthcare problems

S. No. Category Some approaches
Intelligent techniques can solve many Feed-forward neural network
healthcare problems such as skin diseases, Modular neural network
different types of cancers related problems,
Sepsis attack, Alzheimer’s, Parkinson’s, Quantitative approach
chronic kidney diseases and many more Back-propagation neural network\
Fuzzy expert system
Adaptive neuro-fuzzy inference systems
(ANFIS)
Evolutionary artificial neural network
Recurrent neural network
Ensemble approaches

Early disease identification is one of the biggest obstacles faced by healthcare

providers. Therefore, they are a wide variety of illnesses such as cancer, heart attacks,
sepsis attack, Alzheimer’s, Parkinson’s and COVID-19, among others. The precision
of many conventional applications is not adequate to disclose this type of disease.
This chapter seeks to implement an intelligent model which combines two or more
technologies to tackle this challenge.
Below is the name of many intelligent techniques that can be applied for the
solution of many healthcare problems which is shown in Table 1.

5 Conclusion

There are the tremendous scope and opportunity for developments made in the hybrid
intelligence system to revolutionize healthcare practices. This chapter’s essential
purpose was to make one aware of the implications of recent hybrid technology and
the ongoing research and its findings on hybrid computational intelligence applica-
tions in healthcare practices. This chapter will offer readers a broad and consistent
perspective on the development and usage of this area. The chapter offers a thorough
insight into how hybrid intelligent systems are much better than the conventional
method. First, the chapter discusses the need for an intelligence structure followed
by a comprehensive hybrid intelligence description and then a comparative analysis
between the two. A thorough insight into health care is also presented, which also
illustrates the progress that health care has made by introducing intelligence system
techniques. The need for both the intelligence system and the healthcare sector’s
hybrid intelligence system was then met.
With such a fast healthcare transition rate, healthcare institutions increasingly
adapt to the new technology to control and fulfil patient demands. The artificial
intelligence (AI) field provides a fantastic opportunity to learn from the past and make
better future decisions, as seen above. The genetic algorithm, neural networks, fuzzy
Hybrid Intelligent System for Medical Diagnosis in Health Care 47

logic, machine learning and various basic computational intelligence techniques are
hybrid to each other to construct the hybrid method. The continued emphasis and
attention on quality, cost and treatment would ensure that developments are made
in various AI innovations to increase the quality of the services delivered across
different healthcare systems.

References

1. Meesad, P., & Yen, G. G. (2003). Combined numerical and linguistic knowledge representation
and its application to medical diagnosis. IEEE Transactions on Systems, Man and Cybernetics,
Part A: Systems and Humans, 33(2), 206–222. https://doi.org/10.1109/TSMCA.2003.811290
2. Pavlopoulos, S. A., & Delopoulos, A. (1999). Designing and implementing the transition to
a fully digital hospital. IEEE Transactions on Information Technology in Biomedicine, 3(1),
6–19. https://doi.org/10.1109/4233.748971
3. Chabat, F., Hansell, D. M., & Yang, G. (2000). Computerized decision support in medical
imaging. IEEE Engineering in Medicine and Biology Magazine, 19(5), 89–96. https://doi.org/
10.1109/51.870235
4. Tsipouras, M. G., Voglis, C., & Fotiadis, D. I. (2007). A framework for fuzzy expert
system creation—Application to cardiovascular diseases. IEEE Transactions on Biomedical
Engineering, 54(11), 2089–2105. https://doi.org/10.1109/tbme.2007.893500
5. Kwiatkowska, M., Atkins, M. S., Ayas, N. T., & Ryan, C. F. (2007). Knowledge-based data anal-
ysis: First step toward the creation of clinical prediction rules using a new typicality measure.
IEEE Transactions on Information Technology in Biomedicine, 11(6), 651–660. https://doi.
org/10.1109/titb.2006.889693
6. Isola, R., Carvalho, R., & Tripathy, A. K. (2012). Knowledge discovery in medical systems
using differential diagnosis, LAMSTAR, and k-NN. IEEE Transactions on Information
Technology in Biomedicine, 16(6), 1287–1295. https://doi.org/10.1109/TITB.2012.2215044
7. Çomak, E., Polat, K., Güneş, S., & Arslan, A. (2007). A new medical decision-making system:
Least square support vector machine (LSSVM) with fuzzy weighting pre-processing. Expert
Systems with Applications, 32(2), 409–414. https://doi.org/10.1016/j.eswa.2005.12.001
8. Luukka, P. (2011). Feature selection using fuzzy entropy measures with similarity classi-
fier. Expert Systems with Applications, 38(4), 4600–4607. https://doi.org/10.1016/j.eswa.2010.
09.133
9. Kahramanli, H., & Allahverdi, N. (2009). Extracting rules for classification problems: AIS
based approach. Expert Systems with Applications, 36(7), 10494–10502. https://doi.org/10.
1016/j.eswa.2009.01.029
10. Sekar, B. D., Dong, M.-C., Shi, J., & Hu, X. Y. (2012). Fused hierarchical neural networks
for cardiovascular disease diagnosis. IEEE Sensors Journal, 12(3), 644–650. https://doi.org/
10.1109/JSEN.2011.2129506
11. Markowska-Kaczmar, U., & Matkowski R. (2006). Experimental study of evolutionary based
method of rule extraction from neural networks in medical data. In P. Perner (Eds.), Advances
in data mining. Applications in medicine, web mining, marketing, image and signal mining.
ICDM 2006. Lecture notes in computer science (Vol. 4065, pp. 76–90). Springer. https://doi.
org/10.1007/11790853_7.
12. Economou, G. P. K., Goumas, P. D., & Spiropoulos, K. (1996). A novel medical decision
support system. Computing and Control Engineering Journal, 7(4), 177–183. https://doi.org/
10.1049/cce:19960404
13. Kovalerchuk, B., Vityaev, E., & Ruiz, J. F. (2000). Consistent knowledge discovery in medical
diagnosis. IEEE Engineering in Medicine and Biology Magazine, 19(4), 26–37. https://doi.
org/10.1109/51.853479
48 M. Sharma et al.

14. Kinney, W. C. (2003). Web-based clinical decision support system for triage of vestibular
patients. Otolaryngology-Head and Neck Surgery, 128(1), 48–53. https://doi.org/10.1067/mhn.
2003.33
15. Huang, K., Yang, H., King, I., & Lyu, M. R. (2006). Maximizing sensitivity in medical diagnosis
using a biased minimax probability machine. IEEE Transactions on Biomedical Engineering,
53(5), 821–831. https://doi.org/10.1109/TBME.2006.872819
16. Zadeh, L. A. (1998). Roles of soft computing and fuzzy logic in the conception, design and
deployment of information/intelligent systems. In O. Kaynak, L. A. Zadeh, B. Türkşen, & I.
J. Rudas (Eds.) Computational intelligence: Soft computing and fuzzy-neuro integration with
applications. NATO ASI Series (Series F: Computer and Systems Sciences) (Vol. 162). Springer.
https://doi.org/10.1007/978-3-642-58930-0_1.
17. Kandel, A., & Langholz, G. (Eds.) (1992). Hybrid architectures for intelligent systems. CRC
Press. ISBN-9780849342295.
18. Grosan, C., & Abraham, A. (2011). Hybrid intelligent systems. In: Intelligent systems. Intel-
ligent systems reference library (Vol. 17). Springer. https://doi.org/10.1007/978-3-642-21004-
4_17.
19. Jouffe, L. (1998). Fuzzy inference system learning by reinforcement methods. IEEE Transac-
tions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), 28(3), 338–355.
https://doi.org/10.1109/5326.704563.
20. Procyk, T. J., & Mamdani, E. H. (1979). A linguistic self organising process controller,
automatica. Journal of IFAC, 15(1), 15–30. https://doi.org/10.1016/0005-1098(79)90084-0.
21. Wang, L. X. (1994). Adaptive fuzzy systems and control. Prentice Hall Inc. ISBN 978-0-13-
099631-2.
22. Wang, L. X., & Mendel, J. M. (1992). Backpropagation fuzzy system as nonlinear dynamic
system identifiers. In Proceedings of the First IEEE International Conference on Fuzzy Systems,
San Diego, USA (pp. 1409–1418). https://doi.org/10.1109/FUZZY.1992.258711.
23. Sugeno, M., & Tanaka, K. (1991). Successive identification of a fuzzy model and its applications
to prediction of a complex system. Fuzzy Sets Systems, 42(3), 315–334. https://doi.org/10.1016/
0165-0114(91)90110-C.
24. Abraham, A., & Nath, B. (2000). Evolutionary design of fuzzy control systems—A hybrid
approach. In J. L. Wang (Ed.), The Sixth International Conference on Control, Automation,
Robotics and Vision (ICARCV 2000), CD-ROM Proceeding. ISBN 9810434456.
25. Pedrycz, W. (Ed.). (1997). Fuzzy evolutionary computation. USA: Kluwer Academic
Publishers.
26. Sanchez, E., Shibata, T., & Zadeh, L. A. (Eds.). (1997). Genetic algorithms and fuzzy logic
systems: Soft computing perspectives. Singapore: World Scientific Publishing Company.
27. Cord´on, O., Herrera, F., Hoffmann, F., & Magdalena, L. (2001). Genetic fuzzy systems: evolu-
tionary tuning and learning of fuzzy knowledge bases. World Scientific Publishing Company.
https://doi.org/10.1142/4177.
28. Abraham, A. (2003). Business intelligence from web usage mining. Journal of Information and
Knowledge Management (JIKM), 2(4), 1–15. https://doi.org/10.1142/S0219649203000565
29. Abraham, A. (2003). i-Miner: A web usage mining framework using hierarchical intelligent
systems. In The 12th IEEE International Conference on Fuzzy Systems FUZZ—IEEE’03 (Vol.
2, pp. 1129–1134). IEEE Press. https://doi.org/10.1109/FUZZ.2003.1206590
30. Bhattacharya, A., Abraham, A., Grosan, C., Vasant, P., & Han, S. (2006). Meta-learning evolu-
tionary artificial neural network for selecting flexible manufacturing systems. In J. Wang, Z.
Yi, J. M. Zurada, B. L. Lu, & H. Yin (Eds.), Advances in neural networks—ISNN 2006. Lecture
notes in computer science (Vol. 3973). Springer. https://doi.org/10.1007/11760191_130
31. Millman, M. (1993). Access to health care in America (1993). The National Academies Press,
US National Academies of Science, Engineering and Medicine. ISBN-10: 0-309-04742-0.
32. Kjoelen, A., Thompson, M. J., Umbaugh, S. E., Moss, R. H., & Stoecker, W. V. (1995).
Performance of AI methods in detecting melanoma. Engineering in Medicine and Biology,
14(4), 411–416. https://doi.org/10.1109/51.395323
Hybrid Intelligent System for Medical Diagnosis in Health Care 49

33. Schölkopf, B., Burges, C., & Vapnik, V. (1995). Extracting support data for a given task. In U.
M. Fayyad, & R. Uthurusamy (Eds.), First International Conference on Knowledge Discovery
and Data Mining, Proceedings. AAA I Press.
34. Vapnik, V. N., Golowich, S. E., & Smola, A. (1998). Support vector method for function
approximation, regression and signal processing. In Advances in neural information processing
systems (Vol. 9, pp. 281–287). MIT Press.
35. Singh, H., Meyer, A. N. D., & Thomas, E. J. (2014). The frequency of diagnostic errors in outpa-
tient care: Estimations from three large observational studies involving US adult populations.
BMJ Quality and Safety, 23(9), 727–731. https://doi.org/10.1136/bmjqs-2013-002627
36. Ridley, E. L. SIIM: AI poised to enhance all aspects of radiology. 2017. Available at www.aun
tminnie.com/index.aspx?sec=sup&sub=aic&pag=dis&ItemID=117495! Accessed 21 March
2018.
37. Ciampi, A., & Zhang, F. (2002). A new approach to training back-propagation artificial neural
networks: Empirical evaluation on ten data sets from clinical studies. Statistics in Medicine,
21, 1309–1330. https://doi.org/10.1002/sim.1107
38. Xiaoguang, C., & Lilly, J. H. (2004). Evolutionary design of a fuzzy classifier from data. IEEE
Transactions on Systems, 34(4), 1894–1906. https://doi.org/10.1109/TSMCB.2004.831160.
39. Srinivasaa, K. G., Venugopala, K. R., & Patnaikb, L. M. (2007). A self-adaptive migration model
genetic algorithm for data mining applications. Information Sciences, 4295–4313. https://doi.
org/10.1016/j.ins.2007.05.008.
40. Shadabi, F., Sharma, D., & Cox, R. (2006). Learning from ensembles: Using artificial neural
network ensemble for medical out-comes prediction. In IEEE Conference: Innovations in
Information Technology. https://doi.org/10.1109/INNOVATIONS.2006.301896.
41. Kala, R., Shukla, A., & Tiwari, R. (2009). Fuzzy neuro systems for machine learning for
large data sets. In Proceedings of the IEEE International Advance Computing Conference
(pp. 541–545). https://doi.org/10.1109/IADCC.2009.4809069.
42. Augusteijn, M. F., & Harrington, T. P. (2004). Evolving transfer functions for artificial neural
networks. Neural Computing and Applications, 13(1), 38–46. https://doi.org/10.1007/s00521-
003-0393-9
43. He, S., Wu, Q. H., & Saunders, J. R. (2006). A group search optimizer for neural network
training. In Proceedings of Computational Science and Its Applications (pp. 934–943). ICCSA.
https://doi.org/10.1007/11751595_98.
44. Sigillito, V. G., Wing, S. P., Hutton, L. V., & Baker, K. B. (1989). Classification of radar returns
from the ionosphere using neural networks. Johns Hopkins APL Technical Digest, 10, 262–266.
45. Shukla, A., Tiwari, R., Meena, H., & Kala, R. (2009). Speaker Identification using wavelet
analysis and modular neural networks. Conference Proceedings of National Symposium on
Acoustics, 36(1), 14–19.
Remote Patient Monitoring Using IoT,
Cloud Computing and AI

M. V. V. Prasad Kantipudi, C. John Moses, Rajanikanth Aluvalu,

and Sandeep Kumar

Abstract Innovative wireless communication technologies using body sensors and

the advent of Internet of things (IoT) are used to introduce many modern health-
care schemes to provide personalized health management and prevent some acute
diseases. Monitoring and caring for patients remotely using modern techniques
such as IoT, cloud computing (CC), and artificial intelligence (AI) are evolving
proposals on healthcare innovations. Over the past decade, taking care of patients
through remote access is widely suggested by different authors to monitor patients
affected by different illnesses like heart disease, neurological sickness, blood pres-
sure, body temperature, chronic disease, diabetes, and obesity. Further, remote moni-
toring is used to care for post-operative patients and aged patients using smart
sensors and intelligent decision-making technologies. CC is a sophisticated tech-
nology comprised of remote servers which act as a gateway of remote access by
connecting intelligent sensors and intelligent devices with the concept of IoT. AI is
a genius technique of making decisions using deep learning (DL) methodology with
the cloud dataset. This chapter illustrates the concept of cloud and AI-based IoT for
remote health caring. It also discusses different decision-making systems using AI
and the principle of operation of several cloud infrastructures used to access secured
medical records. By learning this chapter, the readers and young researchers will
understand the principle and challenges of IoT, CC, and AI in various healthcare
schemes to identify the suitable architecture of the cloud and AI for different disease
diagnosis and patient monitoring.

Keywords Remote access · Smart sensors · Intelligent decision · Remote servers ·

Blockchain

M. V. V. P. Kantipudi (B) · C. J. Moses · S. Kumar

Sreyas Institute of Engineering and Technology, Hyderabad, India
R. Aluvalu
Vardhaman College of Engineering, Hyderabad, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 51
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_3
52 M. V. V. P. Kantipudi et al.

1 Cloud-Oriented IoT Using AI

1.1 Introduction to Internet of Things

Day by day, IoT plays a vital role in human life and it has the power to change
our lives significantly. IoT is extending the strength of the Internet beyond human
expectations. IoT refers to the physical devices connected to the internet and all
collecting and sharing data. IoT is connecting the entire smart device through the
Internet to communicate all the devices correctly. In other words, we can say IoT is
made up of smart sensors, devices and wearable connected. Nowadays, the arrival
of supercomputer chips and the ubiquity of wireless networks make it possible to
turn anything into a part of the IoT. It is one of the fast-growing and widely used
technologies in our fast-track carrier.
IoT grows day by day because more than 50 billion devices will be connected to
the Internet by 2020 and by 2025 IoT market is expected to reach 6.2 trillion dollars.
Figure 1 shows the growth of IoT in terms of the number of connected devices. IoT
consists of two simple words. The first one is the Internet, and the second one is the
thing. All among these two words. Anything is connected to the Internet of things and
everything in smartphones, cars, laptops, microwaves, and heart monitor implants.
All of these are taken as an application of IoT. It describes a pure interconnected
world. In worldwide, every device is developed with smart capabilities to connect,
interact, and exchange data. It is a world with technology more straightforward,
simple, and better.
Top ten Internet of things (IoT) trends that will rule in 2020,
Voice Assistants: In a smart house, IoT allows all smart devices to communicate
or travel from area to area.

Fig. 1 Growth of IoT in terms of number of connected devices

Remote Patient Monitoring Using IoT, Cloud Computing and AI 53

Wearables Are Finally Diversifying: A smart wearable device provides health

and fitness status to doctors for instant access to saving our lives in daily life.
Smart Mirror on the Wall: A smart mirror can show the day-by-day updates,
i.e., weather, news, and work schedule.
Smart Home Market: In a smart home, IoT-based smart devices can reduce
power, time, energy, and costs.
Do not Forget About Self-Driving cars: Top automotive companies are
ambitious on IOT-based self-driving cars.
Behold, The Flying Taxi: The flying car or taxi will start for traveling in UAE,
and it will be transformed sooner in the whole world.
Edge Computing: The edge computing technology will improve response times,
and it will save bandwidth.
Flexible Displays: The flexible displays technology will play a significant role in
the decade.
5G: Obscuring Reality: 5G technology designed with cellular network concept,
and this IoT application will make a big difference.
Sensor Innovation: Billions of smart sensors are interconnected through smart
sensors to collect the surrounding environment’s data.
The functional structure of IoT is shown in Fig. 2.
As shown in Fig. 2, the plant is any physical system or environment that can
interact with the IoT system. IoT consists of a group of devices connected from the
leaves in the network. A node consists of sensors/actuators, memory, processor, and
every node is connected through an interface. A node may run the protocol of IoT.
Hubs produce the first-level connection between the network and nodes. Hubs are
usually run IP and fog processor operates on a set of local nodes and hubs. Further,

Fig. 2 Structure of IoT

54 M. V. V. P. Kantipudi et al.

cloud servers present computational services of the IoT structure. Databases are used
to store general information and the results of the computation. The cloud can interact
between the users and nodes [1, 2].

1.2 Cloud Computing (CC)

Cloud computing is a pool of configurable resources, i.e., storing, networks, applica-

tions, and organizations, that can be immediately conveyed with unimportant orga-
nization endeavors. Distributed computing would be a profitable market if all around
used [30].
Cloud computing is utilized to get to a bunch of registering assets. The web keeps it
up. Utilizing these assets, clients can get to foundations, figuring force, applications,
and administrations on request. It empowers clients to get to their data, wide range
of applications, and information anyplace dependent on their requirements. More-
over, cloud computing is implied by the straightforward transformation of equivalent
and conveyed systems used for sharing, assurance, and complete of topographically
disseminated ‘self-administering resources intensely at runtime depending upon their
capacity, execution, and customers’ nature of administration necessities [3].
These days advancement of cloud computing is extraordinary in the IT region.
It ensures the limit of a broad scope of data on the web safely. Currently, cloud
computing research is the emerging area for the clinical field, research field, huge
tech and medications, and it is a purposeful anecdote for the web [4].
In the digital world, smart gadgets are used to get all social orders through the web.
So distributed computing plays a significant role in storing and save this real-time
data for various applications, i.e., WhatsApp, Facebook, Twitter, and Amazon. It has
developed as a popular answer for a gracefully modest and straightforward way to
deal with externalized IT assets. Figure 3 shows an essential distributed computing
climate [5].

1.3 Artificial Intelligence (AI)

Artificial intelligence (AI) insight the machines can think like human beings and
take their own decisions based on their surroundings. It is also called machine intel-
ligence. PCs are also ready to perform errands, such as visual discernment, discourse
acknowledgment, dynamic, and interpretation between dialects, ordinarily requiring
human knowledge [6].
In programming designing, plans to make contraptions that can re-enact the human
capacity to think, see, profoundly customized clients encounter and have the alter-
native to tackle different kinds of issues. This limit will allow cycles to end up
being snappier [7]. With mechanical progressions in AI and its potential, we see
Remote Patient Monitoring Using IoT, Cloud Computing and AI 55

Fig. 3 Cloud computing environment

its extending fuse in whole regions of society, for instance, prescription, public
organizations, and the military. Figure 4 [8] shows the eventual fate of AI.
Artificial Intelligence’s Trusts:
AI standards and scenarios, from education to entertainment [9] as dimensions
one may identify with as follows:
Job effect: AI makes our life easy, and it will reduce the workload by coping with
our work schedules, robotic mechanization of tasks.
Education: AI enhances the learning of understudies, for example, through
programmed mentoring or reviewing.

Fig. 4 Future of AI [8]

56 M. V. V. P. Kantipudi et al.

Transportation: New modes of transport, such as self-driving cars, flying cars,

are allowed by AI.
Health care: AI improves human beings’ well-being and prosperity, e.g., by
helping to find, reveal medications, or motivate personalized treatment.
Creation of decisions: AI or master structures allow us to decide on better choices,
e.g., when to hold a meeting or business chiefs’ case-based thinking.
Entertainment: AI gives us pleasure by entertainment, e.g., but in video games,
more astute adversaries.
Singularity: A possible particularity would bring positive social benefits, e.g.,
interminability.

1.4 Deep Learning Architecture

The profound learning concept resolves the real-time issues significantly, and it will
provide the best endeavors to human life for a long time. Deep learning provides
in-depth information in all the applications, i.e., science, management, government,
and business. It has proven better outcomes as compared to other AI techniques. Deep
learning has delivered encouraging outperforms in standard language considering,
i.e., impression, question noting, language interpretation, etc.
The quantity of plans and calculations that are used in profound learning is called
profound learning design. The most mainstream profound learning designs are recur-
rent neural network (RNNs), long short-term memory (LSTM)/gated recurrent unit
(GRU), convolutional neural network (CNN), deep belief network (DBN), and deep
stacking networks (DSNs)—and afterward investigates open-source programming
choices for profound learning. These structures are applied in various circumstances;
however, the accompanying table records part of their typical applications [10].
Recurrent neural organization (RNN)
The RNN is one of the essential organization plans from which other profound
learning structures are planned. An intermittent organization may have associations
that contribute to prior layers (or into a comparable layer). This info licenses RNNs
to record past information sources and model issues true to form. RNNs contain a
rich game plan of engineering. It is chiefly utilized for discourse acknowledgment
and penmanship acknowledgment.
LSTM/GRU organizations
Hochreiter and Schimdhuber proposed the LSTM concept in the year 1997, and later
on, this technology came with RNN for various applications. The LSTM is a neural-
based structure to utilize less memory during execution. We are using this concept
in our daily life for smart devices and later on in 2014, improvement done in the
existing LSTM. For specific applications, the GRU has execution like the LSTM,
yet being less troublesome strategies, less loads, and faster execution. It might be
Remote Patient Monitoring Using IoT, Cloud Computing and AI 57

utilized for picture inscribing, signal acknowledgment, and characteristic language

text pressure.
Convolutional neural organization (CNN)
A CNN is a visual cortex usually handled a combination of multiple layers of neural
that and CNN plays a significant role in engineering to handling various applications.
The LeNet CNN engineering is involved a few layers that execute highlight extrac-
tion and afterward order. These layers are the combination of convolutions, filters,
pooling, an associated order layer and these layers cleared the path for various new
employments of profound learning neural associations. Despite picture preparation,
the CNN has been viably applied to video affirmation and various assignments inside
characteristic language handling. It is mainly utilized for picture acknowledgment,
video investigation, and standard language preparation.
Deep conviction organization (DCN)
The DBN is standard organization engineering yet fuses a novel preparing calcula-
tion. The DBN is including many concealed layer which is related to restricted Boltz-
mann machine (RBM). The significant utilization of DBN is disappointment forecast,
characteristic language understanding, data recovery, and picture acknowledgment.
Deep stacking networks (DSN)
The final design is the DSN, furthermore called a profound arched organization. A
DSN is not precisely equivalent to conventional profound learning structures, and
it is a profound course of action of individual organizations. This design resolves
one of the issues with profound learning: the multifaceted nature of preparing. Each
layer in a profound learning design dramatically expands the multifaceted nature of
preparing. It is utilized for persistent discourse acknowledgment and data recovery.
Innovative wireless communication technologies using body sensors and IoT
are used to introduce many modern healthcare schemes to provide personalized
health management and prevent some acute diseases. Monitoring and caring for
patients remotely using modern techniques such as IoT, CC, and AI are evolving
proposals on healthcare innovations. Over the past decade, taking care of patients
through remote access is widely suggested by different authors to monitor patients
affected by different illnesses like heart disease, neurological sickness, blood pres-
sure, body temperature, chronic disease, diabetes, and obesity. Further, remote moni-
toring is used to care for post-operative patients and aged patients using smart
sensors and intelligent decision-making technologies. CC is a sophisticated tech-
nology comprised of remote servers which act as a gateway of remote access by
connecting intelligent sensors and intelligent devices with the concept of IoT. AI is a
genius technique of making decisions using DL methodology with datasets available
in the cloud. This chapter illustrates the concept of cloud and AI-based IoT for small
health caring. It also discusses different decision-making systems using AI and the
principle of operation of several cloud infrastructures used to access secured medical
records. By learning this chapter, the readers and young researchers will understand
the principle and challenges of IoT, CC, and AI in various healthcare schemes to
58 M. V. V. P. Kantipudi et al.

identify the suitable architecture of the cloud and AI for different disease diagnosis
and patient monitoring.

2 Wireless Body Networks (WBN)

2.1 Overview

The improving usage of a wireless medium of communication and the reduction of

electrical invasive and non-invasive devices has motivated wireless body networks
(WBNs). The WBN is used to monitor any patients continuously without constraint
on the patient’s daily activities. Many technologies have developed to support WBNs
applications like biofeedback, remote monitoring and assisted living to meet their
high-quality services. This section explains the purpose of different body networks
and the supporting sensors and other applications of WBNs.

2.2 Architecture and Applications

WBN comprises different genetic sensors, and these sensors are fixed in various
parts of the human body. These kinds of setups measure various modifications in
human vital signs and determine humans’ emotions like happiness, stress, and fear.
The WBN device can communicate using coordinator nodes, typically a low power,
consume and have high processing capacities. The WBN is responsible for sending
the user’s biological signals to the caretakers to take the required actions. Figure 5
shows the commonly used architecture of WBN communication [11].
Communication between the different levels of WBN is accomplished by using
different technologies like Wi-Fi, ZigBee, IEEE 802.15.6, and Bluetooth [12]. Inter-
WBN communications can also be developed using master node and other personal
devices like home service robots and notebooks. WBN is widely used for medical
and non-medical applications, as shown in Fig. 6 [13]. As shown in Fig. 5 [12], the
WBN can diagnose different human diseases and monitor different user features, for
example, temperature, pulse, and blood pressure. When the anomalous condition is
identified, the wireless sensors’ data can be transferred to the smartphone. This data is
directly transferred to doctors or any other help desk for taking the necessary actions.
The Internet or a cellular network may be utilized to deliver the data from gateway to
destination, as shown in Fig. 5. Further, WBNs are used to take pre-emptive health
measures and remote monitoring.
Remote Patient Monitoring Using IoT, Cloud Computing and AI 59

Fig. 5 Four-level communication of wireless body networks

Fig. 6 WBN applications

2.3 Hybrid Sensor-Based Healthcare Systems

In the past decade, several enterprises created and distributed to the marketplace
various wearable goods in types of bracelets, smartwatches, earphones, and fitness
bands for inconspicuous. The health information gathered by these smart gadgets
60 M. V. V. P. Kantipudi et al.

endures irregularities and reliabilities that influence their effectiveness and fitness
for medical purposes. Wearable sensors used for health checking techniques may
be incorporated into clothes or precisely connected to the human body. The wear-
able sensors are competent for assessing biological factors such as electromyogram
(EMG), electrocardiogram (ECG), body temperature, heart rate (HR), blood pressure
(BP) [15–18].
Wearable sensors can be categorized into two kinds such as non-invasive and
invasive. Non-invasive wearable sensors are utilizing for non-stop monitoring, and
these sensors do not require any physical support with the device, i.e., smart band,
watches, and incorporated into textile. Invasive wearable sensors are categorized
further as slightly invasive, and these types of sensors need clinical involvement to
spot them within the human body [18].
Typically, a WBN node comprises single or extra sensors connected, embedded
microprocessor, limited computational capability and memory, power unit, and
transceiver unit. These units permit every node to converse with the system. Inter-
action among the intersections is central—it may be an interacting platform of
committed servers or secluded (cloud) servers. This network organization relates
to the core of the IoT: to offer instant contact to data at any moment and any location.
Figure 7 shows the different types of sensors and their positions in the human body.
Table 1 shows the different kinds of body sensor, and Table 2 describes the difference
between wearable sensor and implant sensor [18].

Fig. 7 Different types of

sensors and their positions in
the human body
Remote Patient Monitoring Using IoT, Cloud Computing and AI 61

Table 1 Typical kind of sensor in BSN

Type of sensor Function Lifetime
Physiological sensor (Wearable)
ECG Checking the heart electrical activity Weeks
EEG Checking the brain electrical activity Weeks
EMG Checking the electrical muscle activity Weeks
Pulse oximeter Checking O2 concentration in blood Weeks
Blood pressure sensor Checking blood pressure Weeks
Body temperature sensor Checking body temperature Weeks
Implantable sensors
Glucose sensor Checking glucose concentration in the blood Years
Drug delivery Deliver drugs to a target area Years
Cardiac defibrillator Correct cardiac arrhythmias Years
Neuro stimulator Treatment of movement disorders Years
Artificial retina Vision recovery Years
Endoscope capsule Visualization of the digestive tract Hours
Kinetic sensors (Wearable)
Accelerometer Checking body posture Weeks
Gyroscope Checking body orientation Weeks
Ambient sensors (Wearable)
Temperature sensors Checking ambient temperature Weeks
Humidity sensor Checking ambient humidity Weeks
Allergic sensor Checking allergic agent in the air Weeks

Table 2 Difference between

Characteristics Wearable sensor Implantable sensor
wearable and implantable
sensors Volume Larger Smaller
Weight Heavier Lighter
Power High Low
Operation time Several weeks Several years
Recharge ability High Low
Biocompatibility Low High
Example Patch, watch, and Capsule-type
T-shirt type
62 M. V. V. P. Kantipudi et al.

3 Cloud Infrastructure and Processing

3.1 Overview

Cloud computing provides online information for foundation, stockpiling, and appli-
cation [19] to control and design. This part gives the distinctive geography of cloud
computing for correspondence. The fundamental idea about cloud handling includes
parts, models of distributed computing, and their essential qualities. Furthermore,
it plots cloud framework and their administration models. At last, this part gives
ongoing innovation utilized in cloud computing.

3.2 Topology and Network Protocol for Remote Monitoring

Peer-to-peer Inter-Cloud federation: Clouds work together legitimately with

one another yet may utilize circulated substances for registries or expediting, as
shown in Fig. 8. Mists speak with one another and haggle legitimately without
intermediaries [20].
Centralized Inter-Cloud federation: Clouds utilize a focal substance to perform
or encourage asset sharing. The focal substance goes about as a storage facility
where the accessible cloud assets are enrolled. Concentrated inter-cloud federation
is portrayed in Fig. 9.
Multi-cloud Service: Customers access the various cloud through assistance facil-
itated by the cloud customer either remotely or in-house, as shown in Fig. 10. The
administrations contain agent segments.
Multi-cloud Libraries: Clients build up their own intermediaries by utilizing
a brought together cloud API as a library, as shown in Fig. 11. The utilization
libraries encourage the use of mists uniformly between clouds.

Fig. 8 Peer-to-peer
inter-cloud federation [20]
Remote Patient Monitoring Using IoT, Cloud Computing and AI 63

Fig. 9 Centralized
inter-cloud federation [20]

Fig. 10 Multi-cloud service

[20]

Fig. 11 Multi-cloud
libraries [20]

With the fast expansion in distributed computing applications and asset prerequi-
sites, cloud worker remaining task at hand and unpredictability are expanding rapidly,
the designs of cloud worker framework become increasingly perplexing. Checking
and the board for the distributed computing framework are turning out to be increas-
ingly troublesome. Intelligent platform management interface (IPMI) convention
can be applied to customary workers the executives. As it may, colossal CPU, orga-
nization, and capacity assets carry special tests to remote administration of cloud
worker resource [21]. To addresses the issues of far-off checking the board, in 1997,
64 M. V. V. P. Kantipudi et al.

Intel, HP, Dell, NEC four organizations began to create the executive’s norms, they
made a primer IPMI particular in 1998, IPMI2.0 adaptation was accessible in 2004, it
incorporates extra encryption, verification, virtual local area network (VLAN), serial
over LAN (SOL), and different capacities. It was in reverse viable with 1.0 and 1.5
variants of the detail; it gave more prominent security, freedom, adaptability, and
offer-of-band the executive’s ability.

3.3 Cloud Infrastructure

Cloud computing design is arranged into two standard areas: front end and back end,
as shown in Fig. 12. The customers act as front-end users, utilizing cloud supervision
and organization of workers with any PC program as a back end-user.
Figure 13 defines the three service models of cloud computing.
Software as a Service (SAAS)—In the first model of services, a request is facil-
itated as support of a client who gets to it through the web and a few models
of this category are mail administrations, information sharing, Google docs, and
web examination [20].
Platform as a Service (PAAS)—It supplies all the assets needed to fabricate
applications, and there is no compelling reason to download programming; other-
wise, it is called a cloud product. Administrations gave by PAAS are application
plan, improvement, testing, sending, and facilitating. Google App Engine and
Microsoft Azure are two models in this classification.

Fig. 12 Cloud computing architecture

Remote Patient Monitoring Using IoT, Cloud Computing and AI 65

Fig. 13 Cloud computing defines three service models, widely known as ‘as a service’

Infrastructure as a Service (IAAS)—It gives the equipment on which appli-

cation gave by SAAS and PAAS can take a shot at. So, it is otherwise called
hardware as a service (HAAS).

3.4 Cloud Computing Components and Characteristics

Components of Cloud—Customers, information centers, and circulated servers

are the three primary cloud computing mechanisms.
Clients—Clients have smartphones through which they are communicating their
information on the cloud.
Data Center—It is a hub of computers where users can subscribe to the consumers
through the Internet.
Distributed Servers—It will provide an alternate server if one server fails to
provide the services. It will increase the scalability while providing the services.
Essential Characteristics—The cloud computing is describing five essential
characteristics as per NIST.
Rapid Elasticity—It will provide the up and down scalability resources for cloud
computing.
Measured Service—Cloud providers will monitor the cloud services, i.e., billing,
optimization of resources, planning, etc.
On-Demand Self-Service—Cloud computing will provide the services without
any human being.
66 M. V. V. P. Kantipudi et al.

Ubiquitous Network Access—It will provide excellent services throughout the

network for thick and thin clients.
Resource Pooling—Multi-tenant model will provide the cloud services as per the
demand of clients via physical or virtual resources.
Cloud models—Cloud computing will provide excellent services regarding
security issues or privacy.

4 Challenges in Cloud and AI-Based IoT on Remote

Monitoring

4.1 Overview

Cloud computing performs a crucial part in the progressive world and facilitates the
various functions from structure to societal media. Such scheme must survive with
changing load and developing use signifying humanities’ collaboration and enslave-
ment on computerized processing methods while gratifying excellence of provision
(EoP) assurances. Facilitating these techniques is a legion of intangible methodolo-
gies integrated to meet the challenge of advancing computing functions. There is a
necessity to detect new methodology to realize present and upcoming challenges in
cloud computing. This section aims to deliver how emerging technologies such as
IoT, AI, and blockchain influences the cloud computing system.

4.2 Accessing Cloud with Validation

The cloud validation service is a confirmation stage with an obscure crossover design,
and it will control the client’s access assets with unified admittance and confir-
mation approaches. The cloud validation service incorporates straightforward and
intuitive verification techniques for multifaceted personality affirmation. These tech-
niques incorporate biometric strategies, such as unique mark confirmation, equip-
ment gadgets, RSA SecurID Token and FIDO authenticators, and setting-based vali-
dation utilizing elements, such as the client’s area and organization. Trust in a client’s
personality can likewise be set up through danger examination, because of client
attributes, for example, past conduct, authenticators recently utilized for validation,
and different elements. Figure 14 shows the functions of the cloud authentication
service. A cloud validation service arrangement comprises four primary segments:
the Cloud Authentication Service, the Identity Router, the Cloud Administration
Console, and the RSA SecurID Authenticate application introduced on client gadgets.
The cloud validation service will perform real-time resources for the protected
resources. It will improve the validation capability and services run on Microsoft
Azure, a cloud computing platform, etc. For all clients, services are standard because
Remote Patient Monitoring Using IoT, Cloud Computing and AI 67

Fig. 14 Cloud authentication service

of their shared nature. An identity router is a virtual appliance that communicates

with the components such as cloud authentication service, identity sources, and
RSA authentication manager. The cloud validation service components contain a
facilitated, multi-inhabitant cloud administration console that SuperAdmins use to
perform arrangements and the executive’s undertakings [22].

4.3 Block Chain-Oriented Healthcare Records

Blockchain methodology is the latest innovation in intelligence computing

machinery. It operates an absolute disseminated ledger of deals accumulated in a
chain up of cell blocks that may be manipulated to track record deals. Blockchain
developed an extensively recognized idea with the starter of bitcoin. In common, a
blockchain setup has three distinct kinds of nodes such as straightforward node, full
node, and mining node. A specific node is efficient in delivering and accepting deals
in the system and does not store up a ledger copy. A full node delivering justifies all
the deals in the system and stores a copy of the ledger. A full node has the capability
of mining. Mining is well-known as the procedure of creating a new block in the
system [23].
In real time, smart e-health functions can watch patient data by accumulating data
from implantable devices and wearable sensors creating private area networks. Smart
68 M. V. V. P. Kantipudi et al.

gateways accumulate and operate local managing of information accumulated from

appliances, incorporating noise filtering from health devices, information compres-
sion and combination and examines permitting discovery of critical movements in
the patient’s health. For example, a quick gateway accumulating test samples from a
pacemaker may enhance its sample proceeding to a heart attack, discovered via pre-
managing at the fog level. As generally edge machines have calculated and storage
capacity limitations, also difficult restraints occur in delivering an optimum method
for information security and preserving reliability. To guarantee the information is
safeguarded, blockchain methodology has been implemented in the IoT field and
more real-time techniques.
The crucial idea of blockchain meets from numerous disputes while incorpo-
rating with fog processing frameworks. Storage space capacity and scalability are
considered due to elevated cost and preservation overheads. Though only complete
nodes store the complete chain, the storage capacity constraints are considerable.
Also, the blockchain liabilities still involve fog frameworks. Efficient agreement
processes want to be established to authenticate blocks and restricted the distribution
and replication of blocks [24].

4.4 Reliability and Complexity in Computational Intelligence

CC is a model that empowers pervasive, helpful, and on-request network admittance

to a mutual pool of configurable registering assets. It is conceivable to get to both
programming and equipment models distantly in CC applications and practically
no information about their physical or consistent areas. Because of its low sending
and the executive’s costs, the CC worldview is in effect progressively utilized in a
wide assortment of online administrations and applications, including distant calcu-
lation, programming as-a-administration, off-web page stockpiling, amusement, and
correspondence stages. In any case, a few parts of CC applications, for example,
framework plan, advancement, and security issues have gotten too mind-boggling to
even think about being effectively treated utilizing conventional algorithmic method-
ologies under the undeniably high unpredictability and execution requests of current
applications.
CC assets may not be satisfactory because of their portable applications, generally
far from clients topographically. Edge computing (EC) disseminates the applications
with low-inertness and high-unwavering quality prerequisites. As both CC and EC
are asset touchy, a few significant issues emerge, such as leading work planning, asset
designation, and assignment offloading, impacting the entire framework’s presenta-
tion. Numerous improvement issues have been defined to handle these issues. These
improvement issues ordinarily have complex properties, which may not be tended
to by the customary curved advancement-based arrangements [25].
Computational insight (CI), comprising of a bunch of nature-roused computa-
tional methodologies, as of late shows incredible potential intending to these advance-
ment issues in CC and EC. As of late, computational intelligence (CI) methods have
Remote Patient Monitoring Using IoT, Cloud Computing and AI 69

cultivated competent answers for CC applications. CI strategies, for example, fake

neural organizations, profound learning, fluffy rationale, and transformative calcula-
tions, have empowered improving CC ideal models through their capacities of extri-
cating information from high amounts of correct information, in this way further
streamlining their plan, execution, and security concerning conventional methods
[26].

5 Case Studies

5.1 IoT-Based Remote Pain Monitoring System: From Device

to Cloud Platform [27]

Facial appearances are among social indications of agony that can build up a
programmed human agony appraisal apparatus. Such an instrument can be an option
in contrast to oneself report strategy and primarily serves patients. A smart gadget
is proposed to cover the patient’s surface electromyogram (sEMG) in this effort.
The wearable gadget fills in as a remote sensor hub and is coordinated into an IoT
framework for far-off torment observing. Furthermore, both low energy utilization
and wearing solace are thought of through the wearable gadget plan for long-haul
checking. A versatile web application is created for real-time spilling of high-volume
sEMG information, computerized signal handling, deciphering, and perception to
show ongoing agony information to parental figures distantly. The IoT-based distant
agony observing application and its design have appeared in Fig. 15, which is intended
for the real-time focal checking of inpatients and concentrated consideration of
patients in the medical clinic. It is made from four sections: wearable sensor node,
gateway, cloud sensor, and mobile web application.

Fig. 15 Distant pain observing in the clinic for inpatients and ICU patients
70 M. V. V. P. Kantipudi et al.

Fig. 16 Structure of IoT-based biopotential estimation framework and cloud-based information

stream in the programmed pain observing framework

Gateway is the intermediate between sensor hubs and cloud, and the gateway
can be an overall switch, the individual hotspot of a cell phone, or a shrewd passage
supporting with added highlights, for example, heterogeneity, adaptability, and unwa-
vering quality. The framework can profit from intelligent gateways, mainly when
heterogeneous information and correspondence innovations exist in the general
medical care far-off the observing framework. In this design, an HTML5-based
portable application goes about as an intelligent interface among framework and
parental figures. It can deliver continuous waveform, lead lightweight calculations,
spare information to an in-program information base and synchronize with the far-off
information base workers, as appeared in Fig. 16.

5.2 Internet of Things Sensor Assisted Security and Quality

Analysis for Healthcare Datasets Using Artificial
Intelligent-Based Heuristic Health Management System
[28]

In this work, an artificial health management framework has been planned and
created. This technology provides more security in the patient dataset and thera-
peutic administrations’ relationship over its various perspectives. These administra-
tions incorporate the limit for masters, specialists, orderlies, and staff to make better
choices quicker. Using IoT helped fake shrewd-based heuristic health management
frameworks improve and limit the security hazard on medical care informational
collections with IoT sensors. An IoT sensor gathered the clinical information from
different assets.
The dataset gathered from 10 subjects while performing 12 specific
actual exercises (e.g., standing inert, sitting, and unwinding, resting, strolling,
climbing/plummeting steps, midsection twists forward, the frontal rise of arms, knees
bowing, cycling, running, running, and hop front and back) as shown in Fig. 17.
Remote Patient Monitoring Using IoT, Cloud Computing and AI 71

Fig. 17 M-health data assortment measure structure

Among the three sensors, chest-related sensors giving the ECG estimations are
used to decide the various arrhythmias and assess the overall heart observing cycle.
The explanation behind picking the IoT cell phone well-being information has its
capacity to access the well-being data from any patient anyplace, simple to infec-
tion finding, essential patient data, and viably convey the preparation on portable
well-being information. As per the portrayal, the IoT sensor was put as portrayed in
Fig. 17.

5.3 A Survey on Deep Transfer Learning to Edge Computing

for Mitigating the COVID-19 Pandemic [29]

Worldwide health at times faces pandemics as is as of now confronting COVID-19

infection. The dispersion and contamination components of this illness are high.
Countless individuals from most nations are tainted within a half year from its first
report of appearance and continue spreading. The necessary frameworks are not
prepared up to certain stages for any pandemic; subsequently, moderation with the
existing limit gets essential. On the other hand, deep learning (DL) could moderate
COVID-19 like pandemics regarding stop spread, analysis of the sickness, drug and
immunization revelation, therapy, quiet consideration, and some more. However, this
DL requires enormous datasets just as unique processing assets. These sorts of gear
make the frameworks modern and robotized, which assists with adapting to a flare-
up. Be that as it may, these are outfitted with low registering assets; along these lines,
72 M. V. V. P. Kantipudi et al.

Fig. 18 Structure for edge computing

applying DL is likewise somewhat testing; consequently, DTL additionally would

be successful there.
In Fig. 18, a standard current COVID-19 episodes circumstance and conceivable
working model have appeared. In the primary situation, we may utilize a few medical
care sensors like circulatory strain sensors, internal heat-level sensors, webcam, and
so on to detect the information about every patient’s running medical issue. At that
point, all of the gathered information would be distributed off the EC layer. If the
produced report is an essential ailment, a programmed framework-ready message will
be sent with subtleties to the medical clinic control room and all related group special-
ists. In the second situation, a few public spot-checking sensors (like a robot, CCTV,
traffic cameras, and so forth) could be utilized to recognize pointless unlawful swarm
with or without wearing PPE with assistance from the DTL-EC-based model. On the
off chance, the model finds assembling, at that point, a programmed framework-ready
message will be sent with all the subtleties to the close-by power.
Remote Patient Monitoring Using IoT, Cloud Computing and AI 73

References

1. Sepanos, D., & Wlof, M. (2018) Internet-of-Things (IoT) system: Architectures, algorithms,
methodologies. Springer International Publishing.
2. Marques, G., Bhoi, A. K., de Albuquerque, V. H. C., K.S., H. (Eds.) (2021). IoT in healthcare
and ambient assisted living. Springer.
3. Ahsan, M. M., Gupta, K. D., Nag, A. K., Poudyal, S., Kouzani, A. Z., & Mahmud, M. A. P.
(2020). Applications and evaluations of bio-inspired approaches in cloud security: A review.
IEEE Access, 8, 180799–180814.
4. Saran, P., Rajesh, D., Pamnani, H., Kumar, Hemant, S. T. G., & Shridevi, S. (2020). A survey on
health care facilities by cloud computing. In Proceedings of the 2020 International Conference
on Emerging Trends in Information Technology and Engineering (ic-ETITE) (pp. 1–5), Vellore,
India.
5. Rajkumar, K., & Dhanakoti, V. (2020). Methodological survey to improve the secure data
storage in cloud computing. In Proceedings of the 2020 International Conference on Emerging
Smart Computing and Informatics (ESCI) (pp. 313–317), Pune, India.
6. Jagirdar, Reddy, & Qyser. (2014). Cloud computing basics. International Journal of Advanced
Research in Computer and Communication Engineering, 1(5), 1–6.
7. Yeasmin, S. (2019). Benefits of artificial intelligence in medicine. In Proceedings of the 2nd
International Conference on Computer Applications & Information Security, ICCAIS’2019
(pp. 1–6).
8. Oliveira, Lopes, & Govcopp. (2020). What can we expect from the future? The impact of
artificial intelligence on society. In Proceedings of the 15th Iberian Conference on Information
Systems and Technologies (CISTI) (pp. 1–6).
9. Mohammed, Z. (2019). Artificial intelligence definition, ethics and standards. In Elec-
tronics and communications: Law, standards and practice. The British University in Egypt
Publications.
10. Fast, E., & Horvitz, E. (2017). Long-term trends in the public perception of artificial intelli-
gence. In Proceedings of the 31st AAAI Conference on Artificial Intelligence (pp. 963–969).
11. Zhong, G., Zhang, K., Wei, H., Zheng, Y., & Dong, J. (2019). Marginal deep architecture:
Stacking feature learning modules to build deep learning models. IEEE Access, 7, 30220–
30233.
12. Negra, R., Jemili, I., & Belghith, A. (2016). Wireless body area networks: Applications and
technologies. Procedia Computer Science, 83, 1274–1281.
13. Roudjane, M., & Messaddeq, Y. (2020). Innovative wearable sensors based on hybrid materials
for real-time breath monitoring. Journal of Wireless Sensor Networks—Design, Deployment
and Applications, 1, 1–22.
14. Islam, S. U., Ahmed, G., Shahid, M., Hassan, N., Riaz, M., Jan, H., & Shakeel, A. (2017).
Implanted wireless body area networks: Energy management, specific absorption rate and safety
aspects. In Chapter 2, Ambient assisted living and enhanced living environments (pp. 17–26).
Elsevier.
15. Roy, G., Bhoi, A. K., & Bhaumik, S. (2021). A comparative approach for MI-based EEG
signals classification using energy, power and entropy. IRBM.
16. Pramanik, M., Pradhan, R., Nandy, P., Bhoi, A. K., & Barsocchi, P. (2021). Machine learning
methods with decision forests for Parkinson’s detection. Applied Sciences, 11(2), 581.
17. Bhatt, T. V., Patel, R. K., Chitara, H. B., Marques, G., & Bhoi, A. K. (2020). Fuzzy logic
system for diabetic eye morbidity prediction. International Journal of Computer Applications
in Technology, 64(4), 339–348.
18. Bhoi, A. K., Sherpa, K. S., & Khandelwal, B. (2018). Arrhythmia and ischemia classification
and clustering using QRS-ST-T (QT) analysis of electrocardiogram. Cluster Computing, 21(1),
1033–1044.
19. Wang, Y., Wang, H., Xuan, J., Dennis, Y. C., & Leung, D. Y. C. (2020). Powering future body
sensor network systems: A review of power sources. Journal of biosensors and Bioelectronics,
166, 1–23.
74 M. V. V. P. Kantipudi et al.

20. COV 2020: Cloud computing overview. https://www.tutorialspoint.com/cloud_computing/

cloud_computing_overview.html [5 Nov 2020].
21. Rani, K., Rani, P., & Babu, V. (2015). Cloud computing and inter-clouds—Types, topologies
and research issues. In Proceedings of the 2nd International Symposium on Big Data and Cloud
Computing (ISBCC’15) (Vol. 50, pp. 24–29).
22. Hongsong, C., & Xiaomei, M. (2015). Design and implementation of cloud server remote
management system based on IMPI protocol. In Proceedings of the 2015 IEEE 12th Inter-
national Conference on Ubiquitous Intelligence and Computing and 2015 IEEE 12th Inter-
national Conference on Autonomic and Trusted Computing and 2015 IEEE 15th Interna-
tional Conference on Scalable Computing and Communications and Its Associated Workshops
(UIC-ATC-ScalCom) (pp. 1475–1478), Beijing.
23. Dinesha., H. A., & Agrawal., V. K. (2012). Multi-level authentication technique for accessing
cloud services. In Proceedings of the 2012 International Conference on Computing, Commu-
nication and Applications (pp. 1–4), Dindigul, Tamilnadu.
24. Ismail, L., & Materwala, H. (2020). Block—A blockchain-based framework for health records
management. Proceedings of ICCMS, 164–168.
25. Gill, S. S., Tuli, M., Xu, M., Singh, I., Singh, K. V., Lindsay, D., Tuli, S., Smirnova, D., Singh,
M., Jain, U., Pervaiz, H., Sehgal, H. B., Kaila, S. S., Misra, S., Aslanpour, M. S., Mehta, H.,
Stankovski, V., & Garraghan, P. (2020). Transformative effects of IoT, blockchain and artificial
intelligence on cloud computing: Evolution, vision, trends and open challenges. Journal of
Internet of Things, 10, 1–30.
26. Asim, M., &Wang, Y. (2020). A review on computational intelligence techniques in cloud and
edge computing. IEEE Transactions on Emerging Topics in Computational Intelligence, 1–22.
27. Labati, Genovese, Piuri, Scotti, & Vishwakarma. (2020). Computational intelligence in cloud
computing. In Proceedings of Conference on Recent Advances in Intelligent Engineering,
Topics in Intelligent Engineering and Informatics (pp. 111–126).
28. Yang, G., Jiang, M., Ouyang, W., Ji, G., Xie, G., Rahmani, A. M., Lijeberg, P., & Tenhunen,
H. (2018). IoT-based remote pain monitoring system: From device to cloud platform. IEEE
Journal of Biomedical and Health Informatics, 22(6).
29. Amoon., M., Altameem., T.,& Altameem., A. (2020). The Internet of Things sensor assisted
security and quality analysis for health care data sets using an artificial intelligence-based
heuristic health management system. Journal of Measurement, 161, 1–9.
30. Sufian, A., Ghosh, A., Sadiq, A. S., Smarandache, F. (2020). A survey on deep transfer learning
to edge computing for mitigating the COVID-19 pandemic. Journal of Systems Architecture,
108, 1–11.
31. Meera, A. J., Kantipudi, M. P., & Aluvalu, R. (2019). Intrusion detection system for the IoT: A
comprehensive review. In International Conference on Soft Computing and Pattern Recognition
(pp. 235–243). Springer, Cham.
32. Subhadra Bose Shaw, A. K. Singh. (2014). A survey on cloud computing, 2014 International
conference on green computing communication and electrical engineering (ICGCCEE).
33. Gill, S. S., Shreshth Tuli, M. Xu, Inderpreet S., et al. (2019). Transformative effects of IoT,
Blockchain and Artificial Intelligence on cloud computing: Evolution, vision, trends and open
challenges, Internet of Things.
34. Juan A. Añel, Diego P. Montes, Javier Rodeiro Iglesias. (2020). Chapter 3 From the Beginning
to the Future, Springer Science and Business Media LLC.
35. Karthik, S., Kumar, S., Prasad, K. M., Mysurareddy, K., & Seshu, B. D. (2020). Automated
Home-Based Physiotherapy. In 2020 International Conference on Decision Aid Sciences and
Application (DASA) (pp. 854–859). IEEE.
36. Prasad, K. M., Pradeep, K. N., Kashyap, S. S., & Anusha, V. S. (2021). Time Series Data
Analysis using MachineLearning-(ML) Approach. Library Philosophy and Practice, 1–7.
An Analytical Study of the Role of M-IoT
in Healthcare Domain

Bidisha Chanda, Pradeep Kumar Mallick, and Gyoo-Soo Chae

Abstract This study describes mobile health in context of Internet of Things. The
Internet of Things applied in development of healthcare systems have reached an
evolutionary process. It describes how IoT was introduced in healthcare. It mainly
focuses on how mobile IoT gained prominence in the healthcare and medical sectors.
This chapter brings to light several applications of M-IoT in healthcare in measuring
body temperature, monitoring blood glucose level, ECG, etc. The effect of Internet
of Things has been evolving in every aspect of living; however, the effect it has at the
healthcare industry is massive because of its growing need and accuracy. The Internet
of Things will become extra dominant, and it will be accompanied by the capabilities
of mobile computing. It enhances the capability of IoT in healthcare surroundings
via means of bringing a large aid within the scope of mobile fitness (m-fitness).
Mobile computing assists IoT packages in healthcare, contributes to the present day
within the healthcare industry. Mobile computing provides aids to IoT applications
in healthcare. It has also mentioned about various benefits and advantages of this
system. Security is very critical in IoT-based m-healthcare systems. So this study also
includes a section that addresses the privacy related issues, security, and challenges
of the M-IoT system. Overall integration of M-IoT with healthcare will drastically
reduce healthcare costs and unnecessary hospitalizations.

Keywords Internet of Things (IoT) · Mobile IoT · Healthcare · Blood pressure ·

Regression

B. Chanda · P. K. Mallick (B)

School of Computer Engineering, Kalinga Institute of Industrial Technology, Deemed to be
University, Bhubaneswar 751024, India
B. Chanda
e-mail: [email protected]
P. K. Mallick · G.-S. Chae
Division of Smart IT Engineering, Baekseok University, Cheonan-si, South Korea
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 75
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_4
76 B. Chanda et al.

1 Introduction to IoT and Healthcare

Conventional healthcare and medical services are based on the technique of oper-
ating with common and qualitative statistics, which no longer work well. The drastic
boom in population has introduced many demanding situations within the health-
care device which in turn has made the prevailing healthcare services inefficient.
This is where Internet Of Things comes to the rescue [1]. A network of gadgets and
different devices, embedded with sensors, electronics, and community connectivity
and software which permits those gadgets to gather and transfer statistics is what we
call Internet of Things. The effect of IoT within the discipline of scientific research
is significant. IoT has enabled advent of larger and scientific treatments. This guar-
antees that the patients may be monitored from their homes, which reduces the
haste to visit doctors. Various gadgets use IOT to make development within the
quality of the healthcare and clinical services acquired by the patients [2]. IoT-based
healthcare offerings target at imparting rich consumer satisfaction at low cost. From
clinical perspective, actual time statistics might permit quicker and rational deci-
sion making. The Internet of Things (IoT) has made it feasible to carry a world of
opportunities within the discipline of medication and healthcare services. Since the
involvement of IOT in healthcare services humans has extra control on their lives and
remedies because it includes actual time statistics. It has been anticipated through
Cisco Systems that by 2020, IOT will encompass almost 50 billion gadgets linked to
the Internet. These gadgets consist of smart devices like actuators, mobile phones,
and computer systems that have the potential to switch statistics in a secured way
[3]. A form of gadgets along with RFID tags, clinical gadgets, cell phones, and so
on accommodates the surroundings of Internet of Things. These devices are related
through particular identifiers and interact with each other. The transfer of facts among
numerous IOT devices results in an ultra-modern derived fact. Healthcare monitoring
devices and wireless communications and IoT infrastructure provide considerable
improvement within the field of healthcare and medical offerings [4]. There are clas-
sifications of uses of IOT and below listed are ten such fields where the use of Internet
of Things is widely used.
• Participatory sensing: It involves mobile system which encourages users to main-
tain records and distribute information and thus helping in recreation of advanced
knowledge.
• Eco-feedback: It makes the use of mobile applications that generates feedback
for the customers about various environmental events or about their personal
consumption.
• Actuation and control: Applications like mobile application and other mobile
applications that control some physical movements or actuates some instances.
• Health: Mobile applications are used to get health status of the sensors that are
attached to the patient body. These applications update the doctors regarding the
current status of the patients.
An Analytical Study of the Role of M-IoT in Healthcare Domain 77

• Sports: Includes an aggregate of bodily sensors and cellular programs which can
be used all through game activities to file diverse metrics and assist to enhance
the overall performance of the athlete user.
• Farming: Related to smart farming practices to enhance productivity, control of
farm animals and increase consumer delight and transparency.
• Gaming: This class is about digital video games which remember the physical
presence or fame of the mobile user to decorate the gaming experience.
• Transportation: A developing area in which the sensing capabilities of the cellular
smartphone are harnessed for better experience and comfort of parking.
• This is the wider category because it pertains to efforts which prioritizes interacting
with web-enabled physical entities to be had with inside the nearby environment,
as an instance tagging technologies.
• Social interaction with people: Mobile applications play an important role on
interacting with the everyday lives of the people by interacting with them socially.
The first section of this chapter is about introduction of IoT. The next section
elaborates integration of M-IOT in healthcare. It describes how the M-IOT has been
integrated with healthcare so that it can offer improved healthcare facilities and
patients will enjoy these facilities at low cost. The subsequent segment describes
about the operating precept of M-IOT and the way the IOT tracks gadgets and assist
to preserve the actual time facts. Certain current technology within the area of m-IoT
and m-fitness has additionally been highlighted within the following segment which
offers an assessment among the exceptional technology. The next section offers with
the diverse packages of IOT in healthcare. The next part is all about literature surveys.
The seventh segment offers a case look at on diverse advantages of IoT in health-
care. The subsequent segment deals with disadvantages, demanding situations and
protection issues of IoT and M-IoT in healthcare. It offers the demanding situations
of IOT in information just like the security of information. Next segment has case
studies. At the end of this study, the future scope of IOT in healthcare is in short
discussed [4] (Table 1).

2 Integration of M-IOT in Healthcare

The new technological tendencies related to cellular computing and its applications
make it affordable and sufficient to combine it with healthcare and clinical services.
The main idea behind the use of mobile IOT in healthcare sectors is to obtain better
and error-free readings by making effective and efficient utilization of all possible
limited and available resources so as to provide quality services to the patients.. The
cellular computing impacts the IoT with the aid of using imparting diverse offerings,
programs, and gadgets. These smart devices are successful to screen numerous fitness
frameworks like blood pressure, degree of sugar in blood, and pulse rate. These
mobile fitness gadgets, for the reason of doing those tasks, are related to the IoT server.
The use of mobile devices in collecting and tracking real-time data from patients and
78 B. Chanda et al.

Table 1 Sensing technologies at each category of IOT/WOT-based mobile applications

Category Popular technologies Types of data analysis Market value
used
Participatory sensing GPS, camera, Location-based search, N/A
microphone map visualization,
information sharing
Actuation and control Smart electricity Rule based inference, $1.3–4.6 Billions
meters, smart adaptive reasoning,
appliance, light optimization
switches, smart
factory
Health Body area sensors, Personalization $110–900 Billions
GPS, microphone, profiling, anomaly
accelerometer, detection, emotion
communication, and detection, ML
conservation sensing
Sports Body area sensors, Personalization $200–450 Billions
motion sensors, GPS, profiling, anomaly
pressure shoes, detection, historical
pedometers comparisons,
performance
visualization, statistics
Agriculture Wearable collars, Historical comparisons, $140–200 Billions
GPS, barcodes, RFID stream data processing,
tags retrieval of info
Gaming Accelerometer, Activity recognition, $450–635 Billions
camera, pedometer, machine learning,
barcodes information sharing
Transportation GPS, compass, career Location-based search, $500–740 Billions
connectivity, RFID big data analysis,
tags stream data processing
Interaction with things NFC technologies, Location-based search, $70–150 Billions
RFID tags, barcodes, information retrieval,
QR codes information sharing
Social interaction with Bluetooth, PS, Location-based search, $170–450 Billions
people camera, microphone information retrieval,
information sharing
Eco-feedback Camera, energy Historical comparisons, $200–750 Billions
monitors, smart stream data processing,
meters, barcodes, retrieval of info
environmental
sensors, e.g., air
quality
An Analytical Study of the Role of M-IoT in Healthcare Domain 79

storing it on the Internet is referred to as m-health [5]. This stored data can be collected
by mixed group of clients like the doctors, medical staff, health insurance companies,
hospitals, etc. The prime goal of m-health is to obtain information immediately to
diagnose illness, keep a track of diseases, take urgent medications for emergency
cases. Mobile health is especially important in rural areas where it becomes difficult
for the doctors as well as the patients to maintain regular clinical visits. Below given
are some healthcare utilization trends:
(1) It is believed by more than two-thirds of healthcare technology executives that
use of m-health will have great impact in the future
(2) According to health IT outcomes, global m-health revenue will reach $49.12
billion by 2020.
(3) It is believed by 40% of surveyed physicians that m-health applications such
as tele-health could cut down on a number of clinical visits.
(4) According to 93% of the doctors m-health apps can improve patient’s health
and having a m-health application connected to emergency health services is
valuable.
Mobile fitness will develop through creating country-based e-Health techniques
that include it into the present e-Health domain. Policies want to be complemented
with the aid of using standards, architectures, and stable partnerships to assist m-
health tasks mature and comprehend their complete potential—utilizing cell and
Wi-Fi technology to enhance fitness and well-being. Two important instances are m-
fitness and cellular learning (m-learning). Most m-fitness programs are utilized by the
healthcare specialists for diverse duties like analysis of disease, drug reference, and
clinical calculations. Utilizing IoT-MD to screen continual ailments is a brand new
approach of gathering a step forward in clinical records. Internal video display units
and wearable sensors provide a regular collection of records to people and clinical
providers, which alert for adjustments in fitness situations even as constructing a
clinical profile for destiny healthcare needs. Cell telephones and cellular gadgets
also are adapting to IoT-MD through medically authorized attachments and apps used
to screen blood samples or provide reminders to carry out improvement exercises.
Figure 1 gives a graphical representation of the global market size share of M-IOT in
heal in dollars. It presents a clear picture of what percentage of mobile IOT is used
in various healthcare domains.

3 Working Principle of M-IOT in Healthcare

The precept of m-IoT is primarily based on tracking gadgets. These devices track
the patient’s condition remotely after which send this precise records to a back-end
device or a cloud based device which stores it. Then the back-end device minutely
examines and analyzes the records and generates the precise indications. The clini-
cians then come across the disease according to the acquired alert and may effortlessly
determine any case of emergency and approximately the sort of remedy this is for use
80 B. Chanda et al.

Fig. 1 Application categories

for the treatment [6, 27–35]. The tracking tool needs not be a scientific tool; however,
it may additionally be gadgets or devices like smart watch or a few sort of cell tele-
phones that can hook to the human beings. Recently cell telephones have emerged to
be the handiest IOT gadgets. IoT gadgets accumulate critical records and switch that
records to medical doctors for real-time tracking and notifies human beings approxi-
mately the severity of the disorder through cellular apps and lots of different gadgets
connected with it. Medical gadgets sense, reveal, and transmit private records via
the gadgets to steady clouds or platforms in which findings can also additionally
be analyzed to create actionable, records-related measures and recommendations.
Figure 2 gives a diagrammatic representation about the working principle of M-IoT
in healthcare. Here, all the data collected from various sources like hospitals, labora-
tories, etc, are stored in a back-end system or the clouds. All the databases are stored
in the cloud. While examining a patient, a doctor needs to retrieve the data about
the patient from the database itself. The doctor writes a prescription using the E-
prescribing system which is again stored in the clouds. On the other hand, whenever
required by the doctors, a pharmacist accesses the database to get details about the
prescribed medicines. The hospital can use the back-end system or the cloud to store
patient health records as well as details regarding contacting the patents for care.
An Analytical Study of the Role of M-IoT in Healthcare Domain 81

Fig. 2 Global IoT in

healthcare market size, by Sales
application
11%
20%
12%

19%
21%

17%

Telemedicine Medication Management

Clinical operations Patient monitoring
Connected imaging Others

4 Comparative Analysis of Existing M-IoT Technologies

Adaptive Network Technology(ANT)

It is an ultra-low-power, short-variety Wi-Fi generation that has the potential to permit
gadgets to run on a battery for years. It is a multicast Wi-Fi sensor generation that lets
in the tracking gadgets to connect with everyone effortlessly in sensible and precious
ways. The ANT protocol is the generation that powers the health ANT+sport, health
and health product ecosystem. It has the potential to aid AES-128 encrypted chan-
nels for stable connections and may join topologies like peer-to-peer, star, tree, and
mesh typologies. These gadgets can also additionally use freq between 2400 and
2524 MHz; however, 2457 MHz is excludeds. Its number one purpose is to permit
clinical and sports activities sensors which allows conversation with a show unit, for
instance, an eye or cycle computer.

Zigbee
This generation is solely designed to cope with the specific desires of low-value,
low-power Wi-Fi IoT networks, the sensors and manage gadgets that employ low-
value connectivity. These Zigbee’s WPANs perform at 868 MHz, 902–928 MHz, and
2.4 Ghz frequencies inside 10-a hundred meter inside range. Zigbee is able to control-
ling and monitoring. Zigbee networks may be prolonged with the usage of routers
and permit many nodes to interconnect with every other for constructing an extensive
region network. Zigbee can join topologies like point-to-point, point-to-multipoint,
and mesh networks. Zigbee operates on IEE 802.15. four radio specification. Zigbee
has many programs in healthcare monitoring.
Near field communication(NFC) Near field communication (NFC) is a contact-
less communication technology primarily based on a radio frequency (RF) that uses
a base frequency of 13.56 MHz. NFC era is flawlessly designed to change statistics
among gadgets through an easy contact gesture that makes use of magnetic discipline
induction to permit conversation among gadgets when they are touched together or
delivered inside some centimeters of every other. The NFC tag may be incorporated
82 B. Chanda et al.

with smart mobile phones and can be used as a digital fitness file to keep the patient’s
clinical history, prescriptions and real-time sensor data.

Wi-Fi
Wi-Fi is known to be the most power efficient. It is best known for ease of deploy-
mentt. Wi-Fi connections can be integrated with mobile phone to provide much
better and efficient healthcare services. Standard Wi-Fi, being the plain preference
for IoT, has barriers in each variety and electricity efficiency. These shortcomings
was addressed by IEEE for publishing specifications for 802.11ah and 802.11ax.

Bluetooth
Bluetooth makes use of short-wavelength UHF waves among 2402 and 2480 MHz.
Bluetooth helps each star topology in addition to mesh topology. It additionally helps
factor to factor, one-to-many, many-to-many topologies. If the low-energy require-
ments, theoretically limitless scalability, and self-recovery reliability of emerging
Bluetooth mesh networks are given, well-designed Bluetooth may be incorporated
in cell to develop applications that are powerful for such indoor-asset-tracking
scenarios. But the identical Bluetooth findings could be gold standard for accom-
plishing different commercial enterprise goals. Smart tracking gadgets may be related
with Bluetooth can assist the clinicians to asset tracking.
Table 2 presents a comparison between different technologies of mobile IoT.
The comparison is based on various factors like frequency, channels, modula-
tion, complexity, power profile, extendibility, etc. There are other technologies like
LPWANs, RFID. Only small blocks of data can be sent by LPWANs at much lower
rate so they are suited for use cases that do not require high bandwidth and are not
time-sensitive. Compared to LPWAN, Zigbee ensures higher rates of data transfer.
As it uses mesh configuration, it is much less power efficient.

Table 2 Comparison of different M-IoT technologies

Low-energy Zigbee NFC Low power Wi-Fi
bluetooth
Frequency (MHZ) 2402–2482 2402–2482 13.56 2400–2500
902–928
Channels 3 16 1 3
Modulation GFSK BPSK and QPSK ASK 64 QAM
Max potential data 1.1 Mbps 251 Kbps 25 Kbps 55 Mbps
rate
Range 11 m 101+m 11 cm 31 m
Power profile Days Month/year Month/year Hours
Complexity Complex Simple Simple Complex
Nodes/master 8 65001 1+1 –
Extendibility No Yes No Yes
An Analytical Study of the Role of M-IoT in Healthcare Domain 83

Radio frequency identification (RFID) works using radio waves for transmitting
small amounts of data from an RFID tag to a reader within a very short distance.

5 Applications of M-IOT Devices in Healthcare

Continuous glucose monitoring(CGM)

An advanced way that monitors glucose readings is continuous glucose monitoring.
The glucose level in the patients’ blood is monitored by the CGM device. The connec-
tion of the tool is made to an insulin pump with an automated suspension of insulin
infusion. It connects to an analytic platform to monitor health data over time to
recognize patterns/triggers for abnormal levels. There are systems that collect and
store glucose readings in hospital database system, which in turn makes it easy for
clinical staff and the doctors to access real-time data from their patients. Statistics
have demonstrated that CGM may reduce patient’s long-term complications between
40 and 75% of the time. These gadgets are computerized and may file blood strain
for 24/7, while sufferers can do their regular each day activities. A photoplethysmo-
graphic (PPG) primarily based technique is used to screen blood strain. The use of
wearable sensor is provided by wearable, cuff-less PPG-based blood pressure monitor
with novel height sensor. Thus, IoT applications can help to generate correct glucose
level reading with minimized error.

Electrocardiogram(ECG)
An easy test that diagnoses heart disease is electrocardiogram (ECG). A normal
heart’s activity tracking with the aid of using a hand-held ECG tool is powerful and
affords long-term financial savings for heart patients. Existing electrocardiogram
(ECG)tracking gadgets which include Holter are not convenient for long-time period
use because of their length and twist in the wires. Wireless ECG tracking gadgets
that could hook up with IoT server are rising into market. A wearable ECG tracking
and systems that provide continuous alerts like the “I-Heart” that constantly video
display units ECG is advanced. The I-Heart video display units affected person’s
ECG and troubles an alert to the affected person if it identifies odd behavior of heart.
The tool makes use of a Wi-Fi ECG sensor and a smart phone. Thus integrating
IoT technologies heart behavior can be easily tracked with accuracy and minimized
errors.

Body Temperature Monitoring

MosChip has designed an IoT- based hardware and software program for health-
care structures which guide the healthcare employees and caretakers to record their
patients’ fitness situations with the help of the Internet remotely. The temperature
patch is hooked up to the area of the frame to make sure of the correct temperature.
The records are amassed from a sensor and dispatched to the telephone over Blue-
tooth 4.0. The smart software designed for iOS and Android structures will carry out
a record evaluation and ship real-time records to the unique cloud software via over
84 B. Chanda et al.

a GPRS/Wi-Fi backbone. The medical doctors can examine the records and endorse
medicine in real-time. The software consists of fitness reviews for faraway diagnosis,
alarm onset temperature limit, agenda indicators to take medicine, and scheduled
medical doctor visits. Another technology is the use of Arduino in measuring the
body temperature. It uses sensors connected to the Arduino board
Other than these, there are several other applications of IoT in healthcare like blood
pressure monitoring, oxygen saturation monitoring, ECG monitoring, etc. These
applications when integrated with the healthcare sectors prove to be efficient and
accuracy.

6 Benefits of M-IoT

Minimized errors—IOT checks risk of errors. It also deals with precise and exact
collection of data, automated workflows and minimized waste. IoT technologies are
good at minimizing the human mistakes without removing human touch. Suffering
from exhaustion from long hours of work, medical staffs often commit medical-
related mistakes. This is where IoT comes to the rescue.
Better patient experience—A connected healthcare system satisfies the needs of
each and every patient. Committed procedures, improved treatment options, and
upgraded diagnosis accuracy make for a better patient experience. Patients get satis-
fied with the quality of medical facilities they receive and also with the efficiency of
the technologies as well as the accuracy.
Reduced expenses—With IoT, patient no longer needs to be physically present in
a clinic for visits, instead monitoring can be done in real-time. Connected home
care facilities will also bring down the expenses related to hospital stays and re-
admissions. Thus, this efficient autonomous system can reduce expenses when it
comes to patient cost savings because of fewer health center trips in addition to
elevated diagnostics and treatment.
Improved ailment management—With real-time facts, healthcare carriers can
constantly screen patients. This way that they are able to spot any disorder earlier
than it spreads and turns into serious. IoT promises in helping to improve the health
of the patients with chronic diseases. It provides an error-free monitoring system
using sensors, gateways etc.
Remote tracking—Real-time far off tracking through related IoT gadgets and clever
signals can diagnose illnesses, deal with sicknesses and shop lives in case of a clinical
emergency. With the help of remote monitoring and virtual visits, IoT technologies
keep patients better connected to the doctors.
Prevention of diseases—Smart sensors examine fitness conditions, way of life selec-
tions and the surroundings and advocate preventative measures, a good way to lessen
An Analytical Study of the Role of M-IoT in Healthcare Domain 85

the prevalence of sicknesses and acute states. Frequent monitoring by the IoT devices
helps in keeping a track of the present condition of the body.
Accessibility—Doctors can view all of the essential records on command and take a
look at real-time affected person situations without leaving their office. The database
is stored in the cloud from where the doctors can access the required data anytime.
Medical facts accessibility—Accessibility of digital scientific statistics permits
sufferers to get hold of fine care and assist healthcare vendors make the proper
scientific choices and save you complications.
Improved remedy management—IoT gadgets assist the management of medicine
and the identifies the remedy and decrease clinical error. IoT has plethora of benefits
which ensures healthcare system to be more reliable, fast, and quickly accessible.
Improved healthcare Management—Using IoT gadgets, healthcare government
can get precious statistics approximately gadget and workforce effectiveness and
use it to signify innovations.
For studies purpose—IOT gadgets have an excessive potentiality for clinical
research purposes
And they are capable of collect and analyze a big number of useful data.

7 Challenges of M-IoT

Still there are many issues that need to be checked before the concept of IoT is widely
accepted. It is claimed by Atzoria et al. that people will not take risk by IoT adoptions
as long as there is no public confidence that IoT will not cause serious threats and
violations to their privacy.
Security issues—There are instances where the systems get hacked. There is a dire
need to focus on the data security and unauthorized access. There is a risk that fraud
people may also get entry to centralized structures and recognize a few merciless
intentions.
Worldwide healthcare regulations—The government medical establishments that
are planning to integrate IOT in their workflow should follow the guidelines that are
already issued by international health administrations
Data overload and oversight—Studies show that aggregation of data is not eassy
due to the use of different communication protocols and standards. Inspite of data
overload, IoT devices still have an enormous number of data. The statistics gathered
through IoT gadgets are applied to benefit important insights. However, the quantity
of statistics is so superb that deriving insights from it have become extraordinarily
tough for doctors which, in the long run impacts the decision making
86 B. Chanda et al.

Human error—Exchange of information among medical staff can introduce human

errors, which can end up in risking the life of patient. Communicating with patients
who lacks verbal communication skills is another great challenge, and information
collection problems are even more challenging.
Integration of multiple devices and rules—Integration of a couple of gadgets
additionally causes hassle in the working of IoT within-side the healthcare sector.
The motive for this disturbance is that tool producers have not reached a consensus
concerning communique protocols and standard.
Existing software infrastructure is outmoded—IT infrastructures in many hospi-
tals are obsolete. They do not support the idea of integrating the IOT devices for
betterment of healthcare services. Therefore, IoT in healthcare has to face a lot of
challenges and complexities before producing the better and efficient results. No
doubts if the complexities can be reduced using certain methodologies, in Fig. 3, a
graphical representation of factors contributing to the challenges is presented (Fig. 4).

Fig. 3 Working of M-IOT

An Analytical Study of the Role of M-IoT in Healthcare Domain 87

Fig. 4 Categories of issues contributing to challenges

8 Relevant Studies on M-IoT in Healthcare

Skin disease detection from the image processing is a crucial subject technique,
however additionally for computational intelligence and image processing. There
also exist studies indicating the pores and skin tumors for chemical evaluation. In
[7], the chapter mainly focused on how plasmacytoid dendritic cells, Langerhans
cells and inflammatory dendritic epidermal cells, influence anti-viral pores and skin
protection technique. Here, the main theme of the chapter mentioned the diagnosis
which is preferred for pores and skin cancer detection primarily based on the evalua-
tion of auto-antibodies and mucous membranes including usage of commercialized
tools and immunofluorescence microscopy. The authors provided how the antigens
influence pores and for serological analysis. In [8], the chapter mainly mentioned a
technique to locate atopic dermatitis, psoriasis, and call dermatitis characterized with
the aid of using transcriptomic profiling. Potential sickness become detected with the
aid of using the protein expression levels. In [9], a method was proposed for cancer
pores and skin cancers detection. Authors mentioned a device which integrates deep
learning knowledge of with pores and skin lesions ensemble technique. In [10], the
study provided optoacoustic dermoscopy version primarily depends on a feature of
excitation power and perforation intensity measures for pores and skin evaluation
with the aid of using extreme imaging. It became proven how to research absorp-
tion spectra at more than one wavelengths for visualization of morphological and
practical pores and skin features. A method to locate pores and skin cells and cancer
primarily based on deep learning knowledge of version become accepted in [11]. The
authors have mentioned segmentation with regards to the middle of cancer with the
aid of using lesion indexing network which makes use of fully convolutional residual
network as a chief detection technique. In [12], we are able to discover an updated
evaluation over diverse gadget gaining knowledge of techniques carried out to pores
88 B. Chanda et al.

and skin analysis and detection. While in [13], we are able to discover an extensive
contrast of cell apps for pores and skin tracking and cancer detection. The authors have
drawn a comparison between numerous applications, mentioned their ability in effi-
cient image processing and sensing technology and additionally provided elements
of ethical, pleasant, and obvious improvement of apps for scientific use. Ahn et al.
[14] mentioned a device by which phsiological alerts can be measured which includes
ECG and BCG. It makes use of a smart chair that is capable of sensing non-limited
bio-alerts. Almotiri et al. [15] themed a chapter which discuss about the devices of
m-fitness that takes the help of m-gadgets to gather real-time records from sufferers
and stores it on network servers. This records may become useful for the scientific
checking of sufferers and is attained by taking help of some of wearable gadgets and
frame sensor community. Barger et al. [16] designed a smart residence facility that
enables the usage of a sensor community to reveal and analyze the movements of
the patient and a prototype of the equal is likewise being tested. The first and the
most important goal in their work is to test if their device is succesful to outsmart the
behavioral styles and feature mentioned. Chiuchisan et al. [17] deviced a framework
to eradicate the spread of diseases from the affected person in smart ICUs. Dwivedi
et al. [18] evolved a framework in an effort to stable the scientific records that needs
to be communicated over the Internet for Electronic Patient Record (EPR) struc-
tures wherein they suggest a multi-layered architecture of healthcare records device
framework that is an aggregate of public key infrastructure, smartcard, and biomet-
rics technology. Gupta et al. [19] proposed a version which measure the fitness of the
affected person the usage of Raspberry Pi and may be of an extraordinary usage for
the medical sectors and sufferers in addition to their own circle of relative’s members.
Gupta et al. [20] introduce a method that makes use of Intel Galeleo improvement
board that gathers the different records and updates it to the database from wherein
it is accessed by the doctors

9 Discussion of the Use M-IoT in Cancer Detection

Cancer is the out of control increase of skin cells. It develops when an unpaired DNA
damages the pores and skin cells due to exposure to the ultraviolet radiation of the
tanning beds and sun, causing mutations (genetic defects) which leads to the skin cells
to multiply swiftly ensuing the formation of malignant tumors. Sometimes, the pores
and skin cancers may even unfold and harm the nearest cells [21]. Also, in a few cases,
pores and skin most cancers might also additionally develop on vital organs. [22]
Sun exposure may be referred to as one of the fundamental purpose for improve-
ment of tumors in pores and skin cells, even though there are numerous different
elements like surroundings threats, radiation evaluation, or even inheritance can
play a role. The Internet of Things and data and generation carried out in improve-
ment of healthcare and clinical structures have reached an evolutionary process.
Nowadays, skin related cancers result in the demise of a huge population which
is consistent with the healthcare assessment system. To take a look at the cost of
An Analytical Study of the Role of M-IoT in Healthcare Domain 89

the treatment, there is a demand for a computerized device to assess an affected

person’s risk of cancers. The discussed system includes the detection of pores and
skin cancer with excessive degree of accuracy and rapid execution time with the
utilization of support vector system and particle swarm optimization algorithm and
moreover developing the Twitter Doctors Community if you want to transfer the
affected person’s health related information to the individuals in their family the
use of mobileular phone and different clever devices or doctors with the assist of
Arduino. Additionally, doctor can suggest appropriate remedy for the patient and
can interact with sufferers with the aid of IoT and moreover enhancing and making
sure safety-based healthcare system in which affected person’s statistics bases are
measured in a strong manner with the use of Anti SQL injection to save server
statistics from being hacked.
Discussed Methodology: The discussed methods describe how skin cancer can be
detected with more accuracy and efficiency by implementing IoT techniques. These
methods include the following methodologies:
Image Acquisition: Image acquisition dermoscopic images are basically digital
images of magnified pores and skin lesion, captured with traditional camera well
equipped with unique lens extension. The lens connected to the dermatoscope act
like a microscope magnifier with its very mild source that illuminates the pores and
skin layers evenly. Dermatologist can create accurate documentation of amassed
photographs, maintaining a sequence of steps for analysis, wherein photographs are
processed that allows you to extract records that can later used to categorize those
images.
Pre-Processing Techniques: Pre-processing of image is an important step that
permits you to reduce noises and improve the standard of the image. It is impor-
tant to be implemented to limit the inside aspect the background having an effect
on of the result. The main motive of this step is to enhance the standard of cancer
images via a way of method of casting-off unrelated and surplus elements inside
aspect the background of the image for in addition image pre-processing. Good
choice of pre-processing strategies can substantially enhance the accuracy of the
system. In segmentation techniques, the process of segmentation includes to the
separation of an image into disjoint regions that are uniform with respect to some
property such as color, luminance, and its texture. Thresholding-based segmentation
encompasses bi-stage and multi-thresholding. Thresholding technique includes: (1)
Histogram and (2) adaptive thresholding. In region-based segmentation, the image
is splitted into smaller and smaller additives and then merged into sub-images which
can be adjoining and comparable in a few sense. It consists of statistical location
merging, multi-scale location.
Feature Extraction Using PSO Algorithm: In automatic analysis of pores and skin
lesions, characteristic extraction is primarily based on the so-referred as ABCD-
rule of dermatoscopy. The extracted parameter is optimized to the use of particle
swarm optimization algorithm. Particle swarm optimization (PSO) is a stochastic
international optimization technique evolved through Eberhart and Kennedy in 1995.
90 B. Chanda et al.

Fig. 5 Block diagram of the discussed methodology

In PSO, a set of solutions traverse the problem space with a speed primarily based on
their own experience and the experience of their neighbors. During every spherical of
traversal, the speed, thereby the location of the particle, is up-to-date primarily based
totally at the above parameters. This technique is repeated until a most beneficial
answer is obtained (Fig. 5).
Table 3 provides a contrast among the readings taken through the conven-
tional method as opposed to readings measured with the proposed technology. This
effortlessly concludes the efficiency, correctness and acccuracy of the proposed
method.
Observation: This version proposes approximately a stable IoT primarily based
totally pores and skin most cancers detection system. Compared to the traditional
bioscopic approach, this proposed approach proves to be more efficient. The diag-
nosing method uses superior strategies like the digital image processing techniques
combined with support vector machine similarly to the particle swarm optimization
algorithm for the class of the photograph from normal skin photograph. The conse-
quences confirmed accuracy. The alert concerning the illnesses may be despatched
trough twitter alert or mail alert. Through ESP8266 Wi-Fi module, the database is
despatched to the IoT platform. By the use of SVM_PSO algorithm, the cancerous
cell is separated from healthy pores and skin.
An Analytical Study of the Role of M-IoT in Healthcare Domain 91

Table 3 Comparison of
1 18 12
readings between traditional
approach of skin cancer 2 20 15
detection versus discussed 3 25 21
methodology
4 28 21
5 32 26
6 17 13
7 22 17
8 30 25
9 28 22
10 25 20
11 23 18
12 19 15
13 20 16
14 23 18
15 26 22
16 33 28
17 16 13
18 27 22
19 32 28

10 M-IoT in Blood Pressure Measurement from Heart Rate

High blood pressure also called hypertension will increase the risk of mortality
for the adults. [23–26] Generally, high blood pressure is described as blood strain
above 140/ninety, and it will become extreme if the strain is above 180/120. A latest
observation has proven that people of China who have an excessive heart rate in
among 80–90 bpm are much more likely to have a short lifestyles in comparison
to the humans with heart rate among 60–69 bpm. Thus, a linear relationship exists
among heat rate and blood pressure. Systolic blood pressure measured from the heart
rate based on IOT is discussed that makes use of LM358 sensor and NRF module to
calculate the blood strain the usage of regression techniques.
Devices and components: Some of the components and sensors used in constructing
the devices are Arduino Uno(AT328), heart rate sensor(LM328), NRF module, LCD
display, and power supply. This device aims at calculating blood pressure from
measured heart rate. Arduino Uno(AT328) works as the main processing unit for
the proposed device. The arrangements for the components have been shown in the
given figure. The power supply block sends the power supply to the device. The
sensor used for the heart rate measurement records the heart rate data and transfers it
92 B. Chanda et al.

to the Arduino Uno for further processing followed by which the transmitted infor-
mation is sent serially to the NRF module, and the readings of the blood pressure
and heart rate are shown on the LCD.
Discussed Methodology: In this discussed methodology, more than some data sets
were collected from students within age group of 20–28. Figure 6 is a block diagram
that represents the arrangement of devices in the discussed methodology. Their corre-
sponding blood pressure and heart rate values were collected from the heart rate
sensor via Aurdino Uno. Figure illustrates the discussed regression model. At first,
the information of heart rate and its corresponding blood pressure is gathered after
which divided the data set into three classes inclusive of excessive blood pressure,
excessive blood pressure, ordinary blood pressure. The degrees for every class are
given in Table 4. Based on the pattern shown, a regression model is generated in
Table 4. This table is developed to show regression model for each class. From the
generated regression pattern, equations are made which are enforced on the device
to get the measure of blood pressure from the heart rate. In Fig. 7, the block diagram
represents the arrangement to obtain the regression pattern (Tables 5 and 6).
It may be visible from Table 7 that the distinction among blood pressure and blood
strain calculated with the aid of using the proposed version may be very minimal. This
offers accuracy to be 99% accurate. On the alternative hand, comparable statement
has been made for class 2 and class three sufferers. The accuracy rate of class 2
sufferers has been 90% while that of class three sufferers turned into discovered to
be 89%. Table gives the effectiveness of the proposed version. Thus, this approach
proves to be correct in phrases of the calculations and accurate readings of the blood
strain.

Fig. 6 Heart rate measure

by using this device

Table 4 Category of blood

Category/class Blood pressure (BP) Range
pressure ranges
First category/class 1 High BP BP>=140
Second category/class Pre high BP 120<=BP<140
2
Third category/class 3 Normal BP BP<120
An Analytical Study of the Role of M-IoT in Healthcare Domain 93

Fig. 7 Regression model

flow

Table 5 Regression model

Class Regression model
for blood pressure using heart
rate Class 1 Y= 0.382X + 107.220
Class 2 Y= 0.2410X +105.25
Class 3 Y= 0.2480X + 97.0865

Table 6 Calculated BP
Original SBP—calculated SBP (mm Status
considered status
Hg)
0–3 Very accurate
4–5 Slightly inaccurate
6–67 Moderately inaccurate
>7 Very inaccurate

Table 7 Category 1 BP data in this method

HR (BPM) Original SBP (mm Hg) Calculated SBP via Original SBP—calculated
proposed method (mm Hg) SBP (mm Hg)
106 149 149.473 0.437
94 147 145.224 1.776
105 149 149.054 0.054
106 151 149.437 1.563
87 143 142.16 0.84

Observation: A method that can be used to measure blood pressure from heart rate
with the help of IOT has been discussed in this chapter. For this tested analysis„
data sets were collected. Patients were categorized into three classes. This regression
model which has been discussed provides near acceptable results with great correct-
ness. Furthermore, these models can be applied to determine skin moisture and body
temperature with the inclusion of some other required parameters.
94 B. Chanda et al.

11 Conclusion and Future Scope

With better analytics, faster broadband speeds, more competitors in space and effi-
cient technical improvements, the future of mobile IoT has a good opportunity to
make a positive impact on the healthcare industry. Despite of the worry of privacy and
cyberattack, IoT continues to get promising day by day, IoT packages will develop
and become more efficient within the coming years. Ultimately, we are able to say
the IoT future scope or the destiny of IoT could be very bright. There will a number
of IoT gadgets in an effort to be coming shortly. These IoT devices prove to be very
efficient and accurate. It will routinely make the corporations take strain and make the
whole healthcare system lot secure for the users. However, the risk will stay, with the
big boom within the IoT gadgets usage. It has additionally expanded affected person
engagement and pleasure as interactions with doctors have become more efficient
and ended up being less complicated. This can also additionally result in safety and
liberty problems getting magnified through the IoT gadgets.
The future is going on in the front of our eyes. Nowhere is that so obvious as in IoT
within the clinical sector where, over years, many technological improvements have
happened, inclusive of feature-wealthy gadgets or micro-sensors. Even though IoT
has demanding situations and limitations, it will now no longer prevent the generation
from growing. IoT additionally has a chief effect on lowering healthcare charges
drastically and enhancing remedy outcomes. The demand for IoT in healthcare is
massive and it is able to assist the enterprise immensely. It can attain each affected
person from everywhere in the international and join doctors with patients. Yet, even
greater is anticipated to return back within the subsequent decade. Recent forecasts
from the US Census Bureau record that through 2025, 1.2 billion human beings can
be elderly. This reality suggests the population’s destiny desires for fitness wearables,
the call for so as to be better than ever. According to the information provided by
Grand View Research, it is expected that the global IoT remote health monitoring
market will make more than $300 billion by 2022. The technology thus holds a future
by providing independent and mobile health monitoring while reducing stress and
charges to visit doctors and health personnel.

References

1. Mishra, S., Tripathy, H. K., Mishra, B. K., & Mohapatra, S. K. (2018). A succinct analysis of
applications and services provided by IoT. In Big Data Management and the Internet of Things
for Improved Health Systems (pp. 142–162). IGI Global.
2. Mishra, S., Patel, S., Panda, A. R., & Mishra, B. K. (2019). Exploring IoT-Enabled smart
transportation system. In Goyal, D., Balamurugan, S., Peng, S., & Jat, D. S. (Ed.), The IoT and
the Next Revolutions Automating the World (pp. 186–202). IGI Global. http://doi:10.4018/978-
1-5225-9246-4.ch012.
3. Mishra, S., Mishra, B. K., Tripathy, H. K., & Dutta, A. (2020). Analysis of the role and scope
of big data analytics with IoT in health care domain. In Handbook of Data Science Approaches
for Biomedical Engineering (pp. 1–23). Academic Press.
An Analytical Study of the Role of M-IoT in Healthcare Domain 95

4. Mishra, S., Sahoo, S., & Mishra, B. K. (2019). Addressing security issues and standards in
Internet of things. In Emerging Trends and Applications in Cognitive Computing (pp. 224–257).
IGI Global.
5. Rath, M., & Mishra, S. (2019). Advanced-level security in network and real-time applications
using machine learning approaches. In Machine Learning and Cognitive Science Applications
in Cyber Security (pp. 84–104). IGI Global.
6. Mishra, M., Mishra, S., Mishra, B. K., & Choudhury, P. (2017). Analysis of power aware
protocols and standards for critical E-health applications. In Internet of Things and Big Data
Technologies for Next Generation Healthcare (pp. 281–305). Springer, Cham.
7. Wollenberg, A., Günther, S., Moderer, M., Wetzel, S., Wagner, M., Towarowski, A., et al.
(2002). Plasmacytoid dendriticcells: A new cutaneous dendritic cell subset with distinct role
in inflflammatory skindiseases. Journal of Investigative Dermatology, 119, 1096–1102.
8. Schmidt, E., & Zillikens, D. (2010). Modern diagnosis of autoimmune blistering skindiseases.
Autoimmunity Reviews, 10, 84–89.
9. Wang, J., Suárez-Fariñas, M., Estrada, Y., Parker, M. L., Greenlees, L., Stephens, G., et al.
(2017). Identification of unique proteomic signatures in allergic and non-allergic skin disease.
Clinical and Experimental Allergy, 47, 1456–1467.
10. Codella, N. C., Nguyen, Q. B., Pankanti, S., Gutman, D., Helba, B., Halpern, A., & Smith, J.
R. (2017). Deep learning ensembles for melanoma recognition in dermoscopy images. IBM
Journal of Research and Development, 61(5), 1–5:15.
11. Schwarz, M., Soliman, D., Omar, M., Buehler, A., Ovsepian, S. V., Aguirre, J., et al. (2017).
Optoacoustic dermoscopy of the human skin: tuning excitation energy for optimal detection
bandwidth with fast and deep imaging in vivo. IEEE Transactions on Medical Imaging, 36,
1287–1296.
12. Li, Y., & Shen, L. (2018). Skin lesion analysis towards melanoma detection using deep learning
network. Sensors, 18, 556.
13. Chao, E., Meenan, C. K., & Ferris, L. K. (2017). Smartphone-based applications for skin
monitoring and melanoma detection. Dermatologic Clinics, 35, 551–557.
14. Ahn, B.G., Noh, Y. H., Jeong, D. U. (2015). Smart chair based on multi heart rate detection
system. In 2015 IEEE Sensors (pp. 1–4).
15. Almotiri, S. H., Khan, M. A., & Alghamdi, M. A. (2016). Mobile health (m-health) system in
the context of iot. In 2016 IEEE 4th International Conference on Future Internet of Things and
Cloud Workshops (FiCloudW) (pp. 39–42).
16. Barger, T. S., Brown, D. E., & Alwan, M. (2005). Healthstatus monitoring through analysis
of behavioral patterns. IEEE Transactions on Systems, Man, and Cybernetics Part A: Systems
and Humans, 5(1):22–27. ISSN 1083–4427.
17. Chiuchisan, I., Costin, H. N., & Geman, O. (2014). Adopting the internet of thingstechnologies
in health care systems. In International Conference and Exposition on Electrical and Power
Engineering (EPE), (pp. 532–535).
18. Dwivedi, A., Bali, R. K., Belsis, M. A., Naguib R. N. G., Every, P., & Nassar, N. S. (2003).
Towards a practical healthcare information security model for healthcare institutions. In IEEE
EMBS Special Topic Conference on Information Technology Applications in Biomedicine
(pp. 114–117).
19. Gupta M. S. D., Patchava, V., & Menezes, V. (2015). Healthcare based on iot using raspberry pi.
In: International Conference on Green Computing and Internet of Things (ICGCIoT) (pp. 796–
799).
20. Gupta, P., Agrawal, D., Chhabra, J., Dhir, P. K. Iot based smart healthcare kit. In International
Conference on Computational Techniques in Information and Communication Technologies
(ICCTICT).
21. Mishra, S., Tripathy, H. K., Mallick, P. K., Bhoi, A. K., & Barsocchi, P. (2020). EAGA-MLP—
An enhanced and adaptive hybrid classification model for diabetes diagnosis. Sensors, 20(14),
4036.
22. Mishra, S., Mallick, P. K., Tripathy, H. K., Bhoi, A. K., & González-Briones, A. (2020). Perfor-
mance evaluation of a proposed machine learning model for chronic disease datasets using an
96 B. Chanda et al.

integrated attribute evaluator and an improved decision tree classifier. Applied Sciences, 10(22),
8137.
23. Dutta A., Misra C., Barik R. K., & Mishra S. (2021). Enhancing mist assisted cloud computing
toward secure and scalable architecture for smart healthcare. In: Hura, G., Singh, A., & Siong
Hoe L. (eds) Advances in Communication and Computational Technology. Lecture Notes
in Electrical Engineering, vol. 668. Springer, Singapore. https://doi.org/10.1007/978-981-15-
5341-7_116.
24. Mishra, S., Tripathy, H. K., & Acharya, B. A. (2021). precise analysis of deep learning for
medical image processing. In Bio-inspired Neurocomputing (pp. 25–41). Springer, Singapore.
25. Mallick, P. K., Mishra, S. & Chae, G. S. (2020). Digital media news categorization using
Bernoulli document model for web content convergence. Personal and Ubiquitous Computing.
https://doi.org/10.1007/s00779-020-01461-9.
26. Mishra, S., Mallick, P. K., Tripathy, H. K., Jena, L., & Chae, G.-S. (2021). Stacked KNN with
hard voting predictive approach to assist hiring process in IT organizations. The International
Journal of Electrical Engineering & Education. https://doi.org/10.1177/0020720921989015.
27. Roy, G., Bhoi, A. K., & Bhaumik, S. (2021). A comparative approach for MI-based EEG
signals classification using energy, power and entropy. IRBM.
28. Nayak, S. R., Sivakumar, S., Bhoi, A. K., Chae, G. S., & Mallick, P. K. (2021). Mixed-
mode database miner classifier: Parallel computation of graphical processing unit mining.
The International Journal of Electrical Engineering & Education, 0020720920988494.
29. Pramanik, M., Pradhan, R., Nandy, P., Bhoi, A. K., & Barsocchi, P. (2021). Machine learning
methods with decision forests for parkinson’s detection. Applied Sciences, 11(2), 581.
30. Panigrahi, R., Pramanik, M., Chakraborty, U. K., & Bhoi, A. K. (2020). Survivability prediction
of patients suffering hepatocellular carcinoma using diverse classifier ensemble of grafted
decision tree. International Journal of Computer Applications in Technology, 64(4), 349–360.
31. Bhatt, T. V., Patel, R. K., Chitara, H. B., Marques, G., & Bhoi, A. K. (2020). Fuzzy logic
system for diabetic eye morbidity prediction. International Journal of Computer Applications
in Technology, 64(4), 339–348.
32. Marques, G., Bhoi, A.K., Albuquerque, V.H.C. de, K.S., H. (Eds.) (2021). IoT in Healthcare
and Ambient Assisted Living. Springer.
33. Bhoi, A. K., Mallick, P. K., Liu, C. M., & Balas, V. E (Eds.) (2021). Bio-inspired
Neurocomputing. Springer.
34. Bhoi, A. K., Sherpa, K. S., & Khandelwal, B. (2018). Arrhythmia and ischemia classification
and clustering using QRS-ST-T (QT) analysis of electrocardiogram. Cluster Computing, 21(1),
1033–1044.
35. Bhoi, A. K., & Sherpa, K. S. (2016). Statistical analysis of QRS-complex to evaluate the
QR versus RS interval alteration during ischemia. Journal of Medical Imaging and Health
Informatics, 6(1), 210–214.
Hybrid AI and IoT Approaches Used
in Health Care for Patients Diagnosis

Shruti Mathur, Akhilesh Kumar Sharma, and Phayung Meesad

Abstract Computerized reasoning alongside the IoT makes insightful machines that
control the keen conduct and helps in dynamic with a little human impedance. The
AI has its application in different regions like in the smart family unit, medical care,
programmed driving, and in passionate collaborations. These are the different half
and halves AI approaches utilizing IoT in various zones; initially, we move around
current remote correspondence advancements utilizing low-power wide area zone
(LPWA) with an unlicensed range. Under the decision of various remote innovations,
we apply the AI calculation to oversee savvy applications and administrations. To test
the proposed AI-empowered LPWA, the AIWAC cooperation framework has been
designed to assemble the cognitive LPWAN. With the little human interference,
artificial intelligence along with the IoT creates intelligent machines that control
smart behavior and helps in decision making. The AI has been applied in various areas
like intelligent household, advanced health care, automatic driving, and emotional
interactions. There are various hybrids AI approaches using IoT in different areas;
firstly, we focus on current wireless communication technologies using (LPWA) low-
power wide area with an unlicensed spectrum. By applying the AI algorithm, we
get a smart control of wireless communication technology, intelligent applications,
and services for the choice of different wireless communication technologies. For
this, the AIWAC emotion interaction system has been designed to build the cognitive
LPWAN which further tests the proposed AI-enabled LPWA hybrid method. Another
approach to evaluate IoT framework in AI is a security which is a vital factor for
any IoT network. IoMT is also known as the Internet of medical things which is
an integration of IoT and the healthcare environment. If any eavesdropping comes
during transmission, it will lead to serious mutation, so for this a proper security is
required. AI along with IOT also proposed an ISA framework along with an IoHT-
based device in the healthcare environment. We actualized the crossbreed MCDM
strategies, for example, (AHP) analytical various leveled process and a strategy for

S. Mathur · A. K. Sharma (B)

Manipal University Jaipur, Jaipur, Rajasthan, India
P. Meesad
King Mongkuts University Thailand NB, Bangkok, Thailand
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 97
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_5
98 S. Mathur et al.

request inclination by closeness to an ideal arrangement (TOPSIS). To accomplish

coronary illness which is the main source of mortality these days, AI plays a major
role. ML alongside the IoT has demonstrated a new advancement in anticipating
coronary illness. To improve the precision in anticipating cardiovascular illnesses
model being delivered with the precision level of 88.7 and for coronary illness with
the crossbreed irregular backwoods with a direct model (HRFLM).Please confirm if
the inserted city name is correct. Amend if necessary.

Keywords MCDM: multi-criteria decision making · HRFLM: hybrid random

forest with a linear model

1 Introduction

The most developing innovation in the coming period depends on man-made

reasoning. The innovation influences the client cooperation with and by the Internet.
Computer-based intelligence has the forthcoming to change how people interact
with the high-tech world as well as with one another through their work and socially.
As the human cerebrum is related, yet the machines are computerized. The signif-
icant contrast between human and man-made brainpower is that the people adjust
to varying climate by utilizing diverse consistent cycles while man-made reasoning
forms machines that mirror human conduct and perform human-like activities.
Half breed clever framework characterizes a product framework that partici-
pates in equal a combination of strategies and methods from artificial knowledge
as subfields, for example, neuro fluffy frameworks. Artificial intelligence has been
applied to different territories as follows:
• Intelligent family.
• Newfangled medical services.
• Programmed driving.
• Emotional interactions.
Before IoT came into the innovation world, patients and expert’s collaborations
were restricted to telegraph and message correspondences. There was no real way to
suggest and screen patients ceaselessly. With the assistance of IoT, we can undoubt-
edly do distant checking in medical care, releasing potential and engaging doctors.
IoT has additionally been applied to medical care that advantage patients, families,
doctors, clinics, and insurance agencies. There are several ways in which IoT has
been applied to various areas [1–11]:
• For Patients: Devices as wearables like wellness groups and other remote asso-
ciated gadgets like blood pressure, glucometer, and so on. These gadgets are
additionally acclimated to remind carbohydrate level, practice check, and so on.
• For Physicians: With the assistance of checking instruments, doctors monitor the
medical issue of patients. They can hold fast to the therapy plan or needing quick
clinical arrangements.
Hybrid AI and IoT Approaches Used in Health Care … 99

• For Hospitals: IoT gadgets outfitted with sensors, constant area of clinical
hardware like wheelchairs, oxygen siphon, and so on.
The use of artificial intelligence along with the IoT works in different applications
for health care.

2 Methodology Used

(i) Cognitive LPWAN:

Toward Intelligent Wireless Services in Hybrid Low-Power Wide Area Network
(1)
The chapter (1) centers around the current remote correspondence innovations
including cell correspondence advances (4G, 5G) and low force wide area (LPWA)
innovation with an unlicensed range (Lora, SigFox) and other LPWA innovation
upheld by 3GPP alongside an approved range (ECGSM). From the viewpoint of the
traffic signal, we utilize the AI-empowered LPWA half breed strategy for the corre-
spondence postponement and energy utilization of the heterogeneous IoT gadgets. To
give the brilliant control of remote correspondence innovation, savvy applications,
we apply the AI calculation. By the idea of IoT implies the association between the
individuals and things or the other way around dependent on the remote or wired
association. The main investigations using the concept of cognitive LPWAN are as
follows:
• By exploring various LPWA technologies includes the LoRa, SigFox, and other
wireless communication technologies.
• Cognitive LPWAN is a combination of multiple LPWA technologies along with
AI in the network layer to have a more convenient and efficient user experience.
• The researcher puts forward the AI-enabled LPWA hybrid method by applying
an AI-based algorithm.
• Through the comparison of AI-enabled single technology using the single
technology mode is based on the AIWAC emotion interaction system.
There are various heterogeneous LPWAN technologies as follows:
1. SigFox: This technology is used to create low-power consumption and low-cost
wireless IOT special networks.
2. Lora: This helps us in facing large-scale IoT including energy management,
disaster management, etc.
3. LTE-M: It consists of three basic standards LTE-M, EC-GSM, and NB-IoT
based on LTE evolution. LTE carriers also include an upstream & downstream
rate of 1Mbps.
4. EC-GSM (Extended Coverage–GSM) proposed five major objectives: Improve-
ment in indoor coverage performance, large-scale device connectivity, and the
CIoT (cellular IoT) communication should be redefined.
100 S. Mathur et al.

5. NB-IoT: It is not only based on GSM but also a clean state solution. From RAN,
two terms are generated as NB-CIOT and NB-LTE (Table 1).
We compared and analyzed all these five technologies on the wireless communi-
cation technology and reached the following conclusion:
• Because of the high data rate and battery life, the LTE possesses the smallest
coverage. The factors involved are spectrum, bandwidth, data rate, and battery
life.
• AsLoRa works with an unauthorized frequency band, so it does not exhibit a fixed
frequency spectrum.
• LoRa, NB-IoT-IoT, and NB-CIoT are lengthy in distance and have low data
covering up to 10 km along with the data transmission rate which are 100 kbps.
• For low bandwidth and long distance, this communication method is suitable.
The LPWA includes the mobile cellular network technology along with the short-
distances communication technologies such as the BLE and Wi-Fi. Likewise, the
chapter proposes the cognitive LPWAN is an assortment of LPWA innovations that
gives more effective and advantageous keen administrations. It likewise proposed
the AI-based LPWA mixture strategy from the viewpoint of the traffic light. To lead
the business shunt and separating remote correspondence advancements, insightful
applications, and administrations for the decision of correspondence, the scientist
at that point took the AIWAC feeling collaboration framework for instance and
fabricated the design of the intellectual LPWA crossover technique.
(ii) For Smart Hospital Management Using Hybrid Approaches [12]:
Likewise, the chapter proposes the cognitive LPWAN is an assortment of LPWA
innovations which give more effective and advantageous keen administrations. It like-
wise proposed the AI-based LPWA mixture strategy from the viewpoint of the traffic
light. To lead the business shunt and separating remote correspondence advance-
ments, insightful applications, and administrations for the decision of correspon-
dence, the scientist at that point took the AIWAC feeling collaboration framework
for instance and fabricated the design of the intellectual LPWA crossover technique.
The main working of the hospital rather a smart hospital is shown (Fig. 1):
The model comprises the following features:
• According to availability of doctor able to schedule appointments.
• Doctor rating, reviews, and able to treat the problem.
• Able to give medicines dose-wise using medical inventory management by
undergoing previous medical records of the patient.
• Along with suggesting treatment plans and prescription assistance.
The above features were deeply applied into as follows [12]:
i. Patient unique medical ID: To identify and update the patient’s previous record
or new patient, a unique ID is given to the patient. Through patient smartphones
and other wearable devices, he/she can be able to check workout schedules,
sleeping patterns, etc.
Table 1 Comparison of LPWA technologies [16]
Technologies LTE Evolution Narrowband Non-3GPP
LTE-M NB-loT EC-GSM LoRa SigFox
NB-LTE NB-Clot
Coverage <1 km <15 km <15 km <15 km <20 km <13 km
Spectrum Licensed Licensed Licensed Licensed Unlicensed Unlicensed
(7–900 MHz) (7–900 MHz) (8–900 MHz) (7–900 MHz) (867–869 MHz or (900 MHz)
902–928 MHz)
Hybrid AI and IoT Approaches Used in Health Care …

Bandwidth 1.4 MHz 200 kHz 200 kHz 2.4 MHz 125 kHz, 250 kHz, 100 kHz
500 kHz
Date rate <1 Mbps 150 kbps <400 kbps 10 kbps <50 kbps <100 kbps
Battery life >10 years >10 years <10 years >10 years <10 years >10 years
101
102 S. Mathur et al.

Fig. 1 Process flow diagram [12]

ii. Doctor ID: Just like patients, doctors are also assigned a unique ID which gives
us the details of doctor performance, availability, and ratings review.
iii. Scheduling and appointment: Through this model, the patient can enter his
time slot, can check the availability, book the bed, etc.
iv. Medical inventory management: To get the treatment as prescribed by the
doctor, the model uses the smart medical dispensers. To get the medicine,
patient enters its ID and gets the medicine by linking to it. The main function
of AI comes into the role when doctors are not available and provide the
recommendation for alternate medications.
v. Smart assistance for tests: For this, the author creates the private hybrid cloud
of the hospital where the X-ray, scan, etc., are diagnosed.
vi. Treatment and prescription assistance: To analyze the entire possible problem
in the body, the above all the methods are used. Based on pros and cons, the
problem is identified and recommended.
vii. Working process: This model helps us to know the doctor’s suitability and
availability of the medicines and also to provide the prescribed actual treatment.
The proposed model is helpful in many ways to improve the entire working of the
hospitals. We would be able to access the record of a patient, its care, and the overall
cost suffered by the hospital and by the patient. It can be achieved by reducing the
medical related errors, waiting time for the patients that are treated per day. Due to
the automation the system can easily work without manual support.
(iii) Automatic Ocular Disease Screening and Monitoring using a Hybrid
Cloud System [13, 19]:
By using advanced image and data analytics along with the IoT, there is a lot of
improvement in health care, especially in ocular disease. Using the concept of hybrid
cloud [19], Chapter (3) introduced a platform that connects medical experts, patients,
ophthalmologists, and intelligence of ocular disease. Through multiple channels, web
portals, and APIs, the retinal fundus images and patient data are uploaded through the
public cloud tier. By utilizing the high-level example characterization calculation, the
information is moved to the private cloud level for evaluation and investigation. The
degree of danger is recognized through this, and appropriately the ophthalmologists
are eluded.
Because of the expansion in the worldwide populace, visual infections influence
a greater amount of the populace. As per the information of WHO, age-related
macular degeneration (AMD) and diabetic retinopathy are the top reasons for vision
Hybrid AI and IoT Approaches Used in Health Care … 103

misfortune and visual deficiency whose treatment is expensive. The individual feels
obscured vision, then it is the manifestation of visual sickness. This issue happens
because of the maturing populace in the well-being frameworks. For better considera-
tion, numerous patients need to travel significant distances in huge urban areas. Some
of them cannot book an arrangement to visit a trained professional. This outcome is
into seeing a ton of non-basic patients and basic patients sitting tight quite a while for
the medicines. Another issue is the manifestation-based visual infection recognition
which is neither touchy nor hearty. Because of the absence of early visual infection
screening programs, all the issues are caused.
(IV) The Proposed Model (4) for Ocular Disease Screening Model:
Bewitch the huge use of clinical imaging advances, the clinical gadget gets less
expensive and less expensive. The prevalence of cell phones and tablets drives the
web entrance to a more significant level. Alongside the web, distributed computing
is likewise developing quickly.
In light of the components, the creator proposed a visual sickness screening model
that frames an environment of patients, centers, optical shops, and so forth. The shops
having fundus cameras for eye screening will go to the optical shop. These fundus
photos taken will be transferred to the cloud worker for the programmed investigation
of visual sickness. Cloud-based ocular disease assessment has the following features:
• Based on the pattern recognition and abstraction, we have a rich suite of algorithms
to improve the performance.
• Ocular disease objective assessment.
• Poor assessment of ocular disease.
• Movement observing.
• Professionally trained not required.
• 24 h high throughputs required.
• Large-scale screening.
• Easy to access the local and international partners by hybrid cloud architecture.

Architecture
For hosting the patented ocular disease assessment, the public cloud tier deals with
every private cloud tier.

Data collection
It might be acted numerous; the most notable course is to eliminate retinal fundus
pictures and patients data from the standard fundus camera which are consolidated
with the PC for data amassing and planning. Not exactly equivalent to standard
fundus cameras, PDAs are used for picture getting and limit. Data amassed from
all before the Internet of Things, patients’ associations with experts, were limited to
visits and telegraph and text exchanges. There was no way subject-matter experts
or centers could screen patients’ prosperity continually and make recommendations
properly. Data secured by all ways will be moved to the public cloud level through
HTTP rest API either by online interfaces or adaptable applications.
104 S. Mathur et al.

Public cloud tier: The public cloud assumes a significant job in the transmission,
stockpiling, and association with the patients and ophthalmologists. The primary
parts of public cloud level are online interfaces, load balancer, understanding record
information base, and so forth. A portion of the segments are associated with focal
information base. The web-based interface was fabricated utilizing cake PHP open-
source structure following the MVC system.
We can utilize these entries for few purposes by using Xcode and Android. Right
off the bat, patients transfer their retinal pictures and individual data, for example,
race, age, weight, sex, and so forth, doing the visual screening. Furthermore, patients
can screen their ailments by following and observing clinical reports. Third, for
making arrangements, a reference motor is accessible. Finally, specialist organiza-
tions can transfer crude information to the framework either by the singular patient
or by the clump.
Ocular disease assessment engine: It comprises a rich setup of calculations that
comprises glaucoma, AMD, diabetic retinopathy, and obsessive nearsightedness.
Through the glaucoma hazard measure, glaucoma is identified. The optic circle is
first fragmented and reproduced utilizing a novel inadequate methodology. The four
elements of this are as follows: Optic plate is divided and remade, at that point process
the danger measure, the optic circle is first recognized as a kind of perspective point,
and lastly a presence of drusen is accomplished by an arrangement by a help vector
machine (SVM). In the proposed model, it as a rule takes 2 min to robotize, and
investigations of pictures are quick and reliable. It improves the admittance to eye
care services. Finally, early identification and early intercession of visual infection
are in a cost-effective way.
(V) The Fifth Wave and i-Sustainability plus Theories as Solutions for Socio-
Edu Consequences of COVID-19:
The new present-day progressions like IoT and AI improved an incredible arrange-
ment in the field of clinical administrations. The world has changed from a socioe-
conomy academic to a data and educational system. The chapter [14] aims to cover
an impact on the overall crisis by the illness of COVID/post-COVID-19 Era on
the enlightening industry and besides the socio-Edu results. For managing the keen
SMEs, smart society, splendid clinical administrations and government help, etc.,
the IoT, ML, AI, and other advancement which are used for digitalization culture are
called as work government that further helps as 5.0 in metropolitan and 6.0 as game
plans.
The fifth wave or tomorrow age speculation is connected to the proceeding of
the possible destiny of 14.0(15.0) as a picture for the west, same as society 5.0(6.0)
as a picture for non-west and edge of tomorrow was first introduced by Prof. Dr.
Hamid Doost Mohammadian in the year 2010. In light of this speculation, tutoring
has a huge task to show up at social commitment to improve the idea of sensibility
and life subject to the 7PS model for progression. The hypothesis of the fifth wave
is like an instrument to accomplish one among the seven columns for instruction
maintainability.
Hybrid AI and IoT Approaches Used in Health Care … 105

The top three reasons that give a torch bearing impacts in the COVID-19 era are
as follows:
• Up close and personal visit of teachers, understudies, and scientists impractical.
• In research and training laboratory many experimental activities.
• Reducing the international mobilization for the short term.
• The positive impacts are as follows:
• To spread the importance of digital transformation and IT among the teachers,
students, and staff.
• Significant distance communitarian work.
• For building up the chances and inspirations for creating jump frog techniques.
To increase the eagerness of academic society applied in the post-COVID-19 era
by:
• Sharing the best practice among the entities and nations.
• To trade information and experience.
• To build up instructive joint projects and workshops.
• Establish an expert organization of the specialists filled in the filled structure by
two nations.
• Applied and built-up exploration projects.
• For the advancement in a post-COVID-19 climate.
• Environment, social, economic, educational, cultural, technical, and political
aspects are the seven pillars of the sustainability (7PS) model.
Socio-edu consequences caused by COVID/post–COVID-19 era is related to the
following:
• Absence of development in scholastic ecosystem.
• Acts as a STEM for professional and specialized preparing
• Entertainers (teachers, understudies, and chairmen)
• Monetary and economics constraint
• Nature of distance learning/E-learning
• Viable courses
• Operational courses
• Operational coordination
• Hole among hypothesis, work market, and industry
• New position souk in COVID/post-coronavirus era
• Absence of inspiration to utilize IoT training
• Social restrictions.

(VI) Latest IoT-based Solutions in Health Care:

To spread among the edge/mist well-being arrangements by utilizing appropriate

registering stages to disseminate well-being pointers for information handling, it
is stockpiling among the numerous hubs, situated at various degrees of hubs. To
circulate well-being sensors information handling and to store among numerous hubs,
we center around the edge/fog health arrangements. In nowadays, medical services
106 S. Mathur et al.

are gripping the data innovations for spreading the brilliant framework focused on
well-being diagnostics and therapy in this manner permitting the significant decrease
of doctor expenses and patient consideration quality. The IOMT [15] has at last
changed how medical services are moving toward around the world, so the number
of medical care gadgets utilizing IoT and different innovations is expected to arrive
at 162 million by the end of 2020. A serious level of vigor and availability is needed
for end clients for example quiet with the basic and time delicate requirements. The
principle objective is to misuse the total strength of end hubs and low degree of haze
hubs to deal with not just practical undertakings likewise information preparing,
investigation, connection, and impedance.
To spread among the edge/mist well-being arrangements by utilizing appropriate
registering stages to disseminate well-being pointers for information handling, it
is stockpiling among the numerous hubs, situated at various degrees of hubs. To
circulate well-being sensors information handling and to store among numerous hubs,
we center around the edge/fog health arrangements. These are located at different
levels to users:
• Edge registering is generally performed on the gadgets to which the sensors are
appended or a passage gadget which is near the sensors like cell phones, savvy
watches, or a versatile “ad hoc” inserted frameworks like a microcontroller.
• Fog computing works similar to the local area network level, which works with
bigger and more powerful devices such as PCs, local sensors and gateways be
more distant from the sensors and actuators.
• In this, the author proposed IoMT solutions for health monitoring. IoT-based
healthcare systems involve functional tasks based on the edge/fog computing
for smart health. A general solution employs a multilevel architecture, in a very
simplified way:
• Edge level: It collects data on (WSBN) wireless body sensor networks and
performs preprocessing on portable devices [17].
• Fog level: In this, data collection and processing both are done on PCs and
servers/gateways.
• Cloud level: In this, cloud administrations are called for superior figuring
undertakings and distant information stockpiling.

3 Conclusion

As per the review, the authors proposed a different model and platform and also
compared the performance and efficiency of different IOT and LPWA technologies.
They also discussed about the post-COVID era consequences.
We concluded that
(a) For better interaction in the communication technologies, we discussed
different LPWA technologies on the basis of different parameters and also
cognitive LPWAN providing the efficient and convenient smart services. We
Hybrid AI and IoT Approaches Used in Health Care … 107

can also elaborate the same further in the security, energy consumption and
also the real-time management, resource optimization, and deployment of
infrastructure.
(b) The model discussed changed the scenario of health care by giving a proper aid
to the hospital by applying hybrid cloud, IoT, AI, and ML. This model results
it into a reduced waiting time, and more patients are treated per day.
(c) By using the medical image-based classification technique, the researcher
presented an online cloud-based service platform for the automatic screening
of ocular disease. It helps in analyzing the early involvement and disease
management.
(d) There will be a huge disaster for further years because of contagion of the
COVID-19 challenges, socio-edu consequences, recession disaster, climate
change calamity, technology, etc. Also it has its impact on social factors also
like digitalization, sustainability, comprehensive strategic, and various other
impacts.
This chapter gives the insights related to the latest advancements in the field of
AI and healthcare domain. The recent impact of the COVID-19 affected the various
intercontinents, and the emergence of the healthcare-related remedies went so popular
and people interested to use these IOT devices within their reach so as to protect
themselves and their families.

References

1. Roy, G., Bhoi, A. K., & Bhaumik, S. (2021). A comparative approach for mi-based EEG signals
classification using energy, power and entropy. IRBM.
2. Nayak, S. R., Sivakumar, S., Bhoi, A. K., Chae, G. S., & Mallick, P. K. (2021). Mixed-
mode database miner classifier: Parallel computation of graphical processing unit mining.
The International Journal of Electrical Engineering & Education, 0020720920988494.
3. Pramanik, M., Pradhan, R., Nandy, P., Bhoi, A. K., & Barsocchi, P. (2021). Machine Learning
Methods with Decision Forests for Parkinson’s Detection. Applied Sciences, 11(2), 581.
4. Panigrahi, R., Pramanik, M., Chakraborty, U. K., & Bhoi, A. K. (2020). Survivability prediction
of patients suffering hepatocellular carcinoma using diverse classifier ensemble of grafted
decision tree. International Journal of Computer Applications in Technology, 64(4), 349–360.
5. Bhatt, T. V., Patel, R. K., Chitara, H. B., Marques, G., & Bhoi, A. K. (2020). Fuzzy logic
system for diabetic eye morbidity prediction. International Journal of Computer Applications
in Technology, 64(4), 339–348.
6. Yadav, A., Kumar Singh, V., Kumar Bhoi, A., Marques, G., Garcia-Zapirain, B., & de la
Torre Díez, I. (2020). Wireless Body Area Networks: UWB Wearable Textile Antenna for
Telemedicine and Mobile Health Systems. Micromachines, 11(6), 558.
7. Marques, G., Bhoi, A.K., Albuquerque, V.H.C. de, K.S., H. (Eds.) (2021). IoT in Healthcare
and Ambient Assisted Living, Springer.
8. Bhoi, A. K., Mallick, P. K., Liu, C. M., & Balas, V. E (Eds.) (2021). Bio-inspired
Neurocomputing, Springer.
9. Bhoi, A. K., Sherpa, K. S., & Khandelwal, B. (2018). Arrhythmia and ischemia classification
and clustering using QRS-ST-T (QT) analysis of electrocardiogram. Cluster Computing, 21(1),
1033–1044.
108 S. Mathur et al.

10. Bhoi, A. K., & Sherpa, K. S. (2016). Statistical analysis of QRS-complex to evaluate the
QR versus RS interval alteration during ischemia. Journal of Medical Imaging and Health
Informatics, 6(1), 210–214.
11. Marques, G., Miranda, N., Kumar Bhoi, A., Garcia-Zapirain, B., Hamrioui, S., & de la Torre
Díez, I. (2020). Internet of Things and enhanced living environments: measuring and mapping
air quality using cyber-physical systems and mobile computing technologies. Sensors, 20(3),
720.
12. Kumar J. N. A., & Suresh, S. (2019). A Proposal of smart hospital management using hybrid
Cloud, IoT, ML, and AI.
13. Yin, F., Lee, B. H., Yow, A. P., Quan, Y., Wong, D. W. K. (2016). Automatic ocular disease
screening and monitoring using a hybrid cloud system.
14. The 5th Wave and i-Sustainability plus Theories as Solutions for Socio-Edu Consequences of
Covid-19, 2020.
15. Greco, L., Percannella, G., Ritrovato, P., Tortorella, F., & Vento, M. (2020). Trends in IoT
based solutions for health care: Moving AI to the edge.
16. Chen, M, Miao, Y., Jian, X., Wang, X., Humar, I. (2018). Cognitive–LPWAN: Towards
intelligent wireless services in hybrid low power wide area networks.
17. Usman, R., Kulkarni, P., & Sooriyabandra, M. (2017). Low power wide area networks: An
overview. IEEE Communication Surveys Tuts, 19(2), 855–873. 2nd Quart 2017.
18. Archenaa, J. & Anita, E. A. M. (2015). A survey of big data analytics in healthcare and
government. Procedia-Procedia Computer Science, 50, 408–413. https://doi.org/10.1136/qshc.
2006.022046.
19. Journal of Ambient Intelligence and Humanized Computing. https://doi.org/10.1007/s12652-
017-0659-1: The impact of the hybrid platform of the internet of things and cloud computing
on healthcare systems: opportunities, challenges, and open problems.
RADIoT: The Unifying Framework
for IoT, Radiomics and Deep Learning
Modeling

Sakinat Oluwabukonla Folorunso , Joseph Bamidele Awotunde ,

Femi Emmanuel Ayo , and Khadijah-Khuburah Adebisi Abdullah

Abstract The IoT revolution reshapes contemporary healthcare systems by incor-

porate economic, social, and technological prospects. It is progressing from conven-
tional healthcare systems to more personalized healthcare systems, where patients
can be monitored, diagnosed and treated effortlessly. Radiomics is a sub-field of
machine learning (ML) that mines quantitative features from radiological images
relying on an image-based approach with ML models, which procure information
surpassing orthodox medical imaging analysis as diagnosis, prognosis, prediction
and response to therapy. The upsurge in the number of radiological images increases
the workload of radiologist which in turns decreases their performance, thus they
can only detect and evaluate a small portion of information present in images within
a short-time. Hence, there is need for a better method for the increase in radiological
image selection, detection and evaluation processes thereby reducing the workload
of experts. Therefore, this chapter discusses the different types and sources of radio-
logical data, feature extraction and selection method for image analysis. The chapter
also presents different ML models ideal for the radiomics and parameter tuning.
The challenges, applicability and limitations of Radiomics are also described in this
chapter. The radiomic process involves radiological image gathering, segmentation,
feature extraction and selection, model building and evaluation. Each of the stage
of the process workflow is carefully evaluated for development of a reliable, effec-
tive and robust model to be shifted into medical practice for disease diagnosis and
prognosis response to treatment.

S. O. Folorunso (B) · K.-K. A. Abdullah

Department of Mathematical Science, Olabisi Onabanjo University, Ago-Iwoye, Nigeria
e-mail: [email protected]
K.-K. A. Abdullah
e-mail: [email protected]
J. B. Awotunde
Department of Computer Science, University of Ilorin, Ilorin, Nigeria
e-mail: [email protected]
F. E. Ayo
Department of Computer Science, McPherson University, Ogun State, Nigeria
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 109
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_6
110 S. O. Folorunso et al.

Keywords Radiomics · Internet of Things · Deep learning · Image segmentation ·

Texture descriptors · Quantitative imaging

1 Introduction

Medical imaging is a clinical invention that have been largely transformed how
clinicians measure, manage, diagnose and treat diseases [1]. Radiomics is the proce-
dure of transforming a digital medical image into an exploitable high-dimensional
data using data-characterization algorithms [2]. The radiomics concept is inspired
by the fact that medical images contain information that reflects the underlying
pathophysiology and that these relationships can be revealed via quantitative image
analyses [3]. The rapid extraction of these countless numerical characteristics from
radiographic phenotype [1], tomographic images like positron emission tomography
(PET), computed tomography (CT), X-Ray, magnetic resonance imaging (MRI)
and cone beam CT (CBCT) is made possible with high-throughput computing. The
imaging may reveal metabolic data (PET, MRI), or structural data (CT, MRI), and
have variable image resolution. These characteristic features that are mined from
medical images are made amenable for analysis and decision support by machine
learning models. Radiomics being a medical decision support tool can be combined
with other patients’ clinical information for better decision-making and evidence-
based clinical decision support. Recently, smart systems are expediating the health
diagnosis and treatment. These systems render intelligent services for health tracking
and medical automation in various situations and places like home, hospitals and
offices, thus improving timely intervention and reducing the cost of Doctor’s visita-
tion [4]. The design of a robust embedded hardware alongside with smart medical
sensors and devices for universal healthcare has drastically made the Internet of
Medical Things (IoMT) widely popular, so much that the count of healthcare devices
using IoT and wearable devices is projected to reach 162 million by the year ending
in 2020 [4, 5]. Medical imaging data obtained by ingestible, wearable and embedded
sensors can be accurately and efficiently analyzed by machine and deep learning
models for prognosis and diagnosis [5].
Medical imaging informatics is experiencing a dramatic change as the volume,
variety, and the velocity of clinical text and imaging data have increased greatly.
Tools for data capturing and analysis are increasing as well [6]. The cost of both
commercial and custom-made smart devices [7] like Raspberry Pi (RPi) [8] have
reduced drastically due to the reduced cost of microcontrollers and single-board
computers (SBCs) [7].
Hence, this chapter proposed a new hybrid of AI and IOT unifying framework
RADIOT. It comprises of IOT, radiomics and deep learning-based model for classi-
fication and prediction of diseases based on medical images which serves as input
into the IOT real-time application. This chapter also discusses the different types
and sources of radiological data, feature extraction and selection method for image
analysis. This chapter also presents different ML models ideal for radiomics and
RADIoT: The Unifying Framework for IoT, Radiomics … 111

parameter tuning. The challenges, applicability and limitations of radiomics, deep

Learning and IOT are also discussed.
The chapter is structured as follows. Section 2 describes the IoT and IoT
for radiomics. Section 3 describes radiomics and its process work flow, while
Sect. 4 describes machine learning models for radiomics. Section 5 presents chal-
lenges, issues and opportunities, while Sect. 6 presents the proposed framework and
implementation of RADIOT. Finally, Sect. 7 presents the conclusion of the chapter.

2 The Internet of Things (IoT) in smart healthcare system

IoT involves of a group of linked devices that can move data from one device
to another so that performance can be optimized automatically. The IoT network
comprises of four key elements namely sensors, processing networks, data analysis
and system monitoring [9]. The aim of IoT is to build an intelligent and smart system
to improve the way of living. The breakthrough for IoT started when RFID tags
were used often making widely available low-cost-sensors, advanced web technology
and new communication protocols [10]. The IoT device constitutes communication
protocols necessary for connectivity [11, 12–15]. The three main components are
device-to-device (D2D) which communicates between neighboring mobile phones.
Another component is device-to-server (D2S) which transmits all data from far and
near devices to a central server. A popular application of this type of communication
is cloud processing. Lastly, server-to-server (S2S) which communicates and transmit
data from server to server. Cellular networking is a type of this protocol. The process
flow of IoT is to firstly obtain information from the environment using sensors and
IoT devices. Then, knowledge is mined from the data obtained before moving to
other devices, objects, or servers via the Internet.

2.1 Internet of Things (IoT) for Radiomics

Medical imaging denotes the acquisition of internal images of a person for diagnosis
and prognosis of diseases. The IoT in the medical imaging aids the retrieval and
sharing of prognosis and diagnosis information real time. This removes the problem
of quick access to data, repeating examinations, consultation and wrong diagnoses
[16]. The machine-machine correspondence provided by the IoT aids the imaging
equipment to increase the image quality and lessen operation expenses. The various
sources of the medical imaging that can be shared/retrieved from IoT equipment for
radiomics analysis includes MRI, X-Ray, CTs, US and PET as shown in Fig. 1.
The equipment shown in Fig. 1 can be used to diagnose or detect diseases or abnor-
malities in human internal body. This equipment will effectively diagnose diseases
very fast, thereby increasing the number of patients and reports generated. Some of
the issues arising from these vast amounts of reports generated can be misplaced
112 S. O. Folorunso et al.

(b) X-Ray [14] (c) CT Scanner

[15]

(a) MRI Scanner

[13]

(d) PET Scanner

[16]

(e) Ultra Sound

Scanner [17]
Fig. 1 Different equipment for different sources of medical imaging

or an error exchange of patients’ record could lead to wrong treatment. One solu-
tion would be to embed IoT into imaging equipment and connect to the internet
to transfer the diagnosed report for radiomic and further analysis. Figure 2 shows
a block diagram of an IoT based healthcare system that can be embedded into a
RADIoT: The Unifying Framework for IoT, Radiomics … 113

Fig. 2 Block diagram of an IoT-based smart radiomic system

radiomics system. Sensors can be embedded into medical imaging equipment (US,
CT, MRI and X-Ray) that can transmit the images to the IoT on the Internet.
Figure 2 shows how multiple sensors embedded into medical imaging equipment
are connected to the IoT client. This endpoint is linked to the back-end server via a
wireless network using MQTT, CoAP protocol. Radiomic and ML/DL analysis can
be performed from the IoT server at the back-end.

3 The Radiomics

Radiomics is mining of quantitative features from medical images like PET and
CT, modeling and analysis in relation to prediction targets [17]. Radiomics is a
multidisciplinary-based technology. The five stages of the radiomic process flow are
presented in Fig. 3.

Fig. 3 Process flow of radiomics

114 S. O. Folorunso et al.

3.1 Dataset Acquisition

Acquisition of dataset is the initial and a vital stage in radiomics. The dataset consists
of images in formats of the picture archiving and communication system (PACS)
and digital imaging and communication in medicine (DICOM). These formats are
suitable for studies in radiology and are popularly used by medical institutes [18].
The quality of images acquired is also significant in determining the precision and
replicability of the ultimate radiomics model. But the lack of standards for image
acquisition results heterogeneity different studies. The different imaging modali-
ties like ultra Sound (US), PET, CT and MRI scanners permits the obtaining and
reformation of images in large variations [19] for analysis.

3.2 Volume of Interests (VOIs) Segmentation

VOIs is the area of the image captured and segmented. This stage determines the
volume to be analyzed in a certain medical image. VOIs segmentation is the core
step of radiomics study because it determines which volume is analyzed within a
medical image. It estimates which voxels inside an image are examined, and thus,
any fluctuation in segmentation can inject bias in the estimation of the resultant
radiomic features [20]. One method to use to control the degree of this bias it to apply
a multiple-segmentation. VOI segmentation could be done physically by radiologist
or automatically by models. Other methods to automatically create segmentation
are level-set methods [21], graph-cut methods [22, 23] and active contour algorithms
[24]. So, there is a high demand for methods for replicability, less expert involvement,
maximum automation and high time efficiency.

3.3 Feature Mining

Feature mining is the third stage of radiomics process flow and the most essential of
all the stages. It largely mine high-throughput data of measurable imaging character-
istics to name VOIs and choose valuable information which will help to differentiate
between image classes.
The different features mined from radiomics features are categorized into four:
1. Shape features: Here, the geometric and shape features of the marked lesions are
mined. This includes the sphericity, the volume of VOI, the maximum surface
area, tumor density and maximum diameter along with different orthogonal
directions.
2. First-order statistical characteristics: This scheme provides the descriptive statis-
tics of each image in the dataset. The measure of central tendencies includes
RADIoT: The Unifying Framework for IoT, Radiomics … 115

the median, skewness, mean, kurtosis, minimum and maximum voxel intensity,
randomness and uniformity.
3. Textual Features: Texture features mines diverse information among the lesions.
This feature is estimated from a statistical correlation between neighboring
voxels using Gray-Level Co-Occurrence Matrix (GLCM) [25].
4. High-order statistical characteristics: The goal of the mathematical transforma-
tion of images at this stage is to identify repetitious or non-repetitious patterns
and subdue noise. Measurable information obtained from mathematical transla-
tion of images comprises Minkowski function, fragment analysis, wavelet and
Laplace transform of gaussian filtered image [23].
Some popular astute feature mining schemes for dataset are histogram of oriented
gradient (HOG) [26], local binary pattern (LBP) [27], elongated quinary patterns
(EQPs) [28], first-order statistical features (FOSFs) [29], local directional number
(LDN) [30], locally encoded transform feature histogram (LETRIST) [31], binarized-
statistical image features (BSIFs) [32], local phase quantization (LPQ) [33] and
oriented-basic image features (oBIFs) [34]. Deep Learning models also have the
capability to extract features for modeling called transfer learning. Some of the
deep feature schemes include VGG16 and 19 [35], ResNet [36], InceptionV3 [37],
MobileNetV2 [38], DenseNet [39], EfficientNet [40] and NASNetLarge [41].

3.4 Feature Selection

This is a necessary action after feature mining during a radiomics approach. This
action will aid to reduce the feature space by removing the redundant and highly
correlated feature [42]. Hence, the curse of dimensionality is lowered and the danger
of overfitting is removed. The resultant feature now contains the most useful and
relevant set of information which is amenable to the model for optimum result. Some
common feature selection methods are principal component analysis (PCA) [43],
analysis of variance (ANOVA) [44], information gain (IG) [45], whale optimization
algorithm (WOA) [46], gray wolf optimization (GWO) [47], multiscale principal
component analysis (MSPCA) and independent component analysis (ICA) [48].

3.5 Model Development

This is an important stage as choosing a model with optimum parameters can aid in
obtaining an accurate and consistent result which is key to medical decision making.
So, developing a good ML model for radiomics analysis involves choosing useful
radiomic features, a suitable ML models and validation of the new model. Features
mined from radiographic images are numerous, so it is good practice to remove
116 S. O. Folorunso et al.

redundant and correlated features before modeling. Figure 2 shows the taxonomy of
machine learning models for radiomics.

4 Machine Learning Models for Radiomics

Machine learning (ML) is divided into two major types namely supervised and unsu-
pervised learning. Supervised learning describes a learning problem that maps the
input vectors to their corresponding target vectors. The two major types of task for
supervised learning is classification and regression. A major drawback of this type of
learning is the scarcity of labeled data as manual data labeling is costly and laborious.
An unsupervised learning models a data with only input vectors. It mines relevant
information and relationship from a data without output vectors. Taxonomy for ML
models popularly used for Radiomics are shown by Fig. 4. The comparison between
ML and DL models for radiomics is presented in Table 1.

Fig. 4 Taxonomy of machine learning models for radiomics

Table 1 Comparing ML models

Criteria Traditional ML models DL models
Running time Lesser Longer
Count of parameters Less parameters The learnable and
hyper-parameters are more
Feature representation Only model feature vector and Does not need to learn feature
reliant on feature engineering for representations from Radiomic
optimum outcome data and not dependent on feature
engineering
Learning capacity Lesser ability to learn faster Learns better and faster, but faces
a higher risk of overfitting
Interpretability They have stronger Deep learning models are black
interpretability [49] boxes [50]
RADIoT: The Unifying Framework for IoT, Radiomics … 117

4.1 Traditional ML Models

The traditional supervised ML for Radiomics largely comprise the base classifiers
and meta learners. Some common base classifiers are decision tree (DT), artificial
neural network (ANN), Naïve Bayes (NB), support vector machine (SVM), logistic
regression (LR) and k-nearest neighbor (k-NN). The meta learners, which comprises
majorly of homogeneous and heterogeneous ensembles are Bagging, AdaBoost,
random forest and stacking. The unsupervised learning category is riven into three
major parts, namely clustering, density estimation (DE), visualization and projection.
These models have been researched and applied to radiomics over the years.
The focus has always been in prognosis and diagnosis of diseases and data manage-
ment. Traditional ML models have extensively used with radiomic feature extraction.
Reference [20] applied the linear SVM prognosis of head and neck tumor using multi-
modal CT and PET images on features extracted by handicraft and deep learning.
The model was assessed based on MCC and accuracy with values 0.848 and 0.965,
respectively, for combined features of handicraft and deep learning.
Also, reference [51] used LR to detect renal cancer in the patient with CT images.
Based on the handicraft feature extraction, LR achieved a good ROC_AUC value
of 0.878. Likewise, reference [52] compared four (4) different ML models, namely
random forest (RF), AdaBoost, CART decision tree (DT) and XGBoost (XGB) in
diagnosing lungs cancer in patients using CT images. The authors applied AlexNet,
ResNet50, LBP and GLCM textual descriptors to extract features from the medical
image for classification. RF with LBP gave a superior performance based on accuracy,
AUC, precision and Recall with values 77.89, 76.92, 74.19 and 90.20, respectively.

4.2 Deep Learning (DL) Models

DL networks are multi-layered feed-forward neural networks inspired by the human

brain to receive 3D images as input trains them end-to-end, while learning extremely
discriminative image features; this removes the bottleneck of handicraft radiomic
mining of features from images approach. The machine model features created by
the DL network from medical image data can also be analyzed for useful insights
and the hidden pattern [53, 54].
DL models comprise of varied deep networks also categorized as supervised
and unsupervised learning. Supervised learning category comprises of convolu-
tional neural network (CNN), deep belief network (DBN), recurrent neural network
(RNN) and deep neural networks (DNNs). Unsupervised learning category is domi-
nated by Autoencoders, generative adversarial networks (GANs) and deep boltzmann
machines (DBMs). DL models have been applied extensively to radiomics over the
either as a feature extraction technique for CT, PET, MRI and ultrasound images or
to learn the feature representation directly for classification. As DL models thrive
well on massive data, they model such data better than traditional ML.
118 S. O. Folorunso et al.

DL had also been applied to radiomics for disease detection and classification.
Reference [55] assessed the performance of CNNs to breast cancer risk assess-
ment where transfer learning was applied to digital screening mammograms to mine
features from the image. This performance was compared with the parenchymal
texture analysis (PTA). The combined feature of DL and PTA gave a superior perfor-
mance to the sole handicraft and the DL mining method in the classification of
cancer cases. Furthermore, [56] proposed a 3D CNN to mine features from chest CT
to categorize the possibility of the malignancy of pulmonary nodules in patients.
The mined features were evaluated on CNN, ensemble and SVM models. The
ensemble model gave a superior performance based on AUC of 0.78 to other
models. Also, [51] applied CNN architecture with data augmentation to categorize
micro-aneurysms in diabetic retinopathy using fundus retinal images. Their proposed
method revealed a high recall and precision based on false positive rate metric. More
so, authors in [52] proposed an evolutionary deep radiomic sequencer discovery
method for accurate classification of more so lung cancer. Their method performed
best based on accuracy and sensitivity values of 82.39% and 93.42%, respectively,
when compared to other radiomics application in the literature.
Application of DL for image segmentation in radiomics has been extensively
researched. Reference [53, 54] proposed stacked denoising autoencoders (SDAEs)
to gliomas from magnetic resonance images (MRI) and brain lesion segmentation by
deploying transfer learning approach. This approach gave a good performance when
compared to the CNN method. Also, [57, 58] proposed automatic segmentation of
prostate cancer in MRI images.

4.3 Performance Indicators for ML Models

Measuring the performances of a ML model, a binary or multi-class problem is

selected, for example, the binary problem of non-COVID-19 and COVID-19 infected
patients. A good classifier categorizes all COVID-19 infested patients as COVID-19
infested patients and all non-COVID-19 infested patients as NORMAL. A misclas-
sification can either be that a NORMAL patient is classified as COVID-19 infested
patients, or a PTB patient is classified as the COVID-19 infected patient. These two
types of cases are tagged classification errors and are identified as Type I and II
errors, respectively [59].
The classification report contains four cases with two classification errors. This is
presented in Table 2 named confusion matrix (CM). The CM table sum up correctly

Table 2 Confusion matrix

Actual class Predicted class
Class A Class B
Class A TP FP
Class B FN TN
RADIoT: The Unifying Framework for IoT, Radiomics … 119

and incorrectly samples belonging to each class [60]. Four groupings in CM are
explained as follows:
True positive (TP): y = i, f (x) = i (COVID-19 patient correctly classified as
COVID-19).
True negative (TN): y = i, f (x) = i (NORMAL patient correctly classified as
NORMAL).
False positive (FP): y = i, f (x) = i (NORMAL patient is classified as COVID-
19).
False negative (FN): y = i, f (x) = i (COVID-19 patient is classified as
NORMAL).
This arrangement could also be extended to multi-class problems. The derived
indicators from CM are shown in Eqs. (1)–(7)

TP + TN
Accuracy = (1)
TP + TN + FP + FN
TP
Precision = (2)
TP + FP
TP
Recall(TPR) = (3)
TP + FN
FP
False Positive Rate (FPR) = (4)
FP + TN
FN
False Negative Rate (FNR) = = 1 − Recall (5)
TP + FN
Precision × Recall
F1 - Score = 2 × (6)
Precision + Recall
1 + (TPR − FPR)
ROC = (7)
2

5 Challenges, Open Issues and Opportunities

This section will concentrate on the limitations of processing methods unique to

radiomics which includes challenges, issues and opportunities as highlighted in the
subsequent subsections.
120 S. O. Folorunso et al.

5.1 Challenges of Handicraft Radiomics

Mining of handicraft features in radiomics is challenging as there are fewer unique

visual variations [61]. A good way to address this problem is to firstly mine very many
diverse high-level and low-level features from the medical image. Then, a simple but
astute feature selection method be applied to reduce the curse of dimensionality
and uncorrelated features. Segmentation is an important part of handicraft radiomics
and is usually provided by radiology experts. But this manual segmentation is time
wasting and costly.

5.2 Challenges of Deep Learning for Radiomics Analysis

Application of DL to mine features from medical images has many benefits to

radiomics including its ability to generalize and operate without experts’ supervision.
Yet, it is not free of shortcomings like only learning well with massive datasets, and
its lack of robustness to some kinds of transformation. One solution to the problem
of small data size is data augmentation. This increases the number of images based
on minimal rotations. Another challenge with DL for radiomics is mining large
images. This increases the computational cost of the model. One popular approach
to resolving this issue is to crop the images to a manageable size. Also, patient’s
clinical data is subject to ethical and privacy regulations which restrict access to train
data. It almost impossible getting the right and amount of data needed for modeling.
Lastly, interpretability and explainability of the deep learning model in radiomics
are essential as human lives are at stake. The absence of these two factors will limit
the capability of deep learning radiomics in medical practice.

5.3 Challenges of IoT in Radiomics

Some of the challenges faced by IoT in radiomic is the vast amount of image data
produced by the sensors. Mining of handicraft radiomic features from these real-time
data are laborious. Also, the number features generated could be extremely large.
One solution could be to use the DL model for classification and prediction. Another
bottleneck is data optimization [62] as analysis at the back-end are computationally
intensive and a fast real-time response is needed. One solution to this could be to
distribute computational resources. Another issue for consideration is the security of
IOT devices.
RADIoT: The Unifying Framework for IoT, Radiomics … 121

5.4 Open Issues and Opportunities

There are many open problems associated with radiomics inspite of recent techno-
logical advancements and extensive medical imaging applications. These includes
a. Radiomics techniques requires a substantial number of training images, but are
hard to collect due to strict data privacy policies.
b. The main challenge for radiomic analysis is combining different data sources
(like image and text clinical) with missing text data for patients. One solution to
this is for radiomics to model sparse data appropriately [63]. Also, the challenge
of when to adopt feature the fusion method for radiomics needs to be addressed.
c. Lack of labeled data for radiomics analysis as this can only be assigned by
radiology clinical experts. This can be remedied by building semi-supervised
models with specifics of the radiomics domain.
d. Imbalance dataset states a problem where there are more examples of one class
than the other. This problem is even bigger when allied with class disjucts and
overlap [60]. This situation may give rise to suboptimum classification. This
class imbalance problem is also prominent with the radiomic analysis as the
count of positive class (presence of disease) is naturally lesser than the negative
class (absence of disease). Many schemes both at data and algorithm level have
been developed to alleviate this problem in clinical dataset. Some of the methods
listed in [64] were SMOTE [65], ENN [66], TomekLinks [67], ADASYN [68].
Other issues related to radiomics are image noise, the image environment and
scanner abilities.

6 Implementation

6.1 The RADIoT Unifying Radiomics Framework

RADIoT was developed as the unifying framework for classification and predictive
model that can accept medical images as dataset from IoT equipment and combine
DL and predefined features. The framework is also structured to administer medical
imaging datasets from different sources (Multimodal) as in the case of study of
COVID-19 epidemic dataset. The stages in the RADIOT pipeline after the pre-
processing stage is shown in Fig. 5 and explained in the next section. Figure 6
shows the sample chest CT images of the COVID-19 infested and Normal patient
used for this chapter.
(i) Radiomic features (RadF): A total of 1,631 handcrafted features (RadF) were
mined for each COVID-19 infested, and normal patients. These set of features
were mined from medical images based on the Radiomics framework [1].
The RadF sets mark out three key image properties: Shape with 13 number of
122 S. O. Folorunso et al.

Fig. 5 The RADIoT framework for Chest CT scan for COVID-19

COVID-19 Chest CT scan image Normal Chest CT scan image

Fig. 6 Sample chest CT scan images of COVID-19 infested patients

features, intensity with 18 number of features and texture with 1600 number
of features. Based on suggestions from [69], 40 kinds of textual features
with each one computed on 40 sets of parameters that describe pixel layout,
quantization technique and count of gray levels;
(ii) DeepF: A sum of 512 features mined from deep learning trained on CT images
(256 from CT images). A transfer learning approach was applied by initial
training of entire network on the dataset. The DL model hyper-parameters
applied for this chapter to manage the training process with Adam [70] opti-
mizer are: batch_size is 32, number of epochs is 40 and the learning rate is
10−3 . Data augmentation procedures were applied to increase the number of
RADIoT: The Unifying Framework for IoT, Radiomics … 123

Table 3 The outline of the

Feature name Count of Initial After
three feature groups
instances uncorrelation
RadF 1134 1631 490
DeepF 1134 256 256
RadF + DeepF 1134 751 732

images, enhance performance and lessen overfitting: i.e., nominal rotations,

conversions and Gaussian noise. The revamped images were resized to cubes
of 64 × 64 × 64 to fit into the GPU of Google Collab memory size.
(iii) RadF + DeepF: Features mined using radiomics and deep learning were
concatenated into a unifying dataset. An accurate model is projected from
this new dataset as it should take on diverse and balancing information from
the input medical images.

6.2 Feature Selection

The feature selection unit in RADIoT leverages on the union of three methods from
scikit-learn [70]. This is achieved by the computation of the Pearson’s correlation
matrix by elimination of highly correlated features greater than 0.9 (ρ > 0.9). The
purpose of feature selection is to transform feature vector from a high D-dimension
to a low H-dimension by removing insignificant, noise and unnecessary features.
The ‘curse of dimensionality’ problem is removed by this process resulting to an
improved classification accuracy for models.
The resultant dataset with its unique feature were learned on machine learning
model for further analysis. It is observed that for DeepF, all features created were
highly uncorrelated as none of its features were removed. Table 3 shows the 3 feature
groups and the outcome of the feature selection method.

6.3 Classification Results and Discussion

Linear SVM [71] (LSVM) classifier was trained on the three feature sets within
the RADIoT framework. The CT COVID-19 image dataset [72] consisting of 1134
image samples is split into train-test sets of 80–10 ratio. So, 80% of the dataset is
used to train the model, while 20% is used to test and assess the model performance.
The performance indicators used in this chapters are accuracy, precision, recall and
F1-Score. For all metrics except accuracy, the closer their values to 1, the better
the model. The highest value of accuracy is 100%. The results of the experiment
are presented in Table 4. It could be observed that based on accuracy metric, the
combined featured dataset RadF + DeepF outperforms other datasets with 93%
124 S. O. Folorunso et al.

Table 4 The outcome of RADIoT analysis on CT images of COVID-19 patients

Dataset Accuracy Recall Precision F1-score
RadF 85.91 0.857 0.87 0.87
DeepF 75.91 0.745 0.75 0.75
RadF + DeepF 93 0.925 0.91 0.91

accuracy. Likewise, for other metrics RadF + DeepF gave a superior performance in
terms of recall, precision and F1-Score with values 0.925, 0.91 and 0.91, respectively.
This unifying framework could lead to better-quality of life and increased satisfac-
tion of patients. The pipeline uses manual segmentation but automated with methods
discussed earlier. This framework is also structured to administer medical imaging
datasets from different sources (Multimodal) as in the case of study of COVID-19
epidemic dataset.

7 Conclusion and Future Research Directions

IoT comprises of a massive number of various devices that are linked together to
transmit vast amounts of image data. Radiomics and ML models depend largely on
this data for further analysis. This chapter proposes a new AI hybrid framework for
the unification of IoT, radiomics and DL named RADIoT; little effort was put on
fine-tuning the parameters of the DL model.
The RADIoT framework proposed in this chapter aimed at the unifying the DL
and radiomics features for clinical image analysis and classification. The inclusion
of the IoT technology in the framework is to supply the needed data and to also
enable real-time result upload. It removes a lot of bottleneck-like different patient
report mix-up and access to consultation. The RADIoT structure is illustrated with a
DL architecture for CT and through IoT; deployed a transfer learning strategy. This
approach is essential valuable when dealing with small sample sized dataset while
achieving a good classification performance. This structure of RADIoT pipeline can
also model the other radiomic task with multiple modalities (CXR, PET, MRI and US)
and anatomical regions like the brain, breast, lungs are also considered. This chapter
also described the approaches to Radiomics process flow with mining handicraft
features, DL features and the combination of both methods, their challenges and
opportunities.
For future work, further investigation is needed to validate the accuracy and
effectiveness of the proposed framework.
RADIoT: The Unifying Framework for IoT, Radiomics … 125

References

1. van Griethuysen, J. J. M., Fedorov, A., Parmar, C., Hosny, A., Aucoin, N., Narayan, V., Beets-
Tan, R. G. H., Fillion-Robin, J.-C., Pieper, S., & Aerts, H. J. W. L. (2017). Computational
radiomics system to decode the radiographic phenotype. Cancer Research, 77(21), e104–e107.
2. Lambin, P., Rios-Velazquez, E., Leijenaar, R., Carvalho, S., van Stiphout, R. G. P. M., Granton,
P., Zegers, C. M. L., Gillies, R., Boellard, R., Dekker, A., & Aerts, H. J. W. L. (2012). Radiomics:
Extracting more information from medical images using advanced feature analysis. European
Journal of Cancer, 48, 441–446.
3. Gillies, R. J., Kinahan, P. E., & Hricak, H. (2016). Radiomics: Images are more than pictures,
they are data. Radiology, 278(2), 563–577.
4. Akmandor, O. A., & Jha, N. K. (2017). Smart health care: An edge-side computing perspective.
IEEE Consumer Electronics Magazine, 7(1), 29–37.
5. Greco, L., Percannella, G., Ritrovato, P., Tortorella, F., & Vento, M. (2020). Trends in IoT based
solutions for health care: Moving AI to the edge. Pattern Recognition Letters, 135, 346–353.
6. Chen, P.-H., & Cross, N. (2018). IoT in radiology: Using Raspberry Pi to automatically log
telephone calls in the reading room. Journal of Digital Imaging, 31, 371–378.
7. Gil, D., Ferrández, A., Mora-Mora, H., & Peral, J. (2016). Internet of Things: A review of
surveys based on context aware intelligent services. Sensors, 16.
8. Upton, E. (2016). Ten Millionth raspberry pi, and a new kit. Raspberry Pi. [Online].
9. Mahdavinejad, M. S., Rezvan, M., Barekatain, M., Adibi, P., Barnaghi, P., & Sheth, A. P. (2018).
Machine learning for internet of things data analysis: A survey. Digital Communications and
Networks, 4, 161–175.
10. Qin, Y., Sheng, Q. Z., Falkner, N. J., Dustdar, S., Wang, H., & Vasilakos, A. V. (2016). When
things matter: A survey on data-centric internet of things. Journal of Network and Computer
Applications, 64, 137–153.
11. Sheng, Z., Yang, S., Yu, Y., Vasilakos, A. V., McCann, J. A., & Leung, K. K. (2013) A survey
on the IETF protocol suite for the Internet of Things: Standards, challenges and opportunities.
IEEE Wireless Communications, 20(6), 91–98.
12. Yadav, A., Kumar Singh, V., Kumar Bhoi, A., Marques, G., Garcia-Zapirain, B., & de la Torre
Díez, I. (2020). Wireless body area networks: UWB wearable textile antenna for telemedicine
and mobile health systems. Micromachines, 11(6), 558.
13. Marques, G., Bhoi, A. K., de Albuquerque, V. H. C., K.S., H. (Eds.), (2021). IoT in healthcare
and ambient assisted living. Springer.
14. Marques, G., Miranda, N., Kumar Bhoi, A., Garcia-Zapirain, B., Hamrioui, S., & de la Torre
Díez, I. (2020). Internet of Things and enhanced living environments: Measuring and mapping
air quality using cyber-physical systems and mobile computing technologies. Sensors, 20(3),
720.
15. Oniani, S., Marques, G., Barnovi, S., Pires, I. M., & Bhoi, A. K. (2020). Artificial intelligence for
internet of things and enhanced medical systems. In Bio-inspired neurocomputing (pp. 43–59).
Springer.
16. Chandy, A. (2019). A review on IoT based medical imaging technology for healthcare
applications. Journal of Innovative Image Processing (JIIP), 1(1), 51–60.
17. Lambin, P., Leijenaar, R. T. H., Deist, T. M., Peerlings, J., de Jong, E. E. C., van Timmeren, J.,
Sanduleanu, S., Larue, R. T. H. M., Even, A. J. G., Jochems, A., van Wijk, Y., Woodruff, H., van
Soest, J., Lustberg, T., Roelofs, E., van Elmpt, W., Dekker, A., Mottaghy, F. M., Wildberger,
J. E., & Walsh, S. (2017) Radiomics: The bridge between medical imaging and personalized
medicine. Nature Reviews Clinical Oncology, 17, 749–762.
18. Bizzego, A., Bussola, N., Salvalai, D., Chierici, M., Maggio, V., Jurmany, G., & Furlanello, C.
(2016) Integrating deep and radiomics features in cancer bioimaging.
19. Court, L. E., Fave, X., Mackin, D., Lee, J., Yang, J., & Zhang, L. (2016). Computational
resources for radiomics. Translational Cancer Research, 5, 340–348.
126 S. O. Folorunso et al.

20. Balagurunathan, Y., Gu, Y., Wang, H., Kumar, V., Grove, O., Hawkins, S., Kim, J., Goldgof,
D. B., Hall, L. O., Gatenby, R. A., & Gillies, R. J. (2014). Reproducibility and prognosis of
quantitative features extracted from CT images. Translational Oncology, 7(1), 72–87.
21. Hui, G., & Oksam, C. (2010). Individual tooth segmentation from CT images using level set
method with shape and intensity prior. Pattern Recognition, 43(7), 2406–2417.
22. Ye, X., Beddoe, G., & Slabaugh, G. (2010). Automatic graph cut segmentation of lesions in
CT using mean shift superpixels. International Journal of Biomedical Imaging, 983963.
23. Chen, X., Udupa, J. K., Bagci, U., Zhuge, Y., & Yao, J. (2012). Medical image segmentation
by combining graph cuts and oriented active appearance models. IEEE Transactions on Image
Processing, 21(4), 2035–2046.
24. Suzuki, K., Kohlbrenner, R., Epstein, M. L., Obajuluwa, A. M., Xu, J., & Hori, M. (2010).
Computer-aided measurement of liver volumes in CT by means of geodesic active contour
segmentation coupled with level-set algorithms. Medical Physics, 37(5), 2159–2166.
25. Zhou, M., Scott, J., Chaudhury, B., Hall, J., Goldgof, D., Yeom, K. W., Ou, I. M. Y., Kalpathy-
Cramer, J., Napel, S., Gillies, R., Gevaert, O., & Gatenby, R. (2018). Radiomics in brain
tumor: Image assessment, quantitative feature descriptors, and machine-learning approaches.
American Journal of Neuroradiology, 39(2), 208–216.
26. Dalal, N., & Triggs, B. (2005). Histograms of oriented gradients for human detection. In
Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern
Recognition (CVPR ‘05), San Diego, Calif, USA, June 2005.
27. Ojala, T., Pietikäinen, M., & Harwood, D. (1994). Performance evaluation of texture measures
with classification based on Kullback discrimination of distributions. In Proceedings of the
12th IAPR International Conference on Pattern Recognition (ICPR 1994).
28. Nanni, I., Lumini, A., & Brahnam, S. (2010). Local binary patterns variants as texture
descriptors for medical image analysis. Artificial Intelligence in Medicine, 49(2), 117–125.
29. Khoshgoftaar, T., Dittman, D., Wald, R., & Fazelpour, A. (2013). First order statistics based
feature selection: A diverse and powerful family of feature seleciton techniques. In Proceedings
of 11th International Conference on Machine Learning and Applications, Boca Raton, FL, Boca
Raton, Florida.
30. Rivera, A. R., Castillo, J. R., & Chae, O. O. (2013). Local directional number pattern for face
analysis: Face and expression recognition. IEEE Transactions on Image Processing, 22(5),
1740–1752.
31. Song, T., Li, H., Meng, F., Wu, Q., & Cai, J. (2018). LETRIST: Locally encoded transform
feature histogram for rotation-invariant texture classification. IEEE Transactions on Circuits
and Systems for Video Technology, 28(7), 1565–1579.
32. Kannala, J., & Rahtu, E. (2012). BSIF: Binarized statistical image features. In Proceedings of
the 21st International Conference on Pattern Recognition (ICPR2012), Tsukuba.
33. Belahcene, M., Laid, M., Chouchane, A., Ouamane, A., & Bourennane, S. (2016). Local
descriptors and tensor local preserving projections in face recognition. In Proceedings of the
6th European Workshop at the Visual Information Processing (EUVIP), Marseille, France.
34. Lillholm, M., & Griffin, L. (2008). Novel image feature alphabets for object recognition. In
19th International Conference on Pattern Recognition (ICPR 2008), Tampa, Florida, USA.
35. Simonyan, K., & Zisserman, A. (2015). Very deep convolutional networks for large-scale image
recognition. In International Conference on Learning Representations (ICLR).
36. He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition.
InProceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las
Vegas, NV.
37. Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., & Wojna, Z. (2016). Rethinking the inception
architecture for computer vision. In Proceedings of IEEE Conference on Computer Vision and
Pattern Recognition (CVPR), Las Vegas, NV.
38. Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L. (2018). MobileNetV2: Inverted
residuals and linear bottlenecks. In Proceedings of IEEE/CVF Conference on Computer Vision
and Pattern Recognition, Salt Lake City, UT.
RADIoT: The Unifying Framework for IoT, Radiomics … 127

39. Huang, G., Liu, Z., Van Der Maaten, L., & Weinberger, K. Q. (2017). Densely connected
convolutional networks. In Proceeding IEEE Conference on Computer Vision and Pattern
Recognition (CVPR), Honolulu, HI.
40. Tan, M., & Le, Q. V. (2019). EfficientNet: Rethinking model scaling for convolutional neural
networks. In Proceedings of the 36th International Conference on Machine Learning (ICML
2019), Long Beach, California.
41. Zoph, B., Vasudevan, V., Shlens, J., & Le, Q. V. (2018). Learning transferable architectures for
scalable image recognition. In Proceedings IEEE/CVF Conference on Computer Vision and
Pattern Recognition, Salt Lake City, UT.
42. Huynh, B., Li, H., & Giger, M. L. (2016). Digital mammographic tumor classification using
transfer learning from deep convolutional neural networks. Journal of Medical Imaging
(Bellingham), 2(3), 034501.
43. Hotelling, H. (1933). Analysis of a complex of statistical variables into principal components.
Journal of Educational Psychology, 24(6), 417.
44. Gil, D., Díaz-Chito, K., Sánchez, C., & Hernández-Sabaté, A. (2020). Early screening of
SARS-CoV-2 by intelligent analysis of X-ray images. arXiv preprint arXiv:2005.13928.
45. Motwani, M., Dey, D., Berman, D. S., Germano, G., Achenbach. S., Al-Mallah. M. H., Chang,
H. J., et al. (2017). Machine learning for prediction of all-cause mortality in patients with
suspected coronary artery disease: A 5-year multicentre prospective registry analysis. European
Heart Journal, 38(7), 500–507.
46. Agrawal, R. K., Kaur, B., & Sharma, S. (2020). Quantum based whale optimization al-gorithm
for wrapper feature selection. Applied Soft Computing, 89(106092).
47. Wiharto, W., Suryani, E., & Cahyawati, V. (2019). The methods of duo output neural network
ensemble for the prediction of coronary heart disease. Indonesian Journal of Electrical
Engineering and Informatics (IJEEI), 7(1), 51–58.
48. Nilashi, M., Bin Ibrahim, O., Ahmadi, H., & Shahmoradi, L. (2017). An analytical method for
diseases prediction using machine learning techniques. Computers and Chemical Engineering,
106, 212–223.
49. Liu, H., & Lang, B. (2019). Machine learning and deep learning methods for intrusion detection
systems: A survey. Applied Sciences, 9(4396), 1–28.
50. Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions.
InProceedings of the Annual Processing Systems, Long Beach, CA, USA.
51. Ordóñez, P. F., Cepeda, C. M., Garrido, J., & Chakravarty, S. (2017). Classification of images
based on small local features: A case applied to microaneurysms in fundus retina images.
Journal of Medical Imaging, 4(4), 041309.
52. Shafiee, M. J., Chung, A. G., Khalvati, F., Haider, M. A., & Wong, A. (2017). Discovery
radiomics via evolutionary deep radiomic sequencer discovery for pathologically proven lung
cancer detection. Journal of Medical Imaging, 4(4), 041305.
53. Vaidhya, K., Thirunavukkarasu, S., Alex, V., & Krishnamurthi, G. (2016). Multi-modal
Brain Tumor Segmentation Using Stacked Denoising Autoencoders. In Brainlesion: Glioma,
Multiple Sclerosis, Stroke and TMultiple Sclerosis, Stroke and Traumatic Brain Injuries.
(BrainLes 2015). Lecture notes in computer science.
54. Alex, K. V., Thirunavukkarasu, S., Kesavadas, C., & Krishnamurthi, G. (2017). Semisupervised
learning using denoising autoencoders for brain lesion detection and segmentation. Journal of
Medical Imaging (Bellingham), 4(4), 041311.
55. Li, H., Giger, M. L., Huynh, B. Q., & Antropova, N. O. (2017). Deep learning in breast cancer
risk assessment: evaluation of convolutional neural networks on a clinical dataset of full-field
digital mammograms. Journal of Medical Imaging (Bellingham), 4(4), 041304.
56. Liu, S., Xie, Y., Jirapatnakul, A., & Reevesa, A. P. (2017). Pulmonary nodule classification
in lung cancer screening with three-dimensional convolutional neural networks. Journal of
Medical Imaging (Bellingham), 4(4), 041308.
57. Shahedi, M., Cool, D. W., Bauman, G. S., Bastian-Jordan, M., Fenster, A., & Ward, A. D.
(2017). Accuracy validation of an automated method for prostate segmentation in magnetic
resonance imaging. Journal of Digit Imaging, 30, 782–795.
128 S. O. Folorunso et al.

58. Cheng, R., Turkbey, B., Gandler. W., Agarwal, H. K., Shah, V. P., Bokinsky, A., McCreedy, E.,
Wang. S., Sankineni, S., Bernardo. M., Pohida. T., Choyke, P., & McAuliffe, M. J. (2014). Atlas
based AAM and SVM model for fully automatic MRI prostate segmentation. In Conference
Proceedings of IEEE Engineering Medical Biology Society (pp. 2881–2885).
59. Runkler, T. A. (2016). Data analytics: Models and algorithms for intelligent data analysis (2nd
ed., p. 158). Springer Vieweg.
60. Folorunso, S. O., & Adeyemo, A. B. (2013). Alleviating classification problem of imbalanced
dataset. African Journal of Computing and ICT, 6(1), 137–144.
61. Afshar, P., Mohammadi, A., Plataniotis, K. N., Oikonomou, A., & Benali, H. (2019). From hand-
crafted to deep-learning-based cancer radiomics: Challenges and opportunities. IEEE Signal
Processing Magazine, 36, 132–160.
62. Kumar, S. M., & Majumder, D. (2018). Healthcare solution based on machine learning appli-
cations in IoT and edge computing. International Journal of Pure and Applied Mathematics,
119(16), 1473–1484.
63. Gillies, R., Kinahan, P., et al. (2016). Radiomics: Images are more than pictures, they are data.
Radiology, 278(2), 563–577.
64. Folorunso, S. O., & Adeyemo, A. B. (2012). Theoretical comparison of undersampling tech-
niques against. In EIE’s 2nd International Conference on Computing, Energy, Networking,
Robotics and Telecommunication (EIE 2012).
65. Chawla, N., Bowyer, K., Hall, L., & Kegelmeyer, P. (2002). SMOTE: Synthetic minority
over-sampling technique. Journal of Artificial Intelligence Research, 16, 321–357.
66. Wilson, D. L. (1972). Asymptotic properties of nearest neighbour rules using edited data. IEEE
Transactions on Systems, Man, and Cybernetics, 2, 408–421.
67. Tomek, I. (1976). An experiment with the edited nearest-neighbor rule. IEEE Transactions on
Systems, Man, and Cybernetics, 6(6), 448–452.
68. He, H., Bai, Y., Garcia, E. A., & Li, S. (2008). ADASYN: Adaptive synthetic sampling approach
for imbalanced learning. In IEEE International Joint Conference Neural Networks, Hong Kong.
69. Vallières, M., Kay-Rivest, E., Perrin, L. J., Liem, X., Furstoss, C., Aerts, H. J. W. L., Khaouam,
N., Nguyen-Tan, P. F., Wang, C. S., Sultanem, K., Seuntjens, J., & El Naqa, I. (2017). Radiomics
strategies for risk assessment of tumour failure in head-and-neck cancer. Scientific Reports,
7(1), 10117.
70. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M.,
Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher,
M., Perrot, M., & Duchesnay, E. (2011). Scikit-learn: Machine learning in python. Journal of
Machine Learning Research, 12, 2825–2830.
71. Vapnik, V. N. (1998). Adaptive and learning systems for signal processing communications,
and control. In Statistical learning theory.
72. https://github.com/ieee8023/covid-chestxray-dataset. [Online].
73. https://commons.wikimedia.org/wiki/Category:Magnetic_resonance_imaging#/media/File:
Petmr.jpg
74. https://en.wikipedia.org/wiki/X-ray_machine#/media/File:Projectional_radiography_compon
ents.jpg
75. https://en.wikipedia.org/wiki/CT_scan#/media/File:UPMCEast_CTscan.jpg
76. https://en.wikipedia.org/wiki/Positron_emission_tomography#/media/File:ECAT-Exact-HR--
PET-Scanner.jpg
77. https://en.wikipedia.org/wiki/Medical_ultrasound#/media/File:AlokaPhoto2006a.jpg
Hybrid Artificial Intelligence and IoT
in Health care for Cardiovascular Patient
in Decision-Making System

M. Safa, A. Pandian, T. Kartick, K. Chakrapani, G. Geetha, and G. Saranya

Abstract Many people became more anxious of a sudden heart arrest lately. As
smart wearable devices becoming more popular, an opportunity has become more
open to deliver the Internet of Things (IoT) solution. Unfortunately, for patients
recovering from sudden heart arrests, hospital survival rates are poor. The goal of
this study is to create an IoT device that collects body area sensor (BAS) data with
the intent of providing early warning about an imminent heart arrest. This study aims
to establish a clinical surveillance method for cardiovascular patients who stay home
(health care) and the advice on treatment (health care). The vital signs of patients
(heart rate, blood pressure and cardiac rhythm) are tracked by means of sensors
that report to the M2M server. The server would then immediately prescribe the
drug using a case-based inference method called artificial intelligence (AI), which
incorporates multiple hospital care reports and other sources. The stress analysis
prediction model uses a machine learning algorithm, including a decision tree, a
classifier K-neighbours and a support vector classifier. In a data ban known as the
Blob storage scheme, the data of the patient are saved and secured and run by the

M. Safa · A. Pandian · T. Kartick (B) · G. Geetha · G. Saranya

School of Computing, SRM Institute of Science and Technology, Kattankulathur, Chennai, Tamil
Nadu 603203, India
e-mail: [email protected]
M. Safa
e-mail: [email protected]
A. Pandian
e-mail: [email protected]
G. Geetha
e-mail: [email protected]
G. Saranya
e-mail: [email protected]
K. Chakrapani
Department of Information Technology, School of Computing, SASTRA Deemed University,
Tanjore, Tamil Nadu 613401, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 129
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_7
130 M. Safa et al.

SQL server. The new patient data can be interacted by processing via Kubernetes
with current patients in data storage using the machine learning model framework.
The comparable findings assess the stress thresholds and therefore the seriousness
of the heart attack. The IoT sensor-based platform represents a safe framework to
monitor, control and interpret patient information and to address the needs of specialty
hospitals.

Keywords IoT · Stress · Classifier · Containers · Sensors

1 Introduction

The quality of health of those who suffer from heart attacks has a huge impact.
With the systematic use of emerging technology, innovative healthcare systems are
theoretically available. Our lives are being revolutionised by the development of an
intelligently wearable IoT device for tracking health [1]. In recent years, emergency
facilities have made substantial strides. The promise of computing and communica-
tion technology is to include a broader spectrum of medical care. Through making
this progress, the quality of life of a patient will increase and support a significant
majority of the community.
By having wearable IoT devices and enhancing them it allows patients to track
their healthcare assessments and review them. An example of the advantages of
having such instruments is that the patient will still be made conscious of the state of
his disease. The healthcare staff will then use this information to immediately adapt
to a condition or to save the user’s life in an emergency. Connected health is a big
technology advancement application.
The idea of linked medical systems and intelligent interconnected IoT applica-
tions has possible advantages for both enterprises and individuals. The aim is to use
emerging technology surveys to build, refine and extend linked health services to
establish a framework that can allow patients to get a better view of their health and
to provide early health alerts.
The aim of IoT is to link everything to everything and anyone through every
network and service at all times [2, 3]. In several fields, including communication
and software, this purpose needs further progress. Most R&D organisations invest
in growth programmes. In other words, IoE is the network of computers and devices
of all sizes, all communication and exchanging of information. Cisco describes the
Internet of All (IoE) as networking for humans, data, items and processes in linking
networks ([3]). Cisco says that by 2020, 50 billion internet-connected computers will
be available [4]. As network of networks, IoT may be identified.
For the IoT architectural system, namely IEEE P2413 [5], a special IEEE speci-
fication is under development. This norm sets IoT for the creation, application and
management of the physical environment through intelligent networking, the compre-
hensive data collection, predictive analytics and optimization, along with information
Hybrid Artificial Intelligence and IoT in Health care … 131

and communication technology (ICT). The IoT standard presents a model of refer-
ence, describes architectural elements and provides implementation processes for
the structures concerned.
If the Internet expands, one of the major facilities is the global introduction of IPv6
[7], which facilitates the omnivorous handling of all “intelligent” communications.
The new IoT link and integration models would provide you with access to billions
of clever objects. However, more needs will be applied to network features, network
maintenance and network composition as a result of network growth. The IPv6
must allow heterogeneous IoT components and heterogeneous applications to be
connected to each other. Low-power wireless networks: 6LoWPAN is a clone of
IPv6 streamlined. It is effectively IPv6 on IoT computers with minimal resources.
IoT protection is one of the key research concerns because the rising number of
connected devices needs to be safeguarded. For example, it must be assured that IoT
devices supply only approved entities with information [20]. The production of IoT
hardware has many associated research challenges, with the launch of new products
and many of them are compact and battery life. In addition, communication protocols
can incorporate the IoT sensor devices in the Internet. The low energy of the sensor
battery must be taken into account, in particular when sensors in remote areas are
being used.
Many of the protocols that consider low-power consumption for IoT devices are
developed and further developed. For example, an appropriate service protocol for
IP-based omnipresent sensor networks for service announcement and discovery [8]
is proposed. The protocol is completely spread, ensuring optimum acquisition times,
low consumption of resources and low overhead generation with a timely response
to changes in the topology. The protocol is able to achieve optimum buying times
with low-energy costs and generated overhead, making it ideal for mobile networks.
In order to bring low-capacity applications smoothly into the Internet, the Internet
Engineering Task Force has carried out a major standardisation work for the
constrained application protocol (CoAP). CoAP can be found on most applications
supporting the user data protocol, and a hot subject of study is the network architecture
using this protocol [22–26]. IoT systems use multiple protocols (Bluetooth, Zigbee,
etc.) and networks (LANs; WANs). The cloud computing is used as a framework to
support IoT networks for integrating vast numbers of computers and sensors. The
IoT platform, thus, has three major building blocks. Instead of introducing different
ways of making all sensors interact directly with each other, IoT-based healthcare
systems may use cloud computing services.
Many research projects aim to describe an irregular heart of the user; however,
others lack key components. Many people already do eHealth studies, and a variety
of organisations have used this work by building systems that bind patients to physi-
cians worldwide. Two separate types of similar technologies are examined: full health
care through embedded systems and linked mobile eHealth apps. We are creating
an application on the mobile that links to a smart IoT computer, and most organisa-
tions rely on robust healthcare networks allowing people to communicate with each
other and to profit from services. Our proposed solution is more linked to eHealth
applications.
132 M. Safa et al.

1.1 Comprehensive Health Care Systems

The aim of “PatientsLikeMe” was to help patients answer the question. “What is the
best result I can hope for given my position, and how do I get there?” We respond
to patients’ questions in different ways such as the interaction and the exchange of
interactions with patients with related conditions [27]. But data protection and device
compatibility were not listed.
Another system is known as “DailyStrength.” It is a social network based on
communities, where members get shared emotional support by communicating to
each other about their challenges and accomplishments. This site provides online
forums coping with numerous medical problems or life threats [28]. “Patients-
LikeMe” is somewhat close in that these are free channels featuring patient and
doctors. The fact that “DailyStrength” does not have research institutes nor has a
smartphone application is two big divides between them. Both systems are also
available as IoT systems are not IoT-based system.
A powerful model, including several pulse parameters, EEGs and skin leadership
sensors into a shirt, was further developed in another work [42]. Another method has
been developed to link facial expressions with EEG patterns and voice recognition
[43]. Other researchers have shown that EEG alone has distinct emotional char-
acteristics [44]. In order to help recognise variations of different emotions, facial
recognition software has been equivalent to heart rate variability [45]. Any pulse
rhythms have also been shown to be linked to physical stress and not mental stress
[46]. However, their applications are handheld, and IoT has not been used as their
device platform.
The “Omnio”, a full application for medical services, is also a robust health-
care system [29]. It offers clinical tools, medical resources, condition guides and
knowledge on drugs, among its programmes. Everyday Health [30] is a website-
based organisation that provides health and fitness content. It has higher scores and
publishes numerous medical papers which can benefit patients a lot. It also has an
intuitive search that allows users to access their materials quickly.

1.2 Connected eHealth Mobile Applications

Although both of the aforementioned programmes provide health care, they do not
provide intelligent devices that can track users’ day-to-day behaviours and warn them
if necessary. A lot of heart monitors send users their ECG messages so that they can
check up on their status, but none of them alerts users to emergencies. A smart elderly
home monitoring system (SEHMS) designed and built with an accelerometer on
Android TM-based smartphones could detect a consumer downtime [31]. It offers
a user interface for smartphones to display the health information obtained from
the device. The biggest asset of the SEHMS is that it has an aged and chronically
Hybrid Artificial Intelligence and IoT in Health care … 133

aggressive patient’s remote screening service. In case of an emergency, though, the

consumer cannot alert him.
The remote mobile health monitoring (RMHM) is a monitoring device for the
medical parameters of a patient, including his/her heart rate, calculated by wear-
able sensors [32]. It helps carers and lovers to track the remote diagnosis by the
patient. The device has no real-time tracking capabilities. Our analysis centres on
the theory of prediction of heart injuries, and this remains a problem. Each group of
researchers defines their own approach to the achievement of their goal. In order to
avoid cardiac abnormalities, we agreed to use a mix of body temperature and ECG.
Other devices with common hardware implementations have different methods. None
of them spoke during data collection about the power consumption levels. Our device
uses a Bluetooth low-power module to capture longitudinal data using a smartphone
wirelessly.
In [33, 34] authors contrasted several methods of data mining for cardiovascular
identification. They have only projection algorithms and no whole structure with a
data collector and a network of programming. The best approaches are as follows:
decision tree, Naïve Bayes, the neural net and k-mean for the estimation of heart
issues. Our testing includes not only a detailed IoT and a virtual platform system,
but also a decision tree that uses data mining technologies to predict heart attacks.
This is unique in our method because we use a low level, intelligent IoT system
and data extraction technologies in our prediction algorithm. All our healthy, toxic
test subjects are identified with considerable precision when we tested our predictive
algorithm. The comparison between our system and related works can be explained
in Tables 1 and 2.
We suggest a smart IoT-based cardiac rate monitoring system in order to overcome
the limitations of the listed research and systems. In this article, our framework is
designed to explicitly resolve some of the disadvantages and theoretically positive
prediction outcomes of current systems. The key feature of our device is the warning,
which helps the customer to avoid damage before it eventually occurs. Our system is
the first intelligent IoT healthcare system to use a subject-specific body area sensor
to detect cardiac damage to the best of our understanding.

1.3 Artificial Intelligence

This research uses the case-based (CBR) methodology to assess a patient’s disease
type to establish artificial intelligence approach. CBR itself is a problem-solving
technique that applies to the experience of common prior cases [42]. The CBR
solution consists of four steps:
• Retrieve: a problem identification process and a search query with related
instances.
• Reuse: a method in which we use knowledge from other related situations to solve
our dilemma.
134 M. Safa et al.

Table 1 Qualitative comparison of existing work based on different features

Approach Use IoT device Mobility Low-power Cyber physical Cost effective
system system
Patientlike [27] ✕ ✓ ✕ ✕ ✕
Daily strength ✓ ✓ ✕ ✕ ✕
[28]
Omnio [29] ✓ ✓ ✓ ✓ ✕
Everyday ✓ ✕ ✕ ✕ ✕
Health [30]
SEHMS [31] ✕ ✓ ✕ ✓ ✕
RMHM [32] ✕ ✕ ✕ ✓ ✕
PHM [33] ✓ ✓ ✕ ✓ ✕
Qardiocore ✕ ✕ ✕ ✓ ✕
[34]
Maksimović ✕ ✕ ✕ ✓ ✕
[35]
Stecker [36] ✕ ✕ ✕ ✓ ✕
Mancini [37] ✕ ✕ ✕ ✓ ✕
Sun [38] ✕ ✕ ✕ ✓ ✓
Communicore ✕ ✕ ✕ ✓ ✕
[39]
Kavitha1 [40] ✓ ✕ ✕ ✓ ✕
Jagtap [41] ✕ ✕ ✕ ✓ ✕

• Revised: a method for measuring, analysing and updating details in order to

provide a more reliable approach.
Retain: indexing, incorporating and collecting the new approach. It is then
stored in the database as a new information (Fig. 1).
If we have the closest case to find in our current case, we will compute the most
widely used nearest neighbour algorithm (NN) [43]. NN is based on locating the
nearest point to current cases, in the new case, where “cases” are symbolised as
“points”.

1.3.1 Patient Side

First of all, an M2M sensor must be installed on the patient side to catch the critical
signals and transmit them to the server. As described above, the M2M sensor consists
of a patient monitoring unit (e.g. the GE Patient Monitor B20). In order to relay the
output data to microcontrollers, the patient display should be updated. It is then
necessary to configure the microcontroller, so that it constantly sends the data to the
server. The first part works like this. The sensor tests the weight of each pill container
Hybrid Artificial Intelligence and IoT in Health care … 135

Table 2 Quantitative comparison of existing work based on different features

Attribute Avg. max Approx Avg. max No. of Power
HR accuracy (%) sampling rate device used consumption
Patients like me 162 91 115 One 500 mW
[27]
Daily strength 154 87 112 One N/A
[28]
Omnio [29] 143 83 105 One N/A
Everyday health 144 81 85 One N/A
[30]
SEHMS [31] 156 75 96 Two N/A
RMHM [32] 166 82 142 Two N/A
PHM [33] 142 74 155 One N/A
Qardiocore [34] 138 73 115 One N/A
Maksimović [35] 154 89 100 Two N/A
Stecker [36] 162 71 135 One N/A
Mancini [37] 157 86 140 Two 600 mW
Sun [38] 164 79 90 One N/A
Communicore 149 73 155 One N/A
[39]
Kavitha1 [40] 157 65 160 One N/A
Jagtap [41] 144 72 140 Two N/A

Fig. 1 Cycle of CBR method

136 M. Safa et al.

for an initial value when a drug is recommended to the patient (w0). A certain short
time is waiting for the sensor (microcontroller). If the weight (w1) is decreased by
the value of the drug (previous to calculate the pill weight with the predetermined
value), the patient would have taken one pill. If the weight appears to be the same, the
patient has not yet taken the drug. The last aspect of the patient side is the Android
smartphone edition programme that shows the patient’s well-being. If the sensors
collect the vital signs of the patient, the data are then sent to the database server and
the details are stored in a text file with information inside (e.g. patient. txt).

1.3.2 Server Side

There is a storage server on the server side to store the data and a web server for
website presentation as well. In addition, an artificial intelligence computing device
is required to assess and prescribe a treatment for the patient. First of all, the user
must login to get access to the web presentation. He/she can only access the infor-
mation inside, and he/she can only view sensitive details with a patient-type user,
while the doctor-type user can view patient information under his/her supervision.
Artificial intelligence is the secret to the decision-making of a recommendation, but it
should eventually be accepted by the psychiatrist. The most significant feature on the
computer is the server in the database. The patient, the doctor and the predetermined
data are kept, which are analysed by artificial intelligence to generate suggestions
for treatment as appropriate. This illustration below is an extract of data from the
server in the archive.

2 Data Source

A large number of medical records have been collected in hospital libraries. In

different circumstances, the word heart attack circulates and is dangerous to the
human heart. One of nature’s deadliest diseases is cardiovascular disease. “Cardio-
vascular disease” refers to the state of the heart and blood vessels and to the effect
of blood pumping and blood supply in the body. The Cleveland, Hungarian, Swiss,
Long Beach VA cardiac condition database reports were accessed (UCI machine
Learning Repository). Data sets distinguish disease-related trends. Data are split
into two data sets: data set training and data set research. A total of 920 documents
were collected along with 76 medical attributes. The list of 14 attributes on which
the device operates is seen in the following table (Table 1.).
Hybrid Artificial Intelligence and IoT in Health care … 137

2.1 Analysis of Data

Data pre-processing functions such as data washing, data integration, missing values
are done, and redundant data are eliminated as the data set contains missing values
and redundant data are processed. The forecast leads to fault.
Now, we explain briefly the KDD steps seen. The first step is the most critical
step in defining the problem and deciding the key criteria influencing the project.
The second step is to analyse the data material. This phase involves mainly data
analysis, data definition, data study and data quality assessment. The third stage is
the preparation of data including data washing, data collection, data analysis and data
integration. The fourth step is the model design from which the model will forecast.
The model’s efficiency is evaluated in the fifth stage. Finally, once it is fit for use,
a range of processes such as software creation, maintenance and the preparation of
the final report should be followed (Fig. 2).
Different approaches were used with data processing tools to the early detection
of heart disease [42–47]. The most serious risk to human well-being is auditory
artery blockage. Many scholars are seeking to derive important lessons from the
perspectives and experimental findings of physicians. Min et al. [49] used multilayer
receptors to build an intelligent device for the determination of five severe cardiovas-
cular disorders. C4.5 was used in [48, 49] to detect the association between cardiac
problems and the possible causes for these conditions. Sundar et al.[50] developed
a computerised Bayes and weighted associative classifier (WAC) method to esti-
mate the risk of a heart condition. Ordonez used [51] to reduce the number of its
recommended association rules using a new algorithm to reduce the search space. In
conjunction with ignorant Bayes, Chenand et al. [52] used an intelligent web-based
approach to enable medical practitioners to improve their clinical decision-making
and early detection of the disease.

Fig. 2 Procedure of knowledge discovering from database

138 M. Safa et al.

3 Materials and Methods

3.1 Data Gathering

For the proposed predictive model, we have used the data set in the UCI machine
learning repository. This archive includes many cancer, asthma, cardiovascular and
other data sets. The data collection for cardiovascular patients is used in the presented
analysis for model assessment. This data set provides 303 patients with details.
Features are categorical, integer and actual in this data set. While 76 features consti-
tute a total of 76, all the experiments published are focused on a subset of 14 as
described in Table 3.
• If the data set regularly fails to contain a function (e.g. more than 50 times), it can
then be excluded.
• If the missing rate is less than 10%, then the absent value is replaced by this
function’s average value. In addition, if the type of function is not numeric, the
missing value is filled with classification methods Table 4.

Table 3 List of features for each patient

Attribute Description
Age An integer value representing the age of each patient
Gender Male (1), Female (0)
CP Chest paint type
trestbps Resting blood pressure (in mm Hg on admission to the hospital)
Chol Serum cholesterol in mg/dl
Fbs Fasting blood sugar >120 mg/dl(1 = true;0 = false)
Ca Number of major vessels (0–3) coloured by fluoroscopy
thalach Maximum heart rate achieved
Slope the slope of the peak exercise ST segment; 1: upsloping; 2: flat; 3: downsloping

Table 4 Missing rate of each

Missing value Feature name Missing rate Feature name
feature in data set
rate (%) value (5)
0 cp 0 age
0 exang 0 Gender
1 testbps 3 chol
12 restecg 3 fbs
11 talach 12 oldpeak
0 num 1 ca
14 slope 15 thal
Hybrid Artificial Intelligence and IoT in Health care … 139

3.2 Feature Selection

In our data set, each patient has 76 tasks. The principal objective of the feature selec-
tion is to remove irrelevant and redundant functions, which in terms of accuracy and
complexity degrades classifiers’ efficiency. The methods of selecting functions are
divided into two methods: methods for filtering and wrapping. The consistency of
the wrapper is used as an assessment measure for function subsets in the wrapper
methods. Filters use heuristic algorithms, on the other hand, to calculate the signif-
icance of subsets. Filtering methods have seen in the literature more quickly than
wrapping. In this analysis, the Pearson correlation coefficient [9] approach was used
to pick from 76 characteristics of each patient the most unequal features. This function
collection is based on a two-set dependence. The greater coefficient of association
means higher functional class dependence.

3.3 Classification

We merged numerous classification methods to structure the pattern recognition

model: decision tree (DT), vector support machine (SVM), neural network (NN),

Fig. 3 Description of the trained decision tree classifier

140 M. Safa et al.

logistics regression, Naive Bayes and raw sets. Figure 3 gives an overview of how a
decision tree can be used to construct a pattern detection process.
SVM is a powerful classificatory especially if we face a two-class problem in order
to discriminate between samples by hyperplanes. This definition is a classification
of disparity that is influenced by vectors of support only. The full range as well as the
hyperplane of the separator does not alter without support vector samples. That is
why SVM makes noise and outliers more stable. The kernel principle, which allows
SVM to distinguish samples that cannot be separated by a hyper-fluid, is an important
theory of SVM. We attempted linear and polynomial kernels in this article.
The neural network inspired the emulation of functions by biological neurons.
This role may serve as retrograde or graded. In this article, we used a hidden layer
of multilayer perceptron (MLP). A weighted number is used to measure the output
of each perceptron.

4 Various Machine Learning Algorithms

4.1 Logistic Regression

Logistic regression is considered to be one of the most efficient algorithms for

machine learning for its bi-classification. The approach to a wide range of prob-
lems is easy to apply and is proposed [11]. It functions on the variable of the type.
The variables, win or fail, are the binary-dependent variables such as 0 s or 1 s. When
these values have more than one outcome, a logistic regression with multinomials or
multiple categories is ordered to implement a normal logistic regression. The role of
logistics

1
P = (y = 1/ X ) =
1 + e−wa

where e is the number of the Euler and an is an entry, we insert it in the function, as
shown in tshe logistic regression ROC curve (Fig. 4).

4.2 Naïve Bayes

The Navies Bayes classification algorithm is one of the best Bayesian master educa-
tion classification algorithms. Naive Bayes’ classification algorithm is extremely
scalable and includes linear variables as predictive variables to render a problem
statement [17]. The ranking and regression are similar and competing with SVM
is difficult. The disease-related specialisation is developed in patients. It shows
Hybrid Artificial Intelligence and IoT in Health care … 141

Fig. 4 ROC curve

the potential of each input attribute to forecast the status and provides the likeli-
hood of events. The likelihood of conclusion A is subject to a certain amount of
evidence/comment B, where there is a relation of dependence between A and B. It is
called P(A|B), in which P(A) equals event A, in which P(B) equals event B, in which
P(B|A) is analogous to event B, and subject to event A.

4.3 Random Forest

Random forest is a learning algorithm for classification and regression. It creates

decision trees for each attribute. You have corrected the overfit of your training
package. The assessment phases often delete the missing values, outliers and data
pre-processing. In this type of methodology, the weak model has to be integrated
into a dynamic model. The random tree (Fig. 5) displays the many machine decision-
making bodies.

(P(B/A) ∗ P(B)
P(A/B) =
P(A)

4.4 Support Vector Machine

Vector help is a classification and regression technology algorithm used in machine

learning. Due to its performance comparison with other algorithms, it is regularly
used as classification methods. This technique draws a hyperplane as a coordinate
for each attribute in the data set [16]. The description of the hyperplane dividing the
one-class class from the other shall lead to classification. It presents a framework
that provides another example and makes a binary linear classification difficult.
142 M. Safa et al.

Fig. 5 Random forest tree

4.5 Gradient Boosting

It is a methodology of machine regression learning and grading that is a set of weak

prediction models like decision-making trees. The function that corresponds with
the estimate of heart attack in this data set has different meaning in this gradient
improvement strategy. The diagram below (Fig. 6) shows the differing relevance of
a heart disease forecast using the boosted tree.

4.6 Accuracy Module

The module uses machine learning algorithms to estimate accuracy. The algorithms
that estimate the highest likelihood of having a cardiovascular problem carry this
module to the maximum accuracy. Each algorithm gives a different rate of accu-
racy for the attributes taken which are responsible for cardiovascular disease. The
consistency of your model can be measured using the precision and sensitivity of the
uncertainty matrix can be obtained as seen below:
P

True Negative Rate, Specificity= P+Q
Q
False Positive Rate= Q+P
Sum to 1

R

True Postive Rate,Sensitivity= R+S
R
False Negative Rate= s+R
Sum to 1
Hybrid Artificial Intelligence and IoT in Health care … 143

Fig. 6 Variable importance graph

where P = no. of true negatives, Q = no. of false positives, R = no. of true positives,
and S = no. of false negatives.
The approach which identifies the people with cardiovascular diseases (true posi-
tive rate) and the specificity identifies those without cardiovascular disease (true
positive rate) (true negative rate).

5 Results and Discussion

The findings and accuracies produced in this section are analysed, and the results are
shown. The most reliable result here is the algorithm that has the greatest precision.
Table (Table 5) shows the exactness of each algorithm.
The precision is determined by the data set attributes gathered. This accuracies
can also be enhanced with more and better operating systems and data sets. Based

Table 5 Comparison of
Algorithm name Accuracy Overall accuracy
accuracies
Logistic regression 0.89161585 0.88651685
Random forest 0.88953252 0.81089888
Naïve Bayes 0.91095528 0.85426966
Gradient boosting 0.91070122 0.86426875
SVM 0.8982622 0.78977528
144 M. Safa et al.

Fig. 7 Comparison of algorithms

on the precision of the algorithms, the best algorithm for the forecast is selected. The
following figure provides a description of the results for both algorithms (Fig. 7).
The UI is designed to allow the user to quickly use it. The user interface for the
cardiovascular disease prediction tool is given below.
Figure 7 indicates a patient’s risk of cardiac attack depending on the algorithms.
In this respect, the 14 attributes which enter the patient yield the correct outcome
which is predicted by algorithms. Figure 7 displays the probabilistic meaning of the
medical spectrum and includes the live selector where the theme changes at every
time.

6 Conclusion

The aim of this research is to develop a clinical screening strategy and treatment
recommendations for the cardiovascular patients who remain at home (health) (health
care). Via monitors which are sent to the server, patients are monitored for their main
indicators of heart rate, blood pressure and heart rhythm. The server will then admin-
ister the medicine using an artificial intelligence (AI) case-based inference system,
which involves many hospital and other sources of prescription. A machine learning
algorithm, including a decision tree, a classified K-neighbours and a supporting
vector classifier, is used for the stress analysis prediction model. The patient’s data
were saved and protected and run by the SQL server in a data ban known as the
Blob storage scheme. Through using the machine learning model system, the new
patient data can be interacted via Kubernets with current patients in data storage.
The equivalent outcomes compare the stress rate and therefore the gravity of the
Hybrid Artificial Intelligence and IoT in Health care … 145

heart attack. The IoT sensor-based interface offers a secure system for tracking,
monitoring, analysing and resolving patient information in specialty hospitals.

References

1. ITU-T Global Standards Initiatives Recommendation ITU-T Y.2060 (06/2012). http://www.

itu.int/en/ITU-T/gsi/iot/Pages/default.aspx.
2. Vermesan, O., & Friess, P. (2013). Internet of Things: Converging technologies for smart
environments and integrated ecosystems. River Publishers Series in Communications.
3. Clarke, R. (2013). Smart cities and the internet of everything: The foundation for delivering
next-generation citizen services. Cisco.
4. Evans, D. The Internet of Things how the next evolution of the internet is changing everything.
Cisco IBSG.
5. IEEE Standards Association, P2413—Standard for an architectural framework for the Internet
of Things (IoT). https://standards.ieee.org/develop/project/2413.html.
6. IEEE Standards Association (IEEE-SA). (2015). Internet of Things (IoT) ecosystem study.
IEEE.
7. IETF. (1998). Internet Protocol Version 6 (IPv6) specification, network working group. The
Internet Society.
8. Djamaa, B., & Witty, R. An efficient service discovery protocol for 6LoWPANs. In Proceedings
of Science and Information Conference, SAI 2013. View at: Google Scholar.
9. Safa, M., & Pandian, A. (2021). A review on big IoT data analytics for improving QoS-based
performance in system: Design, opportunities and challenges. In Lecture notes in networks and
systems (Vol. 130, pp. 433–44).
10. https://robotechshop.com/shop/arduino/arduinoboard/arduino-uno-r3-china/?v=7516fd
43adaa.
11. https://www.sparkfun.com/products/11574.
12. https://www.slideshare.net/AmeerKhan3/zigbee-based-patient-monitoring-system-21031306.
13. Raj, P., Safa, M., Kumar, N., Sahrudh, P. J., Singh, S. (2020). Enhanced smart music controller
by applying CNN in IoT. International Journal of Advanced Science and Technology, 29(06),
2739–2749.
14. Saranya, G., Geetha, G., Safa, M., & Meenakshi, K. (2020). Multi-label and multi-class retinal
classification and comparative model of retinal diseases. International Journal of Advanced
Science and Technology, 29(06), 2319–2329.
15. Vadagave, A., Safa, M., Sokhi, S. S., Mathur, B., & Anirudh, B. (2020). Enhancing smart
precision agriculture in IoT using agro ecologicali analysis. International Journal of Advanced
Science and Technology, 29(06), 2750–2760.
16. Bose, A., Safa, M., Siddi, S. B., Anand, M. R., & Kumar, V.. (2020). Implementation of
dynamic lighting and augmented reality (DLAR) smart home system for deaf and hard-of-
hearing (DHH) residents. International Journal of Advanced Science and Technology, 29(06),
2724–2738.
17. Ganesh, A., Safa, M., Roopanjali, M., Sai Sri Harsha, P. V. K., Anjana Tulasi, D. (2020). An
approach for monitoring and analyzing parking occupancies using IoT. International Journal
of Advanced Science and Technology, 29(06), 3536–3545.
18. http://www.jameco.com/z/MY-1--1-uF-100-Volt-Mylar-Capacitor_26956.html.
19. https://www.aliexpress.com/store/product/50-SETS-JST-XH-2-5-2-Pin-Battery-Connector-
Plug-Female-Male-with-140MM-Wire/506373_32598005148.html.
20. https://upload.wikimedia.org/wikipedia/commons/c/c1/ECG-RRinterval.svg.
21. Penney, J. (2016). Choosing an IoT security provider. http://info.deviceauthority.com/blog-da/
choosing-an-iot-security-provider.
146 M. Safa et al.

22. Internet engineering task force, the constrained application protocol (CoAP). https://tools.ietf.
org/html/rfc7252.
23. Sethi, P., & Sarangi, S. R. (2017). Internet of Things: Architectures, protocols, and applica-
tions. Journal of Electrical and Computer Engineering, 2017, 1–25. Article ID 9324035. View
at: Publisher Site | Google Scholar.
24. Pereira, P. P., Eliasson, J., & Delsing, J. (2014). An authentication and access control framework
for CoAP-based Internet of Things. In Proceedings of the 40th Annual Conference of the IEEE
Industrial Electronics Society. View at: Google Scholar.
25. Khattak, H., Ruta, M., & di Bari, P. (2014). CoAP-based Healthcare sensor networks: A
survey. In Proceedings of the 11th International Bhurban Conference on Applied Sciences
and Technology. View at: Google Scholar.
26. Kovatsch, M. (2013). CoAP for the Web of things: From tiny resource-constrained devices to
the web browser. In Proceedings of the 4th International Workshop on the Web of Things (WoT
2013), UbiComp’13 Adjunct, Switzerland, Zurich, September 2013. View at: Google Scholar.
27. Internet Engineering Task Force (IETF). (2012). The Constrained Application Protocol
(CoAP). https://tools.ietf.org/html/rfc7252.
28. Patients Like Me. https://www.patientslikeme.com/.
29. Sivaranjani, N., Jayabharathy, J., & Safa, M. (2019). A broad view of automation in legal predic-
tion technology. In 2019 3rd International conference on Electronics, Communication and
Aerospace Technology (ICECA). https://doi.org/10.1109/iceca.2019.8822019Omnio, http://
omnio.com/.
30. Everyday Health. http://www.everydayhealth.com/.
31. Lee, J. -V., Chuah, Y. -D., & Chieng, K. T. H. (2013). Smart elderly home monitoring system
with an android phone. International Journal of Smart Home, 7(3), 17–32. View at: Google
Scholar.
32. Zhang, Y., Liu, H., Su, X., Jiang, P., & Wei, D. (2015). Remote mobile health monitoring system
based on smart phone and browser/server structure. Journal of Healthcare Engineering, 6(2),
717–738. Article ID 590401. View at: Publisher Site | Google Scholar.
33. Pawar, P., Jones, V., van Beijnum, B. -J. F., & Hermens, H. (2012). A framework for the
comparison of mobile patient monitoring systems. Journal of Biomedical Informatics, 45(3),
544–556. View at: Publisher Site | Google Scholar.
34. Qardiocore. https://www.getqardio.com/qardiocore-wearable-ecg-ekg-monitor-iphone/.
35. Maksimović, M., Vujović, V., & Perišić, B. (2015). A custom Internet of Things health-
care system. In Proceedings of the 10th Iberian Conference on Information Systems and
Technologies (CISTI ‘15), Aveiro, Portugal, June 2015. View at: Publisher Site | Google Scholar.
36. Stecker, E. C., Reinier, K., Rusinaru, C., Uy-Evanado, A., Jui, J., & Chugh, S. S. (2017). Health
insurance expansion and incidence of out-of-hospital cardiac arrest: A Pilot study in a US
metropolitan community. Journal of the American Heart Association, 6(7). View at: Publisher
Site | Google Scholar.
37. Mancini, M. E., Cazzell, M., Kardong-Edgren, S., & Cason, C. L. (2009). Improving workplace
safety training using a self-directed CPR-AED learning program. AAOHN Journal: Official
Journal of the American Association of Occupational Health Nurses, 57(4), 159–169. View
at: Publisher Site | Google Scholar.
38. Feng-Tso, S., Cynthia, K., Heng-Tze, C. et al. (2012). Activity-aware mental stress detection
using physiological sensors. Mobile Computing, Applications, and Services, 211–230. View
at: Google Scholar.
39. Communicore. (1996). Sudden cardiac arrest: A treatable public health crisis. Retrieved 20
July 2017. http://publicsafety.tufts.edu/ems/downloads/sca_whtp.pdf.
40. Anandan, M., Manikandan, M., & Karthick, T. (2020). Advanced indoor and outdoor navigation
system for blind people using raspberry-pi. Journal of Internet Technology, 21(1), 183–195.
41. Karthick, T., & Manikandan, M. (2019). Fog assisted IoT based medical cyber system
for cardiovascular diseases affected patients. Concurrency and Computation: Practice and
Experience, 31(12), e4861. https://doi.org/10.1002/cpe.4861
Hybrid Artificial Intelligence and IoT in Health care … 147

42. Bhoi, A. K., Sherpa, K. S., & Khandelwal, B. (2018). Arrhythmia and ischemia classification
and clustering using QRS-ST-T (QT) analysis of electrocardiogram. Cluster Computing, 21(1),
1033–1044.
43. Bhoi, A. K., & Sherpa, K. S. (2016). Statistical analysis of QRS-complex to evaluate the
QR versus RS interval alteration during ischemia. Journal of Medical Imaging and Health
Informatics, 6(1), 210–214.
44. Mishra, S., Mallick, P. K., Tripathy, H. K., Bhoi, A. K., & González-Briones, A. (2020). Perfor-
mance evaluation of a proposed machine learning model for chronic disease datasets using an
integrated attribute evaluator and an improved decision tree classifier. Applied Sciences, 10(22),
8137.
45. Bhoi, A. K., Sherpa, K. S., & Khandelwal, B. (2015). Multidimensional analytical study of
heart sounds: A review. International Journal Bioautomation, 19(3), 351–376.
46. Bhoi, A. K., & Sherpa, K. S. (2014). QRS complex detection and analysis of cardiovascular
abnormalities: A review. International Journal Bioautomation, 18(3), 181–194.
47. Bhoi, A. K., Sherpa, K. S., & Khandelwal, B. (2018). Ischemia and arrhythmia classification
using time-frequency domain features of QRS complex. Procedia Computer Science, 132,
606–613.
48. Jagtap, S. (2017). Prediction and analysis of heart disease. International Journal of Innovative
Research in Computer and Communication Engineering, 5(2). View at: Google Scholar.
49. Min, C., Yujun, M., Jeungeun, S., et al. (2016). Smart clothing: Connecting human with clouds
and big data for sustainable health monitoring. Mobile Networks and Applications, 21(5),
825–845. View at: Google Scholar.
50. Chanel, G., Kierkels, J. J. M., Soleymani, M., & Pun, T. (2009). Short-term emotion assessment
in a recall paradigm. International Journal of Human-Computer Studies, 67(8), 607–627. View
at: Publisher Site | Google Scholar.
51. Petrantonakis, P. C., & Hadjileontiadis, L. J. (2010). Emotion recognition from EEG using
higher order crossings. IEEE Transactions on Information Technology in Biomedicine, 14(2),
186–197. View at: Publisher Site | Google Scholar.
52. Chen, Y., & Shen, C. (2017). Performance analysis of smartphone-sensor behavior for human
activity recognition. IEEE Access, 5, 3095–3110.
A Smart Assistive System for Visually
Impaired to Inform Acquaintance Using
Image Processing (ML) Supported by IoT

C. Aravindan, R. Arthi, R. Kishankumar, V. Gokul, and S. Giridaran

Abstract The proposed work suggests to provide a new interface support system
for visually impaired person to independently handle everyday life activities like
recognizing and greeting a friend during a walk in the street. This proposal closely
looks to solve everyday problems of a visually impaired person like identifying their
acquaintance in the road or bus, capturing important moments of the day as pic or
live, informing in case of emergency, and all such tasks can be solved by means of
intelligent system to provide assistance as per situation mentioned. Existing systems
tend to fail to utilize the visually impaired person assistive smart cane and accuracy
to support is not as expected. The data ingestion layer has been used for storing the
new as well as processed samples, whereas the data analysis layer analyses processed
data and calculates the accuracy and efficiency of the system by machine learning.
Finally, the application layer implements a mobile app that can be used to type the
name of the new person whose facial samples need to be stored in the cloud that
provides the visually impaired person with haptic feedback when an acquaintance
has been identified. The intelligence system defined in this proposal touches three
main disciplines: machine learning, imaging processing and IoT system for support,
haptic feedback.

C. Aravindan (B) · R. Arthi · R. Kishankumar · V. Gokul · S. Giridaran

Department of ECE, SRM Institute of Science and Technology, Ramapuram Campus, Chennai,
Tamil Nadu, India
e-mail: [email protected]
R. Arthi
e-mail: [email protected]
R. Kishankumar
e-mail: [email protected]
V. Gokul
e-mail: [email protected]
S. Giridaran
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 149
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_8
150 C. Aravindan et al.

Keywords Internet of Things (IoT) · Visually impaired · Image processing ·

Machine learning · Haptic smart cane

1 Introduction

In contemporary existences, significant study and struggle develop disappeared into

appreciative way to construct a prosperous typical for the visually impaired. Amid
of the swift growth of computational influences and availability of state-of-the-art
suspecting, evaluation and interpreting equipment and knowledges, computers are
fetching supplementary and have an additional intelligence. Several exploratory
assignments and profitable commodities have established the proficiency of a
computer to interrelate with people in a usual approach by contemplating at individ-
uals from side to side cameras, heeding to persons within microphones, accepting
these responses, and responding to persons in an approachable way. So, computers
can recognize face evidently; subsequently, they initiate to truthfully appreciate
person’s opinions and purposes. In a pervasive computing domain, position evidence
is extremely valuable. Numerous contemporary developing claims are centred on
position evidence. For instance, position evidence can be able to be expended for
the benefit of clients that could discover whatever they want and anywhere it is from
the existing position evidence of the customers. A global positioning framework can
be utilized to forestall lost children in a shopping centre by joining area gadgets to
them to find their present area. Additionally, a route framework is utilized to control
clients to a specific area.
Artificial intelligence is the study and creation of computer systems that can
perceive reason and function on its own without any human interference. According
to World Health Organization, there is one billion visually impaired people who have
curable vision impairment, but the people who do not have the cure as an option are
very high in numbers. Considering the improvement in recent technology, visually
impaired person can also have improved lifestyle with assistive devices. In earlier
times, visually impaired person was using white cane as a stick to detect objects
in their path and trained hard to identify things and people by their sounds. With
the advancement of computer science and electronics towards IoT, machine learning
(ML) and artificial intelligence can immensely make things easy for the people who
are visually impaired. The main goal of the proposed work helps visually impaired
users to identify the acquaintance whom they are meeting in day-to-day life with the
latest technology of IoT and ML. Though the first-generation invention this kind of
application that had their own pros and cons. The proposed work highly enlightens
the biggest strength of IoT and its accessibility anywhere, everywhere for the benefit
of visually impaired.
Even though many organizations and scientists are currently working on allevi-
ating the discomfort for visually impaired people, research in this field still remains
an open subject as there are several unresolved issues. Exploration in appearance
A Smart Assistive System for Visually Impaired … 151

identification is inspired not merely by the essential trials of an appreciation draw-

back attitudes but also by several realistic purposes where person bonding is desired.
Identifying others is a foremost task for persons with visual impairments and be able
to obstruct commitment in communal behaviours. There are a lot of ideas coming
up for the solution, but there is no proper device yet for easy and handy use of visu-
ally impaired person. ‘Smart Eye for Visually Impaired—An aid to help the visually
impaired person’ provides a solution to the problem that has been customized only
for a few locations, images and does not use videos that affect the efficiency of the
system while used in crowded places. In addition to this, some of the major problems
faced by ML is the lack of mobility and has become a barrier which stops people
from using it on a daily basis.
The IoT layer exists of seven type of layer separation, they are as follows: The
things layer, connectivity/edge computing layer, global infrastructure layer, data
ingestion layer, data analysis layer, application layer, people and process layer. The
proposed scheme contains of five, Connectivity/edge computing layer, data inges-
tion layer, data analysis layer, application layer, people and process layer, layers
which have been consumed. The purpose of connectivity/edge computing layer is
to decrease the assignment of IoT produced by cloud and that has been executed
by unloading information handling at superior strategies. The data ingestion layer
is where the transfer of data and storing occurs so that it can be used in the future.
The data analysis layer is used for optimization, control processing and increase both
user and receiver numbers. Application layer acts as an interface for the end devices
to communicate with the network. It is referred to the OSI level 5, 6 and 7, which are
application layer, presentation layer and session layer. The process layer is where
the information obtained through the IoT is used, few uses include decision making
and collaboration.
The concept of fusing IoT with machine learning can give more significant solu-
tions to solve the problems stated above. The IoT platforms such as Microsoft Azure,
Amazon AWS, Google Cloud and IBM allow to train and run the models through
the Internet. This really speeds up the process as the platform can remotely run the
model anywhere and obtain the necessary output. Anyone with a system of moderate
CPU and GPU power can acquire graphical CUDA support through the Internet to
run a task of machine learning. Looking into the concept of OS in face recognition,
this has become a key idea for identifying his acquaintance that a visually impaired
person meets on the road or in his journey.
The main advantages of utilizing cloud-based machine learning are as follows: It is
cost efficient, and it is easy to scale up so the companies can start with experimenting
a small part of projects before deploying the complete project. In addition to this for
training machine learning models requires lot of data that usually occupies a lot of
space in devices and that can be overcome through cloud.
Visually impaired people use their strong hearing ability to find the person
conversing with them, but are not always identified exactly. The real challenge here
was to train and deploy a machine that will be really helpful for visually impaired
on regular basis. The proposed scheme works on the principle of text to speech
conversion and face recognition that can be used as a practical method which helps
152 C. Aravindan et al.

in identifying people by informing the same to a visually impaired person via an

earpiece. The assistive system has a significant function in the method of communi-
cation and interprets the faces identified to the visually impaired person and creates
the situation for them to react to situations without depending on someone else.
In this paper, we have proposed to provide a new interface support system for
visually impaired person to independently handle everyday life activities like recog-
nizing and greeting a friend during a walk in the street. For the rest of this chapter,
we discuss some related work in Sect. 2 and discussed our proposed work as system
design in Sect. 3. The experimental results and discussion are explained in Sect. 4.
We conclude our work and suggest for future work in Sect. 5.

2 Related Work

A machine learning model needs the big system with a powerful processor to work
at its maximum efficiency. The major problem was that face pictures were acquired
beneath unregulated situations, such as changes in lighting, posture, facial appearance
and so on as discussed in [1]. For detecting the faces [2] in uncontrolled environment,
the system initially needs to detect sensed images. It has become the herculean task
for the visually impaired person to carry a huge advanced computer system to identify
the people they are meeting. Although mobile devices are coming up with powerful
ML systems, they are not solid enough to be used for any particular task in the long
run. Adding up to the existing problems, it has become difficult to achieve 100%
efficiency in any ML model and while using it in day-to-day basis, any small error
can cause a lot of issues.
As discussed, several attempts have been made to help visually impaired person
in day-to-day activities. In [3], the authors have developed RFID-based system for
assisting navigation in the closed environment. It provides them with the shortest
path from starting location or present location to the target place. The places that this
can be used include museums or a system for rescue in surrounding that becomes
hard to locate an emergency exit. In [4], the authors had proposed a system called
as SmartVision that helps visually impaired person to move around in an unknown
environment such as indoor and outdoor. The input for the system can be obtained
from geographic information system (GIS). In [5], the authors had proposed a system
for recognition of hand gesture to greet such as waving hand to say hi or bye. It also
includes other sign such as thumbs up, no or direction signs. They have used artificial
intelligence neural network classifier trained along with the combined features and
have been subjected to critically analyse multiple descriptors.
The person identification for the face that appears in the input data [6] has
been termed as the face recognition process. Face detection has been treated as
one of the main strides in a person identification system that distinguishes between
semi-automatic and completely programmed face recognizer. The main objective
of an automatic person identification was to investigate for person appearances in
a stationary look and, whenever located, to precisely restore their areas. Towards
A Smart Assistive System for Visually Impaired … 153

constructing the identification completely involuntary, these scheme needs to effort

exclusive of any contribution from the customer. Numerous endeavours have been
used to tackle the issue of appearance discovery endure starting with essential
methodology, and finishing with the strategy was discussed in [7].
The critical task in face recognition and detection algorithms is to rectify the errors
from the previous methods and algorithms and to provide an even more efficient
system. The data acquired for training the system must be taken from real-time
scenarios, similar to where the system is going to be used. Works have been going
on in these fields for improvisation for the past 30 to 40 years, and there has been a
lot of advancements. Post 20 years of development, a lot of scientists [8] accept that
simple two frontal images in controlled conditions where comparison has become
a problem that can be diagnosed in non-theoretical way. With very few changes in
images other than facial expression and with the recognition accuracy being more
than ninety per cent reported over a plethora of papers, the problem becomes trivial
in modern day standards without a chance of questions in the same conditions the
machine done performance far better than human performance (especially if humans
are tested on the images of the unknown persons). Although we discuss about the
development of person identification such as there is a lot of changes that occur due to
ageing, the variation in pose and other conditions that affect the image light intensity
is brought in, and human vision tends to have higher potential to identify faces when
compared to computer vision.
Exhaustive overviews of face location methods can be found in [9, 10]. This
section underlines the primary difficulties of an automated face identifier that needs
to handle and furthermore quickly depicts the person identification utilized in our
trials. Person identification schemes [11] coexisted into four typical taxonomies
such as knowledge-based, feature invariant, template matching and appearance-based
methods. As discussed in [12], the primary burden introduced by most of these prac-
tices as the time needed to distinguish all the appearances in an image. The calculation
has been all approximately determined lately as being some of the quickest and most
precise face location calculations revealed, and by the highest quality level, the other
face identification procedures are benchmarked. Hence, it is better to embrace it to
execute our face identification subsystem.
Identification of skin tone in shading pictures is a well-known, and valuable proce-
dure for face identification has been discussed in [13]. Numerous methods [14, 15]
have been reported for finding skin shading areas in the information picture.
The images captured will be in colour as modern imaging systems are all colour,
and this creates a requirement to convert from colour to grayscale. The intensities of
colours red, green and blue give an integer with which the image is characterized.
An efficient luminance, Y of each pixel, has been estimated as discussed in [16].
Feature abstraction for mutually 1D and 2D has been primarily elucidated by [17],
and the novelist approach consecutively had been executed a portion of research
subsequently. The input and output as text and audio flow in and out through the
local server created by the flask (IoT layers) have been discussed in [18]. Wiimote is
the essential game regulator for Nintendo’s Wii home computer determined support
[19]. The author has surveyed various face recognition methods with respect to face
154 C. Aravindan et al.

databases that are utilized to examine the algorithms as in [20]. The drawback of face
recognition from an enormous usual of metaphors attained around an interval such
as an assignment appearing in several investigation and substantiation purposes as
said in [21]. Several related healthcare work with IoT and machine learning-based
approaches were also taken into consideration while proposing this system [22–28].

3 System Design

Generally, humans have the tendency to easily identify and recognize faces on day-
to-day basis, and the simplified process for face recognition is as shown in Fig. 1.
Face recognition is the farthest step of face detection, and the set of captured images
has been tested on video. The video has been freezed and divided into many frames
where the single frame has been considered to be a picture. This picture has been
first tested for matching the face that was being stored in a folder under a certain
name.
The general architecture of the proposed work has been depicted in Fig. 2. The
model utilizes Raspberry Pi as the central unit of control that initiates the task to
acquire the real-time input. The system has two cameras to procure the live input.
The first camera has been turned on once the system switches on via remote control
attached in the visually impaired person assistive smart cane, and this camera stays on
until the entire system is shut down. This camera has been used for face identification,
and the image input from the camera is pushed to the cloud model. Face discovery is
one of the main strides in a person identification system that distinguishes between
semi-automatic and completely programmed person identifier. The main objective

Fig. 1 Simplified process for face recognition

Fig. 2 General architecture block diagram

A Smart Assistive System for Visually Impaired … 155

for the entire person identifier was to examine for person expressions in an immobile
copy and whenever located towards precisely restore their areas. The frame is divided
into small segments and made into a single sequence for matching purposes. Before
coming into the matching step, the dataset is stored in the form of folders which are
converted into single combination and stored as single yml file. This yml file is the
base for testing the frames obtained from the video.
The codes for detecting faces and training the recognizer file are pre-constructed
in and stored together and arranged as their order in calling the function in usage
order. The files face, face_trainer, is verified and stored. The main function to call
the other code block was written in Raspberry Pi directly. While revealed previously,
the primary response input has been taken by Raspberry Pi camera, and the captured
image is pushed to the server for further processing and recognition.
The output of face detector has a mapped image of where the faces covered in
that image. The value location has been mapped and moved to the next step of
image enhancement for further analysis of face recognition. There are numerous
ways of image enhancement techniques, grayscale to BGR, and blurring the back-
ground image to nullify the noise value in the background is important to the proposed
paperwork. Our experiments had conversion from the BGR image to grayscale image.
Heavy graphics intense hardware is required in image processing and the face recog-
nition as there is huge detail to work on. These graphics intense workload can be
handled by the platform of IoT.
There are companies that provide such service that run and supply anyone the
hardware power they need through the means of Internet. Although anyone with the
knowledge of cloud, Internet of Thing and database management can create their
own cloud server to access their need. Flask and django are the frameworks that
are used to create to web development application and develop which can create
databases using the ORM system; such databases can be accessed through Internet
web application. Once the cloud system is created, all the locally done processes were
redone on the cloud network so that it can be run automatically when a function is
called to be done. This will allow to use process only feeding the input and output
with the mobile system such as a raspberry pi connected to Internet.
To build the identification completely spontaneous, the scheme needs to work
without any contribution from the user database that has the data which is in form
script file of what the values of each face mean when arranged numerically on the
basis of the concept of considering each pixel as ones and zeros.
Once the input of that has undergone the image enhancement process, the image
was now compared with the existing script file of each person in the database. Each
individual has toned the texture of his or her own, thus leading to numerical values
that are unique. But after certain extent, there is possible of values repeating since
people might have an identical face feature in some cases. The process of identifying
and differentiating each person takes a lot of time for the machine to learning after
much number of failures and mistake that it had made in the previous attempts of
recognition.
The output response is constructed based on the input data and the data present in
the database for verification, and the database (DB) contains a list of all the names,
156 C. Aravindan et al.

images and their face values in scripted format. The classifier vector collects the
script from the database, processes to the de-serialize classifier and gets the user
label, i.e. the names. The constructed output is in form of text script format. The
Raspberry Pi takes this text file as text input for text to speech block. The output is
given out in the form of audio to visually impaired person that is the user to hear and
react. This section underlines the primary difficulties, an automated face identifier
needs to handle and furthermore quickly depict the face identification utilized in our
trials.
The secondary camera has been used to take pictures of persons that get stored
as data folders with identification tags, i.e. names of an acquaintance as the folder
names labels in the cloud. The camera has been utilized to take 20 unique pictures
of the individual in front of the webcam. These are stored away in Raspberry Pi’s
SD card for utilizing as acknowledgment in later stages. The motion face detection
process is done as the subject in video is on constant motion. The webcam bundle
has been introduced to interface the camera with Raspberry Pi to capture pictures
that utilize open-source computer vision (OpenCV). OpenCV has a library that has
programming capacities basically focused on constant PC vision. The model has
been trained remotely on a linux environment using Anaconda, Spyder Anaconda or
pycharm, etc.
Once obtained, the model initiates the training to recognize faces with its related
details such as name or stored identification. The efficiency depends on the number
of times, the data was repeatedly trained, and the cloud server provides the necessary
power needed for the processor. The script code uses a cascade classifier called
Haar transform classifier and HOG classifier supported by the libraries of OpenCv,
Dlib and face recognition. An alternative system named YOLO can also be used to
achieve high-efficiency results. The feature extraction is a procedure that depends on
photographing the appearance with a fixed size window from left-to-right and top-
to-bottom. A window of measurements h × w pixels starts checking every separated
face feature from the left top corner sub-isolating the picture into a set number of h
× w estimated blocks. For every specific area, the transformation has been applied to
obtain the distinguishing qualities that signify the observation vector for that specific
area.
The response output has been constructed from the matching process of id value
in database, and the generated output was sent back to Raspberry Pi as text file. And
the Raspberry Pi initiates the text to speech snippet which gives out the final output
as audio prompt. In case of adding a new face to the acquaintances list, the face is
first recognized as an unknown face by the primary camera module, and the process
to capture the new face by the secondary camera is initiated via Raspberry Pi.

4 Results and Discussion

As discussed in the system architecture, the model has been trained under the machine
learning model with database images of acquaintances and family members to detect
A Smart Assistive System for Visually Impaired … 157

their faces. Images are stored in the folder that comprises of facial identities of a
particular person that has the label ids of that particular person. When faces are
detected, the input response has been verified with the ‘facetrainer.yml’ file present
in the local server database.
Table 1 shows epoch numbers vs the current loss. It can be deduced from the
table that the training loss decreases from 2.055 for epoch 0 to 0.00669 for epoch
99. This shows that the training loss is inversely proportional to the epoch numbers.
As the epoch number increases the current loss decreases, as current loss decreases
the accuracy and efficiency of the model increases.
The model was trained with 100 epochs. Figure 3 shows the sample training loss,
and it is visible that at the early stages, the training loss reduces by great amount and
finally comes down to 0.0067.
The graph first shows the lessening in time with the quantity of epochs increasing.
An epoch is a term utilized in machine learning and AI that shows the quantity of
passes of the whole training dataset that AI calculation has finished. Datasets are
generally gathered into clusters (particularly, when the measure of information is
extremely huge). A few people utilize the term iteration freely and allude to getting
one clump through the model as a cycle.
In the event that clump size is the entire preparing dataset, at that point, the
epochs are the number of cycles. For physical reasons, this is typically not the situa-
tion. Numerous models are made with more than one epoch. d*e = i*b would be the

Table 1 Epoch number

Epoch numbers Current loss
versus current loss
0 2.05540347099304223
1 0.98636966643740845
2 0.81370615959167486
… …
… …
98 0.00708159850910605
99 0.00669336365535856

Fig. 3 Training loss versus epoch

158 C. Aravindan et al.

usually used relation where size of dataset is ‘d’, number of epochs is ‘e’, number
of iterations is ‘i’, and size of batch is ‘b’. Deciding the number of epochs, a model
should train which depends on numerous parameters identified with both the infor-
mation itself and the objective of the model, and keeping in mind that there have been
efforts to transform this cycle into an algorithm, regularly a profound comprehension
of the data itself is irreplaceable.
Figure 4 shows that the system has been run before any face was stored as to
improve the efficiency of the system. When adding a new face, the system first
needs to recognize the face aa an unknown face, and then, it can add the face under
the particular name so that it can identify the person during the next identification
attempt. The system detects a face and verifies it with the file, but as the face is not
previously stored, it identifies the person as unknown. This message is sent back to
the Raspberry Pi which finds that no known face is found in the nearby surroundings
and does not give any audio output.
Figure 5 shows the audio output for an unknown face not recognized, and there is
no audio output as the Raspberry Pi stops the text to speech process by not providing
any input but the user receives the prompt that the system has not recognized the
person.
In Fig. 6, the user has been presented with a choice to either add this person to
his acquaintance list by gathering his facial image clip or skip the action. This clip
is converted into multiple frames and stored in the desired folder.
The text to speech block then sends a response to the mobile application to enter the
name of the person whose image samples need to be stored under the acquaintances
list which can be seen in Fig. 7. When the input face is recognized, the label id is

Fig. 4 Unknown face recognized

A Smart Assistive System for Visually Impaired … 159

Fig. 5 Audio output for an unknown face not recognized

Fig. 6 New face added to database and identified as Gokul

constructed as an input response to the text to speech block that has been shown in
Figs. 8 and 9.
In case of a situation where there is a need to add a new face id to the existing
photo database, the user needs to initiate a voluntary action via Wii remote which
has been attached to the visually impaired person assistive smart cane Fig. 10.
The Wii remote is connected to the Raspberry Pi using Wiimote library, and
this remote can either be disconnected or put in standby mode with their respective
response button Figs. 11 and 12. The Wii Remote used in this project is gaming
160 C. Aravindan et al.

Fig. 7 Audio output for new face added to database

Fig. 8 Output for face recognition

console remote with an analogue stick for control. The Wii’s replacement comfort,
the Wii U, bolsters the Wii Remote and its peripherals in games where utilization of
the highlights of the Wii U Gamepad is not commanded.
A Smart Assistive System for Visually Impaired … 161

Fig. 9 Audio output validation for face id Kishan

Fig. 10 Wii remote connected via Bluetooth to wiimote.py

Fig. 11 Raspberry Pi connected to Wii remote using Bluetooth

162 C. Aravindan et al.

Fig. 12 Visually impaired person assistive smart cane with Wii remote control for response and
capture

The Wii mote is the operator input device used for manipulating the input names
that are to be stored in the place of labels for identifying the stored faced of acquain-
tances. The input button for typing is the buttons in the remote that can be used to
perform combination to type an alphabet using one hand Braille typing.
Bluetooth is a well-known low fuelled remote innovation that is intended for
sharing information between two gadgets over a short distance. Figuring out how to
cooperate with Bluetooth gadgets on the Raspberry Pi has a significant task with the
always expanding number of wireless gadgets.
Note: Alternative option input for storing names can be assigned either using
speech to text the library or using mobile input for to type the names, especially the
one invented by Apple iphone in ios 13, Braille type using touch screen tapping.

5 Conclusion and Future Work

A smart assistive system for the visually impaired field of study has unlimited poten-
tial, lots of unexplored areas, unresolved issues which need to be addressed, and
the problem never sees an end. This system can be enhanced by using convolu-
tional neural networks to identify any person and can be recognized as the text and
converted into speech to hear as loud for the visually impaired person that can be of
great use in multiple scenarios.
There is complete hope that this system can be transformed in the companion for
the disabled through complete voice control removing the usage of all the assistive
hardware that is supporting the machine in run. With future advancement in wireless
A Smart Assistive System for Visually Impaired … 163

communication and nanoelectronics, a camera placement in eyes with the help of

contact lenses can be paired with the TWS earbuds through mobile interfacing.
To enhance the detection process and speed it up, we can make use of TinyML in
this project.
TinyML is the latest update on machine learning that is about to change the
future of technology. Training of the machine learning model is computationally
time consuming and high in cost, so tiny ML can be used as a substitute to overcome
this. TinyML is used in places which require full-stack solutions, which are nothing
but, a mixture of hardware, system, software and applications.
The system can be designed to navigate the visually impaired person to reach
the destination that has been given as input via speech. This was recognized by
the system that has been processed to remove the noise. The destination point was
marked on Google maps coupled with the server which can provide the necessary
directions to reach the particular point. In today’s world, the crime rate is increasing
at a constant rate. But unfortunately, a visually impaired person is not considered as
an eye witness. By recording the live feed and storing it on a cloud, it is possible
to provide a solid proof. It can also be of tremendous help to the visually impaired
person in case if he gets attacked or mugged by criminals. This kind of evidence can
be used for the capturing and incarceration of the criminal.
In this assistive system for visually impaired, there is a probability of an acquain-
tance not coming inside the frame because of the varying height of the people, and
hence, face will not be recognized. This can be resolved in the future with the help
of servomotor. The camera can be attached to the servomotor and made to rotate 360
degrees continuously which increases the capture area of the camera.

References

1. Xiao, Q., & Yang, X. (2010). Facial recognition in uncontrolled conditions for information
security. EURASIP Journal on Advances in Signal Processing, 1–9.
2. Kumar, A., Kaur, A., & Kumar, M. (2019). Face detection techniques: A review. Artificial
Intelligence Review, 927–948.
3. Chumkamon, S., Tuvaphanthaphiphat, P., & Keeratiwintakorn, P. (2008). A blind navigation
system using RFID for indoor environments. pp. 765–768.
4. Faria, J., Lopes, S., Fernandes, H., Martins, P., & Barroso, J.(2010). Electronic white cane
for blind people navigation assistance. In 2010 World Automation Congress (pp. 1–7), Kobe,
Japan.
5. Jadooki, S., Mohamad, D., Saba, T., et al. (2017). Fused features mining for depth-based
hand gesture recognition to classify blind human communication. Neural Computers and
Applications, 28, 3285–3294.
6. Kaur, G., Goyal, M., Kanwal, N. (2018). Feature extraction in face recognition: A review.
UKEssays.
7. Viola, P., & Jones, M. J. (2004). Robust real-time face detection. International Journal of
Computer Vision, 57(2), 137–154.
8. Beham, M. P., & Roomi, S. M. M. (2013). A review of Face Recognition methods. International
Journal of Pattern Recognition Artificial Intelligence, 4(27), 1356005.
164 C. Aravindan et al.

9. Yang, M. H., Kriegman, D. J., & Ahuja, N. (2002). Detecting faces in images: A survey. IEEE
Transactions on Pattern Analysis and Machine Intelligence, 24(1), 34–58.
10. Iancu, C., Corcoran, P., & Costache, G. (2007). A review of face recognition techniques for
in-camera applications. In IEEE International Symposium on Signals, Circuits and Systems
Vol. 1.
11. Moghaddam, B., & Alex. P. (1997). Probabilistic visual learning for object representation.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(7), 696–710.
12. Gonzalez, R. C., Woods, R. E., & Eddins, S. L. (2004). Digital image processing using MATLAB.
Pearson Education India.
13. Kim, I., Shim, J. H., & Yang, J. (2016). Face detection. International Journal of Engineering
Research and Applications, 6, 145–150. (Stanford University).
14. Hunke, H. M. (1994). Locating and tracking of human faces with neural networks. Carnegie-
Mellon University. Department of Computer Science.
15. Rowley, H. A., Baluja, S., & Kanade, T.(1998). Neural network based face detection. IEEE
Transactions on Pattern Analysis and Machine Intelligence, 20(1), 23–38.
16. Pratt, William, K. (1991). Digital image processing (p. 634), Wiley.
17. Samaria, F., & Young, S. (1994). HMM-based architecture for face identification. Image and
Vision Computing, 12(8), 537–543.
18. Novosel, R., Meden, B., Emersic, Z., & Peer, P. (2017). Face recognition with Raspberry Pi for
IoT environments. In International Electro Technical and Computer Science Conference ERK.
19. Schreiber, M., von Wilamowitz-Moellendorff, M. & Bruder, R. (2009). New interaction
concepts by using the wii remote. In International Conference on Human-Computer Interaction
(pp. 261–270). Springer.
20. Tolba, A. S., El-Baz, A. H., & El-Harby, A. A. (2006). Face recognition: A literature review.
International Journal of Signal Processing, 2(2), 88–103.
21. Shakhnarovich, G., Fisher, J. W. & Darrell, T. (2002). Face recognition from long-term
observations. In European Conference on Computer Vision (pp. 851–865). Springer.
22. Roy, G., Bhoi, A. K., & Bhaumik, S. (2021). A comparative approach for MI-based EEG
signals classification using energy, power and entropy. IRBM.
23. Nayak, S. R., Sivakumar, S., Bhoi, A. K., Chae, G. S., & Mallick, P. K. (2021). Mixed-
mode database miner classifier: Parallel computation of graphical processing unit mining.
The International Journal of Electrical Engineering & Education, 0020720920988494.
24. Pramanik, M., Pradhan, R., Nandy, P., Bhoi, A. K., & Barsocchi, P. (2021). Machine learning
methods with decision forests for Parkinson’s detection. Applied Sciences, 11(2), 581.
25. Panigrahi, R., Pramanik, M., Chakraborty, U. K., & Bhoi, A. K. (2020). Survivability prediction
of patients suffering hepatocellular carcinoma using diverse classifier ensemble of grafted
decision tree. International Journal of Computer Applications in Technology, 64(4), 349–360.
26. Bhatt, T. V., Patel, R. K., Chitara, H. B., Marques, G., & Bhoi, A. K. (2020). Fuzzy logic
system for diabetic eye morbidity prediction. International Journal of Computer Applications
in Technology, 64(4), 339–348.
27. Marques, G., Bhoi, A. K., de Albuquerque, V. H. C., & Hareesha, K. S. (Eds.) (2021). IoT in
healthcare and ambient assisted living. Springer
28. Bhoi, A. K., Mallick, P. K., Liu, C. M., & Balas, V. E (Eds.) (2021). Bio-inspired
neurocomputing. Springer.
Internet of Things in Health Care:
A Survey

Ahmed Izzat Alsalibi, Mohd Khaled Yousef Shambour,

Muhannad A. Abu-Hashem, Mohammad Shehab, and Qusai Shambour

Abstract The Internet of Things (IoT) is a novel technology that makes smart
devices the essential building blocks in the growth of smart pervasive frameworks.
Health-related problems are considered as one of the major issues that directly affect
the quality of the life. Among other applications enabled by the IoT, healthcare tech-
nologies might be the most important one. Although several pieces of research are
developed for the IoT healthcare field, there are still considerably demanding research
and projects of exploiting the benefits of IoT in the healthcare domain. This chapter
surveys current advances in IoT-based health care focused in the aspects of privacy
and security techniques, e-health and m-health, system design and architecture, cloud,
fog, and evolutionary computing, as well as network and communication techniques.
This chapter will directly support researchers and healthcare professionals in their
understanding of developments in the domain of IoT-based healthcare technologies,
thus enabling them to spark further research in this area.

A. I. Alsalibi (B)
Department of Information Technology, Faculty of Engineering and Information Technology,
Israa University, Gaza, Palestine
e-mail: [email protected]
M. K. Y. Shambour
The Custodian of the Two Holy Mosques Institute for Hajj and Umrah Research, Umm Al-Qura
University, Mecca, Saudi Arabia
e-mail: [email protected]
M. A. Abu-Hashem
Department of Geomatics, Faculty of Architecture and Planning, King Abdulaziz University
KAU, Jeddah, Saudi Arabia
e-mail: [email protected]
M. Shehab
Department of Software Engineering, The World Islamic Science and Education University,
Amman, Jordan
e-mail: [email protected]
Q. Shambour
Department of Software Engineering, Faculty of InformationTechnology, Al-Ahliyya Amman
University, Amman, Jordan
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 165
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_9
166 A. I. Alsalibi et al.

1 Introduction

The Internet of Things (IoT) is group of smart objects that are wirelessly connected
via smart sensors. IoT will consist of billions of smart communicating “things”
and is considered as an integrated part of the future Internet. Recently, the IoT
has drawn significant research attention in which it provides proper solutions for
many modern applications such as waste management, smart cities, transportation,
emergency services, security, retails, automotive industries, agriculture, and health
care [1, 2].
Health care represents one of the most attractive applications that are enabled by
the IoT. The IoT has the prospective to enhance many healthcare applications such
as fitness programs, remote health monitoring, elderly care, and chronic diseases.
Enriching the user’s experience, decreasing costs, and enhancing the quality of
life are expected benefits by IoT-based healthcare services. IoT-based healthcare
providers are expected to be benefited by the reduction of device downtime through
remote provision, the identification of optimal times for reloading supplies for diverse
devices for their continuous operation and fairly distributed the restricted resources
over the patients. In addition, IoT-based healthcare provides many benefits compared
to conventional healthcare solutions, such as reliability, accessibility, and stability
[1, 3].
In recent years, the IoT in the healthcare domain has high broad attention from
researchers across the world to discover the potentials of utilizing the IoT in the health
care and address its challenges. Accordingly, there are currently various applications,
prototypes, and services in the field. Nevertheless, the use of IoT in the healthcare
field stills in its infancy. Thus, a comprehensive detail of existing research on the IoT-
based healthcare is very helpful for a variety of researchers who are greatly concerned
in conducting more research in this field. This chapter reviews current advances in
IoT-based healthcare research and exposes different aspects such as security and
privacy, network and communications, cloud computing and big data, e-health and
m-health, system design and architecture, evolutionary, and fog computing.
The rest of the chapter is organized as follows: Sect. 2 mainly provides a classifica-
tion of the existing research works of IoT-based healthcare technologies based on their
contributions into five categories: (1) privacy and security techniques, (2) e-health
and m-health, (3) system design and architecture, (4) cloud, fog, and evolutionary
computing, and (5) network and communication techniques. Section 3 provides clas-
sification based on optimization objective and solution schemes. Section 4 provides
extensive details about the state-of-the-art techniques as follow: In Sect. 4.1, the IoT-
based healthcare techniques in terms of their privacy and security approaches are
discussed. In Sect. 4.2, the IoT-based healthcare techniques in terms of compression
and encryption are presented. In Sect. 4.3, the methods that focus on improving e-
healthcare and m-healthcare including aging, framework, system, model, platform,
gateway, heart, and medicine are investigated. In Sect. 4.4, the methods that focus on
utilizing the huge amount of healthcare data in order to improve healthcare services
are reviewed. In Sect. 4.5, the methods that use evolutionary algorithms to improve
Internet of Things in Health Care: A Survey 167

Table 1 Classification based on privacy and security techniques

Category References
Encryption [4–21]
Hashing [22, 23]
Access control and authentication [24–31]

the efficiency of healthcare are discussed. In Sect. 4.6, the methods that use fog and
cloud computing are discussed. In Sect. 4.7, the methods that discuss the network
and communication of health care are presented. Section 4.8 provides a comparative
evaluation of data handling technique, energy consumption, combined technique,
electrocardiogram, monitoring, mobile applications, sensor technology, wearable
devices, and smart systems. Finally, summary and future research hints are given in
Sect. 5.

2 Classification and Overview

This section classifies the existing research works of IoT-based healthcare tech-
nologies based on their contributions into five categories: (1) privacy and security
techniques, (2) e-health and m-health, (3) cloud, fog, and evolutionary computing,
(4) network and communication techniques, and (5) system design and architecture.

2.1 Based on Privacy and Security Techniques

Privacy and security are the main concern issue in IoT healthcare systems. Many
techniques were proposed to overcome this issue by encrypting the user data using
different methods such as symmetric and a symmetric encryption, cryptography,
hashing, and granted privileges to an authorized user to access these data. Table
1 presents the techniques that focus on privacy and security techniques including
encryption, hashing, access control, and authentication.

2.2 Based on e-Health and m-Health

Electronic health (e-health) and mobile health (m-health) are recent healthcare tech-
niques which gained significant attention nowadays because of the wide availability
of IoT healthcare devices.
168 A. I. Alsalibi et al.

Table 2 Classification based on e-health and m-health

Category References
Framework [32–40]
Gateway [41–43]
Heart [44–48]
Medicine [49–51]
Age [52]
Medical emergency decision support systems [53–56]

Table 3 Classification based on cloud, fog, and evolutionary computing

Category References
Cloud computing [57–67]
Fog computing [68–72]
Evolutionary computing [73–75]

The massive quantity of data that are gathered by IoT platforms gained the atten-
tion of researchers to analyze and utilize these data in order to enhance the efficiency
of healthcare services.
Table 2 presents the techniques that focus on improving e-healthcare and m-
healthcare including aging, framework, system, model, platform, gateway, heart,
medicine, and age.

2.3 Based on Cloud, Fog, and Evolutionary Computing

Fog and cloud systems are essentially incorporated in IoT healthcare domain. The
data gathered from IoT platforms and devices are stored in fog and cloud systems
for analysis and prediction purpose. Then, the data and information will be accessed
by an authorized user only. Table 3 presents the techniques that focus on cloud, fog,
and evolutionary computing.

2.4 Based on Network and Communication Techniques

In this part, we present three different IoT aspects in terms of networks and commu-
nications, which are: (i) data handling techniques which focused on the commu-
nication protocols that are used and employed to transfer health data, scalability
and fault tolerance, and scheduling; (ii) energy consumption which focused on
the possibilities of improving the power consumption, decreasing overall delay, and
Internet of Things in Health Care: A Survey 169

Table 4 Classification based

Category References
on network and
communication techniques Data handling techniques
Data synchronization techniques [76, 77]
Network communication protocols [78–80]
Scalability and fault tolerance [81, 82]
Scheduling [83, 84]
Energy consumption
[85, 86, 86, 86–89]
Combined technologies
[90–96]

enhancing the overall reliability of IoT systems; (iii) combined techniques which
applied various approaches to manipulate health data measurements such as gath-
ered data, sending, storing, and analyzing stages. Table 4 presents the techniques that
focus on network and communication techniques.

2.5 Based on System Design and Architecture

This part considers evolutionary algorithms and techniques that proposed to improve
the efficiency of using IoT in healthcare industry by performing different data anal-
ysis such as prediction and optimization on the gathered data. Table 5 presents the
techniques that focus on system design and architecture including electrocardio-
gram (ECG), monitoring, mobile applications, sensor technology, wearable device,
and smart systems.

Table 5 Classification based

Category References
on system design and
architecture ECG [97–99]
Monitoring [100–105]
Mobile applications [106–113]
Sensor technology [114–122]
Wearable device [123–128]
Smart systems [129–132]
170 A. I. Alsalibi et al.

3 Classification Based on Optimization Goal

and Evaluation Platform

Table 6 classifies the works based on optimization metric, and it is clear that the
techniques proposed are guided by multiple optimization goals which need to be
carefully balanced.
The selection of tools for evaluating IoT-based healthcare techniques is very
important. While the simulators provide high flexibility to examine the designs which
may be currently not has any real implementation, it’s may be very slow to permit
high exploration of the design-space. In contrast, real systems permits more accu-
racy testing, and because of their high speed, it permits to execute a huge number of
procedures and instructions. Based on the above reasons, Table 6 also classifies the
works based on their evaluation tools and platform.

4 IoT Techniques

4.1 Access and Authentication

This section discusses the IoT-based healthcare techniques in terms of their privacy
and security approaches.
Chibelushi et al. [24] argue that the IoT identification management (IDM) issues
have been raised as an important concern when building an IoT system. Each thing
(device and user) should have a unique identity, and the IDM should have the ability
to differentiate between a user and a device, therefore, ensuring information and
identity context safety. The paper studies the expected issue behind the IDM and
introduces a new framework to examine the identity of things. The proposed frame-
work is embedded inside mobile ad hoc network and designed to serve the wireless
devices only in mobile. The proposed technique considers the limitation of network
bandwidth by exchange little information at the same time. For security purposes, a
sandboxing technique is used to protect the user’s data when device is sharing with
others.
Gope and Hwang [25] proposed a new technology that uses a body sensor network
(BSN) to protect the privacy of the user. As users may be vulnerable for privacy
violation in case, they are not considering the security requirements; the paper high-
lights the major security requirements that should be considered in BSN. Subse-
quently, a new BSN security system is proposed to fulfill such requirements. The
proposed system consists of wearable sensors (i.e., electromyography (EMG), elec-
trocardiogram (ECG), electroencephalography (EEG), and blood pressure (BP)).
The main function of the previous sensors is to gather the data and send it to a local
processing unit (LPU), which works as a coordinator. The LPU can be a smart phone,
a PDA, or any other portal device. The main function of the LPU is to work as a
router between the server and the BSN by using wireless communication medium
Table 6 Classification based on optimization metric and evaluation platform
Category References
Optimization metric
Performance [4, 90, 6, 91, 85, 24, 83, 129, 8, 32, 29–32, 108, 116, 78, 86, 13, 29, 15, 71, 19, 132, 78–80, 82–88, 87, 80, 82, 30, 88, 56, 31]
Energy consumption [81, 131, 26, 41–43, 117, 71, 41–43, 20, 113]
Internet of Things in Health Care: A Survey

Accuracy [123, 44, 92, 69, 32, 9, 32–34, 39–44, 46–48, 50–54, 47, 72, 48, 39, 40, 67, 56, 2]
Security [4, 6, 24, 8, 9, 9, 25, 26, 22, 10, 10, 11, 133, 13, 19–23, 23–26, 29–31, 31]
Hardware cost [76, 97, 119, 105]
Evaluation platform
Real system/prototype [68, 34–37, 40–43, 43, 37, 40, 2, 106, 44, 91, 92, 83, 7, 130, 81, 9, 50, 51, 94, 28, 133, 78, 81–83, 19, 79, 103, 43, 121, 53, 110–112, 47, 82, 56, 134]
Simulator [76, 90, 101, 6, 69, 22, 60, 84, 10, 102, 86, 13, 23, 71, 118, 17, 86–88, 72, 87, 80, 55, 99, 88]
Both simulator and real system [100, 102, 109]
Framework [5, 8, 12, 14–16, 18, 21, 24–27, 29–34, 36–39, 41, 43, 48, 49, 104, 114, 127, 135, 136]
171
172 A. I. Alsalibi et al.

(i.e., mobile networks GBRS/CDMA/3G). In case the LPU observe any abnormal
behavior, thereafter it will send an alert to the patient who is wearing the sensors.
Habib et al. [26] claim that the traditional authentication method that requires the
user to show their credentials only has an issue as it cannot examine the claimed ID of
the user after the login process. The new proposed authentication system is based on
wireless device radio fingerprinting and biometric modalities. The proposed system
has the ability to ensure that the monitored data belongs to the proper user through
the overall session. The proposed framework consists of three phases to increase the
accuracy by correctly validating the patient’s device, patient’s physiological char-
acteristics, and patient’s behavior on continuous basis. The patient’s smartphone
is paired with the medical sensors during an issuing procedure. Thus, the patient
used only the sensor that paired with smartphone. This pairing operation can ensure
only that the smartphone can only connect with pre-authenticated sensors. The three
authentication phases are: (1) patient to smartphone authentication, (2) smartphone
to network authentication, (3) patient to remote medical server authentication.
Hou and Yeh [27] introduced a sensor-based communication architecture. The
proposed technique is based on single sign-on (SSO). In order to ensure system
security and efficiency, one-way hash function is used to ensure the security and
the robustness of the proposed system. Furthermore, a coexistence mechanism is
proposed to insure the correctness of the coexisting medical items. A coexistence
mechanism has the ability to identify many tagged objects existing at the same time
at the same location. This proof can be conducted at the application field of the
medication management and inpatient safely.
Kai et al. [28] argue that to protect the privacy of any patient in IoT-based health-
care system, a systematic mechanism is required. For trust purposes, an interactive
vector is proposed to communicate the user devices with system brokers. The main
idea is to maintain a trustful IoT Application Market (IAM). In addition, features and
behavior of applications in the marketplace and user devices, respectively, can be
exchanged in mathematical value to set up the connection among users and market.
The proposed system works as follows: To establish any healthy IoT application to
the IAM, the application owners as service providers should have been authenticated
first from the PA. Then, the PA should provide cryptography credentials for each
actor. Through the monitoring period, the users generate feedback vector and send
it to the IAM and PA. After that, the IAM gathers the feedback vectors to calculate
and measure the next evaluation vector.
Kumar et al. [29] proposed a system ontology-based IoT. The ontology decision
model makes the critical medical operations easier for all the doctors in emergency
time. This will provide information for the doctors about diseases and help to provide
treatment; thus, the rate of deaths could be decreased. The position of doctor and
hospital will get increased due to the shared information through the database of
ontology. As a result, the details of disease such as treatment, symptoms, causes,
effects are all structured in ontology manner and saved in different databases. Data
are shared among different organizations by using PROTEGE tools [2].
Wortman et al. [30] proposed a new system based on Architecture Analysis and
Design Language (AADL) in order to make an efficient fit for the design of embedded
Internet of Things in Health Care: A Survey 173

systems. The core function of AADL is to link system architecture and design require-
ments to perform analysis at early stage in order to discover and solve error. The
proposed language is flexible; it could be extended to work as user-defined mecha-
nism for integrating customized constraints or existing AADL annexes. This flexi-
bility also provides an ability to add the random networks of wireless sensor definition
to the description of customizable properties, devices, processes, and systems.
Yeh [31] proposes a secure healthcare system based on BSN architecture. To
simultaneously grantee the efficiency of the system with confident to transmit the
data through public IoT devices. They used solid crypto-primitives to establish
two communication methods to grantee confident transmission and provide extra-
authentication between smart objects and devices. When the two channels, i.e.,
(sensors to local processing unit (LPU)) and (LPU to BSN server), are opened, no
guarantee that the transmission of data is secured. An attacker may initiate malicious
behaviors, (i.e., biodata espionage on a person). The result can be much worsen if
the connection is not secure. Furthermore, the proposed system is implemented with
Raspberry PI platform in order to express the efficiency and feasibility.

4.2 Compression and Encryption

Yang et al. [2] tried to enhance the security of the wearable devices based IoT.
The proposed system handles the key exchange using elliptic curve cryptography
(ECC). The ECC composed of three stages: setup, regeneration, verification, and
key exchange. The proposed system identifies the patient using patient’s SIM card
number along with the private key. In order to avoid the replay attack, the private key
is created by legal use without using third party.
Kumar and Gandhi [23] proposed an authentication method by using hash func-
tions, agents, session keys, and random numbers to validate devices using public key
algorithms and a combination of hash and encryption. Furthermore, they proposed
an authentication scheme to validate the IoT healthcare system’s device groups. They
have authenticated devices using five key steps of mutual authentication session.
Another hashing method is proposed in this paper, called transport layer security
(TLS) protocol, to move the records from source to destination in more secure and
reliable pattern. The proposed protocol helps the user to avoid no lost messages.
Datagram transport layer security (DTLS) is also proposed to tolerate unreliability.
The DTLS is composed of a base protocol, handshake protocol, record layer, cipher
layer, and alert protocol. The ability of an attacker to send multi-messages, which
causes a Denial-Of-Service (DOS) attack toward the server, is the main challenge for
this protocol. To overcome this concern, a smart authentication gateway to protect
the sensitive data from both the attacker and malicious is proposed.
Over and above, there are plenty of security methods proposed in the literature
to address the security issue of IoT such as [12], private key [15], public key [7],
private and public [5, 16], symmetric [13, 19, 20] symmetric and asymmetric [14,
17], cipher text [10], cryptographic [4, 6, 8, 9, 11, 21], and handshake [18].
174 A. I. Alsalibi et al.

Table 7 Comparative evaluation of different IoT security techniques

Data encryption (+) IoT can guard exposing healthy and steady improvement
in practical applications. The IoT gateway can solve sensor
communication interoperability problems as well as secure
the process of authenticating the services and patient’s data
transmission
(+) It provides a secure service in which confidentiality,
integrity, and availability are guaranteed for all users
(+) Robust in terms of preventing security breach attacks on
the databases. afford scalable and robust end-to-end security
(−) Managing private key and controlling the access
requires further investigation
(−) User security choices are limited since it does not define
specific cryptography algorithm or key generation function
Hashing (+) Confidentiality is protected by using hash functions and
channel protection during data exchanged in the
communication sessions. Anonymity is assured by using
fixed keys with random numbers and hash functions
(−) An effective and highly secure authentication methods
is required
Access control and authentication (+) Suitable to be utilized as the core protection method for
an IoT-based healthcare systems. Guaranteeing data
exchange confidentiality and affording entity verification
among smart objects, the local processing unit and the
back-end server
(−) More development is required if robust IoT applications
are to execute on diverse mobile
ad hoc network (MANET) in support of healthcare
applications

Table 7 summarizes the advantages/disadvantages of different security techniques.

4.3 E-health and M-health

Electronic health (e-health) and mobile health (m-health) are recent healthcare prac-
tices that growing rapidly due to the spread of IoT healthcare devices. The massive
amount of health data that collected through IoT healthcare devices drags the atten-
tion of researchers to utilize these data in order to improve healthcare services. This
section discusses the methods that focus on improving e-healthcare and -healthcare
including aging, framework, system, model, platform, gateway, heart, and medicine.
Internet of Things in Health Care: A Survey 175

4.3.1 Aging

New technologies contribute directly in improving population aging in the last

decades. As IoT devices become widely used, IoT drags researchers’ interest to
utilize the huge amount of information that is captured and collected using these
devices.
Konstantinidis et al. [52] presented a new idea that aims to utilize the huge amount
of data collected by IoT devices in order to assist active aging as well. The proposed
method employs IoT healthcare devices with the aid of IoT-oriented infrastructures
such as extensible messaging and presence protocol (XMPP). The proposed platform
provides a disease management tool.

4.3.2 Framework System Model and Platform

speeding up the growth of healthcare data leading to a massive amount of data that
require high-quality framework in order to manage and analyze such amount of
data to provide high-quality healthcare services. This section discusses the proposed
frameworks, models, systems, and platforms that aim to utilize IoT devices to increase
healthcare services quality.
Gelogo et al. [32] presented a mobile gateway architecture for u-healthcare as well
as provide a general idea of IoT in terms of u-healthcare applications. The proposed
framework utilizes the heterogeneous data collected from various IoT devices in order
to improve u-healthcare services. The presented work opens new path for researchers
to utilize IoT sources to solve u-healthcare challenges.
A smart healthcare kit is proposed by Gupta et al. [33] to monitor emergency
medical services. The proposed model utilizes patient’s data that were collected
from various IoT devices in order to aid emergency medical services with useful
information. The proposed design considered as a health manager in which the system
could replace the doctor by checking patient’s status and giving alarms if the patient
health is at risk.
Another healthcare model is proposed by Lee and Ouyang [34] in order to utilize
the data collected from personal health devices of IoT. The proposed model presents
collaboration protocols, which detect risky information and share them among other
devices. The proposed model aims to improve the quality of personal health feedback
by increasing and utilizing the collaboration between IoT devices.
Patra et al. [35] targeted another aspect of healthcare by proposing low-power
acquisition and digitization system. The proposed system filters and amplifies the
signals that have linearity above 12 bits. The proposed work comes as a response to
the increasing demands for low-power signal acquisition and digitization system.
As could-based web services spread widely due to its popularity for expanding
the market. Pescosolido et al. [36] proposed a cloud-based web server architecture
for healthcare services. The proposed model can improve the market in e-health and
m-health by aggregating huge amount of data collected from different IoT resources.
176 A. I. Alsalibi et al.

Furthermore, Santos et al. [37] proposed an architecture and a classification for m-

health IoT systems. The proposed architecture uses a wide variety of gateways and
health managers to distribute them on different locations to enable multiple IoT
health devices to access single service.
Building smart healthcare center coins, the main goal that inspired Singh [38] to
propose a mobile e-healthcare system. The proposed model allows doctors to address
the patients through healthcare IoT devices without being in the healthcare center
physically. The main function of the system is to utilize the capabilities of IoT in
order to overcome the shortage of healthcare centers in India’s villages.
Ullah et al. [39] presented a semantic model called k-healthcare for recording
and accessing patients’ health information. The proposed model consists of four
layers: starting with sensor layer, going through network layer, then Internet layer
and ending with services layer. The main purpose of the model is to create a platform
that simplifies accessing patient’s health information.
A disease diagnosis framework using cloud-centric IoT is proposed by Verma
and Sood [40] to build m-healthcare monitoring framework. The proposed frame-
work utilizes IoT services, especially in m-healthcare applications, in order to trace
potential disease and rank its severity level.

4.3.3 Gateway

Healthcare data are massively increasing as IoT personal healthcare devices are
growing rapidly and continuously. The growth of IoT healthcare devices increases the
heterogeneity of collected and transmitted data across the network, which demands
effective and robust healthcare methods to utilize gateways in order to control and
manage this massive amount of transmitted data. This section discusses health-
care methods that utilize gateways in order to control data transmission among IoT
devices.
Niranjana and Balamurugan [41] propose an igateway platform to improve hetero-
geneous data communication. The proposed platform presents iMedBox (intelligent
Medicine Box) and iMedPack (intelligent pharmaceutical Packaging). The proposed
platform affords uncomplicated IoT devices’ connectivity as well as in-home health-
care service. Furthermore, to afford high quality healthcare services, Rahmani et al.
[43] proposed a utilization of smart e-health gateways positions. The purpose of
the proposed method is to utilize the huge number of health monitoring systems.
In addition, Rahmani et al. [42] proposed another means of utilizing smart e-health
gateways by adapting fog computing architecture. The proposed method employs
the strategic locations of smart gateways to enhance the quality of e-health.

4.3.4 Heart

Heart diseases are life-threatening diseases. Early detection of heart diseases leads
to avoid health complications that may result in patient’s death. In this context, many
Internet of Things in Health Care: A Survey 177

researches are conducted to utilize new technology in order to detect any heart-
related anomalies. This section discusses the methods that utilize IoT healthcare
devices in order to improve heart disease detection. Ukil et al. [48] discussed and
analyzed the importance of utilizing IoT healthcare devices for diseases detection
and prevention. They highlighted the importance of anomaly detection accuracy,
especially under high noise conditions. The authors aim to raise the awareness of
the significance of IoT healthcare devices and to encourage researchers to utilize
IoT in healthcare research field. A real-time monitoring system for heart diseases is
proposed by Li et al. [45] to utilize the large number of available healthcare services.
The proposed method consists of two stages: data collection stage that is used to
build a monitoring scheme and data transmission stage. The aim of the proposed
method is to build a monitoring system that transmits patient’s health information in
real time. Azariadi et al. [44] developed an electrocardiography (ECG) system for
heartbeat diagnosis. The proposed method presents an analysis and classification for
ECG signals gained through wearable ECG devices. The analysis is done by using
discrete wavelet transform (DWT), while support vector machine (SVM) is applied
to categorize the signals. Furthermore, heart attack detector is proposed by Manisha
et al. [46] to eliminate or at least reduce heart attack hazardous effects. The proposed
system employs heart rate monitoring as well as IoT blood pressure systems in order
to improve the heart attack detection. The proposed system is aiming to help people
get rid of heart attack or at least reduce its danger by early detection of the disease.
Senthilkumar et al. [47] propose monitoring and emergency healthcare manage-
ment system for monitoring the level of heart beat and finger moisture. The proposed
system allows data transmission between patients and physician through healthcare
IoT devices remotely using GPRS/GSM. Furthermore, the transmitted data are sent
to cloud for further analysis and monitoring. The proposed system aims to reduce
the healthcare management and operational costs as well as providing high quality
services.

4.3.5 Medicine

Medicines have gained an important position in healthcare system as its direct role
in fighting diseases. Even though medicines are important, but prescribing the right
medicine considering patient’s health condition and without side effects is still a chal-
lenge. This section discusses the methods that utilize IoT healthcare devices to collect,
analyze, monitor, and manage patients’ health information in order to prescribe
medicines without jeopardizing patients’ health. Farahani et al. [49] proposed a
patient-centric e-health ecosystem architecture in order to connect doctors with
patients remotely and improve the provided healthcare services. The proposed system
contains three layers: device, fog computing, and cloud. The proposed system comes
as a response to several challenges in health care such as data management, scala-
bility, regulations, and others. Furthermore, drug interaction checker is proposed by
Jara et al. [50] to utilize IoT in pharmaceutical system by testing any harmful side
effects of drugs. The proposed system employs near field communication (NFC)
178 A. I. Alsalibi et al.

and barcode identification in order to match drugs ID with patient health records and
check its suitability. In addition, another drug interaction checker is proposed by Jara
et al. [51] in order to reduce adverse drug reaction and increase drug compliance.
The proposed system utilizes IoT devices in order to monitor medicine for any side
effects of pharmaceutical excipients.

4.4 Big Data and Cloud Computing

The spread of IoT healthcare devices contribute directly in increasing the amount
of collected healthcare data. This increment in healthcare data drags the attention of
researchers to utilize computational capabilities in order to support doctor decisions
by inferring useful information about patients’ health, which reduce the burden on
doctors’ side. This section targets the methods that focus on utilizing the large amount
of healthcare data in order to improve healthcare services.

4.4.1 Medical Emergency Decision Support Systems

Supporting doctors’ decision for emergency cases using computers becomes neces-
sity as the medical data massively grow because of the spread of medical sensors
that tied to patients. This section discusses the decision support methods that utilize
the collected data from IoT devices in order to make proper actions accordingly.
Rathore et al. [53] proposed management system for medical emergency using
Hadoop in order to utilize and manage the massive amount of heterogeneous data
that are collected through IoT devices. The proposed management system employs
enhanced network architecture to collect the data; then, an intelligent system is used
to analyze the data and make the proper actions. The proposed system is supposed to
manage collecting data from a huge number of sensors that are attached to humans’
body in order to improve the quality of healthcare services. In addition, Rathore
et al. [54] presented a real-time emergency system to utilize IoT healthcare devices
for public health using the medical sensors that are attached to human bodies. The
proposed system consists of multiple layers starting with data collection and ending
with intelligent building layer, which handles data analysis decision making. The
purpose of such systems is to utilize and manage the huge amount of heterogeneous
data collected from diverse healthcare sensors.
A semantic interoperability model is proposed by Ullah et al. [55] in order to
manage healthcare data exchange among IoT devices. The proposed model works as
decision support system as its ability to recommend medicine based on information
that collected from various and heterogeneous IoT healthcare devices. Xu et al.
[56] proposed emergency medical services method to utilize the heterogeneous data
that are collected using IoT devices. The proposed method consists of three stages,
starting with semantic data model to save and infer IoT data, then building UDA-IoT
(resource-based data accessing) method to maintain collected data, and ending with
Internet of Things in Health Care: A Survey 179

Table 8 Comparative evaluation of various e-health and m-health techniques

Framework (+) Utilizing IoT devices to increase healthcare
services quality
(−) Due to the size of targeted data, building
efficient systems still a challenge
Gateway (+) Utilize gateways capabilities in order to
control the massive amount of data
transmission among and between IoT devices
(−) Highly dependent on efficient and
customized resources such as sensors’ power
processing, memory, transmission speed, and
energy
Heart (+) Utilizing IoT healthcare devices in order to
improve heart diseases detection
(−) Building efficient detectors still a challenge
due to the large diversity of physical signs that
need to be monitored
Medicine (+) Utilizing IoT healthcare devices to collect,
analyze, monitor, and manage patients’ health
information in order to prescribe medicines
without jeopardizing patients’ health
(−) Applicability, such systems need high
accuracy and flexibility which is a challenge
Age (+) Utilizing the huge amount of information
that is captured and collected using IoT devices
to improve population aging
(−) Further investigation is needed due to lack
of research methods
Medical emergency decision support systems (+) Analyzing and managing the massive
amount of data collected using IoT devices to
enhance the quality of healthcare services
(−) Managing such amount of data requires
robust and efficient platforms

the construction of emergency medical services system to manage data collection,

integration, and interoperation.
Table 8 summarizes the advantages/disadvantages of various e-health and m-
health techniques.

4.5 Evolutionary Computing Algorithms

This part considers evolutionary algorithms to improve the efficiency of using IoT
in healthcare industry by performing different data analysis such as prediction and
optimization on the collected data.
180 A. I. Alsalibi et al.

Bhatia and Sood [73] proposed an IoT healthcare framework for ubiquitous health-
care during workout sessions. The proposed model begins by accruing data from
different attributes throughout workout sessions. The collected data are then trans-
ferred to cloud in order to analyze the real-time health situations and calculate the
vulnerability of health state using artificial neural networks (ANN) model.
Man et al. [74] developed a roadmap for assessing IoT in the context of health-
care industry during the period (2010–2020). The proposed model consists of
seven modules, including radio-frequency identification (RFID) and wireless sensor
network (WSN) devices, IoT middleware, graphic user interface (GUI), cloud
database, core management, ANN, and fuzzy logic (FL). The ANN is used to predict
the demand of healthcare assets, and FL is used to observe the balance between
early results from ANN, emergency level, stock level, and available loan with a high
rank of accuracy. In another study, Zhang et al. [75] used genetic algorithm GA to
observe the best possible data replica replacement solution for cloud storage. Authors
aimed to improve the user time accessibility, load balancing, system scalability, and
reliability of cloud storage. The proposed scheme optimizes the replica placement
strategy through creating different mathematical models according to the different
data types using GA in order to figure out the suitable data center.

4.6 Fog and Cloud Computing

Fog and cloud systems are essentially incorporated in IoT healthcare domain. The
data gathered from IoT-medical devices are allocated in fog and cloud systems for
storing, analysis, prediction, and other processing operations. The processed data are
then delivered to users who have the privilege to access it.

4.6.1 Cloud Computing

A platform is designed by Doukas and Maglogiannis [57] to manage patient records

on the cloud and utilize hardware and software resources. The proposed system
architecture collects patient data from mobile and wearable sensors and sends the
collected information and sensors status to cloud via mobile phone or special-purpose
microcontroller. The cloud hosts an application programming interface (API) to
provide a real-time data visualization about the patient and sensors status.
Garai et al. [58] proposed a crowd-sourcing telemedicine system that is integrated
with the IoT platform. The aim of this research is to establish general interoper-
ability among devices of e-health IoT, telemedicine tools and healthcare information
systems. The proposed system has the ability to capture the intercepted information,
processed, and evaluated immediately.
A cloud-centric architecture is proposed by Gupta et al. [59] to predict physical
activities of users on a regular basis. The proposed architecture used embedded
equipment sensors rather than wearable sensors to record physical activities of users
Internet of Things in Health Care: A Survey 181

while they are using treadmill in a healthcare center. The recorded data are transferred
first to local database servers and then to centric data center using Extensible Markup
Language (XML) web services. The stored data can be used to visualize results and
alerts according to the user’s health condition.
Hassan et al. [60] addressed in their study some challenges of integrating IoT and
cloud computing in healthcare domain including reliable data transmission to cloud,
dynamic resource allocation, and effective data mining tools. The proposed system
uses wearable sensors and mobile devices to gather patient data. The collected data are
sent to a portable computing device through Bluetooth mobile device via Bluetooth
and thereafter to the server on the Internet or cloud. Representational State Transfer
(REST) web services are used to send data and receiving information from cloud.
The collected data can be shared with licensed social networks and medical societies
to looking for personalized trend, growth of disease, process of rehabilitation, and
pharmacotherapy effects.
Hassanalieragh et al. [61] discussed a system architecture for IoT with cloud-based
processing in healthcare field. The proposed system consists of three main compo-
nents: data acquisition which uses a patient smart mobile device to acquire physio-
logical biomarkers measured by wearable sensors; data transmission that is used to
send gathered data to healthcare organization data center with assuring security and
privacy; and cloud processing which is responsible for storing patients information,
analytics the patients’ health condition, and visualization that makes the analytical
data readily comprehend and recognized.
Green IoT agriculture and healthcare application using sensor-cloud integration
model are introduced by Nandyala and Kim [62]. Sensor-cloud employs physical
sensors for data collection and data transfer to the cloud. The general requirements
of the proposed architecture are turning off the unneeded facilities; sending needed
data only; minimizing data path and wireless data path lengths; trading off processing
for communications; emerging advanced communication techniques; and utilizing
renewable green power sources.
Plageras et al. [63] proposed a surveillance IoT framework to widespread moni-
toring of health care. The proposed system architecture composed of three layers:
home environment which uses 6LoWPAN mesh network to connect IoT devices
including sensors edge router, cameras, and local server; the gateway; and cloud
which is responsible for storing and analyzing data and video. Doctors and caregivers
can provide advices to patients according to the analyzed data.
Sermakani [64] discusses in his research: the IoT applications in healthcare
domain; a case study for real-time remote diagnostics at the Syrian refugee camp in
Duhok; a real-world project experience on developing products for rural health care;
the IoT healthcare challenges; and the IoT roadmap and the healthcare future.
Strielkina et al. [65] discussed the possibilities of applying the queuing theory for
modeling IoT system. The authors represented the IoT system process as a Markov
stochastic procedure with discrete conditions and continuous time.
A cloud-based conceptual framework was proposed by Tyagi et al. [66]. The
research aimed to enhance the services provided to patients through improving the
communication and collaboration between patients and the healthcare actors. The
182 A. I. Alsalibi et al.

provider of cloud service offers a platform and infrastructure as services to host IoT
health applications.
Verma et al. [67] proposed an IoT cloud-centric-based smart student m-healthcare
monitoring system. The proposed system explores the diseases status of student
through predicting the level of potential disease. The authors discussed the temporal
data mining for the information gathered by medical and IoT devices and the decision
making according to various critical circumstances.
In another work, an electrocardiogram monitoring based on IoT techniques was
proposed in [2]. Patients data are gathered using wearable sensors and then send to
cloud server via WIFI. The authors used HTTP and MQTT protocols to offer visual
and timely electrocardiogram information to certain users and to store the data for
further analysis.

4.6.2 Fog Computing

Fog computing provides may advantages that can improve the system performance
including low latency, network bandwidth efficiency, mobility, location awareness,
better response time, and high service quality.
A demo system for IoT healthcare in a hybrid cloud/fog environment is proposed
in [68]. The proposed demo aimed to employ the application system of IoT healthcare
in hybrid environment, deploy applications components via cloud and fog nodes, and
manage and execute the demo with respect to orchestration techniques.
Mahmud et al. [71] studied the integration services of cloud and fog environ-
ment in interoperable health care. The proposed architecture forms a number of
clusters that contain multiple fog nodes from the same or different levels. Nodes are
assigned to perform three tasks: executing applications, hosting database, or main-
taining communication with other clusters. All communication within a cluster is
the responsibility of cluster head node. The cluster represents the cloud datacenter,
whereas fog nodes are employed as server nodes. The cluster head node executes
as a manager of resource for the cluster. Fog nodes are responsible of executing the
healthcare applications and handling database operations.
Elmisery et al. [69] proposed a method to preserve the health records privacy which
are used by a recommender healthcare system in the cloud. The authors employed fog
nodes at the client side to host a privacy middleware, which performs concealment
operation using distributed data collection protocol.
Gia et al. [70] discussed the possibilities for improving the healthcare monitoring
system. The authors employed fog nodes to support advanced services including
distributed storage, location awareness, embedded data mining, and real-time
notification service.
Sood and Mahajan [72] proposed an IoT healthcare system to identify and control
the spread of Chikungunya virus (CHV) in the affected region. Fuzzy c-means method
is employed to detect the users who are possibly infected by CHV and send urgent
notifications to users from fog layer. Moreover, social network analysis at the cloud
side is responsible to represent the spreading status of the CHV.
Internet of Things in Health Care: A Survey 183

Table 9 Comparative evaluation of different computing techniques

Cloud computing (+) Accessibility of data and services anywhere and anyplace using
internet connection
(+) Users are free from worrying about several issues including
storage resource, network communication, and computation
requirements
(−) Dependency of network connection
(−) losing the full control of data
(−) Less security and privacy concerns
Fog computing (+) Efficient transmission and processing time
(+) Cost efficient in deploying new data-centers
(+) High mobility
(−) Hard to detect a hacker attack
(−) Hard to detect the hacked file
Evolutionary computing (+) Can be used to improve system extensibility and reliability of
cloud storage
(+) Optimized resources and quick access purposes
(+) Can be employed to predict probabilistic health state
vulnerability
(−) More research investigations in this area are required
(−) The accuracy result of the evolutionary algorithms are highly
dependent on parameter setting

Table 9 summarizes the advantages/disadvantages of different computing tech-

niques.

4.7 Network and Communication

This section includes three main subsections, namely (1) data handling techniques,
(2) energy consumption which argues the possibilities of improving the consumption
rate, decreasing overall delay, and enhancing the overall reliability; the last section
named (3) combined techniques which are applied various approaches to manipu-
late health data measurements from collecting data, sending, storing, and analyzing
stages.

4.7.1 Data Handling Techniques

Data handling techniques focused on the communication protocols that are employed
to transfer health data.
This section includes three main subsections, namely, (1) network and commu-
nication protocols; (2) scalability and fault tolerance; and (3) scheduling.
184 A. I. Alsalibi et al.

Network and communication protocols

Various types of sensors used in home appliances and IoT medical devices are inte-
grated in the sensing layer in [78]. The research aimed to transmit health data in
an efficient way, with considering the limitation of the infrastructure of existing
network. The authors deployed Constrained Application Protocol (CoAP) and Hyper
Text Transfer Protocol (HTTP) at the home gateway layer to reduce bandwidth and
the amount of data generated.
In another work, Oryema et al. [79] designed and implemented a messaging system
using a constrained application protocol for IoT healthcare services. The proposed
system combines the advantages of CoAP and Message Queuing Telemetry Transport
(MQTT) protocols in constrained environments such as small message size and QoS
and supports many-to-many communications in IEEE 11,073.
Ugrenovic and Gardasevic [80] proposed an IoT healthcare system to monitor
the patient’s health condition and to access health indicators using web browser at
any time at any place. The research is focused on implementing CoAP protocol in
Mozilla Firefox web browser. The objective is to overcome the issues associated with
the classical routing protocols.
Scalability and fault tolerance
Gia et al. [81] introduced a fault tolerance technique to cover many system faults in
healthcare domain. The research aimed to overcome the bottleneck at edge routers,
and to improve network fault tolerance. The proposed monitoring system starts by
acquisition bio-signal using analog devices integrated in IPv6 medical sensor nodes
to data representation stored in the cloud server.
OneM2M-based IoT system is proposed by Woo et al. [82] for healthcare moni-
toring device. The proposed system aims to improve the gateway procedure in
converting ISO/IEEE 11,073 protocol messages from PHDs to the IoT servers
oneM2M protocol messages and vice versa. Also, the authors proposed an algo-
rithm for fault-tolerant to support the system reliability such that fault-tolerant Daisy
chaining are created at the same layer and the gateway stores a backup copy in order
to recover from faults on gateways.
Scheduling
Scheduling sensors technique for nullify interference in IoT healthcare environment
is proposed by Dhar et al. [83]. The proposed technique aims to keep the interference
away among different sensors in to guarantee the validity of health data and to ensure
the efficiency of transferring data. The healthcare parameters are passed through
shared channel with an equal priority. In case the parameter data cannot be fitted to
time slots of sensors parameters, then the data will be split to a number of parts in
order to fit between time slots.
Hindia et al. [84] proposed a dynamic scheduling approach with the ability to
serve different sensor category. The proposed approach employs android application
to collect health information using sensors. The application then sends the data over
Femto-LTE network according to the sensitivity of the patient’s condition which
Internet of Things in Health Care: A Survey 185

are classified to three main categories: (Class A: pressure of blood, heart rate, and
temperature), chronic (Class B: ECG data), and normal (Class C: glucose data).

4.7.2 Energy Consumption

In healthcare organizations, the energy consumption is about 6.5 billion each year,
that amount is increasing to meet patients’ needs [137]. Therefore, new ways should
be founded to maintain or minimize the level of consumption. This section shows
some of the relevant research on this issue.
Qiu et al. [77] focused on the instability energy sources of the body sensors, where
the nodes may fail because of insufficient energy supplies. The authors introduced
a self-recoverable time synchronization method (STSM) for health IoT. The STSM
selects the candidate nodes dynamically, and then the candidate node selects new
source node for charging. In the same time, the STSM joins the MAC layer times-
tamp the two-point least-squares method to safely enhance the precision of Pair
Broadcast Synchronization (PBS) protocol. It worth to mention that the sender-to-
receiver protocol (SRP) model and receiver-to-receiver protocol (RRP) are utilized
in STSM. Therefore, the STSM provides more accuracy than PBS with the same
amount of consumed energy [77].
In the wireless connections of the e-healthcare, a fast, reliable, and energy-efficient
fashion should exist. Thus, in [85], the authors divided the work into three parts, (i)
selecting the best level of power for every node by proposing a Power Level Deci-
sion (PLD) algorithm. Therefore, transmitting data per bit will consume minimum
energy and (ii) determining the optimal packet size by introducing a power level and
Packet Size Decision (PPD) algorithm. Thus, reducing the consumption of energy or
reducing the transmission power delay, and (iii) increasing the capability of battery
energy within nodes by designing a Global Link Decision (GLD) scheme that leads
to decreasing overall delay and enhancing the overall reliability at the same time.
The energy efficiency is very important for Wireless Body Area Network (WBAN)
which is considered as a highly convenient connection instrument for IoT healthcare
devices.
Consequently, Kim et al. [86] proposed a multi-hop WBAN structure scheme for
to minimize the total control messages number as an energy-efficient feature. The
proposed scheme contains three operations: mobility support, setup the clustered
topology, and improvement of transmission efficiency [86]. In the same context
(i.e., WBAN), Wu et al. [88] proposed a combination of a wearable sensor node,
Bluetooth low energy, and solar energy harvesting which can be implemented on
the autonomous WBAN. The multiple sensor nodes enable to measure the subject’s
body such as heartbeat and temperature distribution because of the possibility of
deploying them on various body positions. In addition, there is a tracking technique
that is used to utilize power of the sensor node [88].
A prototype for the IoT is introduced in [87], where the authors proposed a routing
protocol to choose the appropriate neighbor to forward data, which was gathered and
kept in the neighbor discovery process. While [89], review the radar sensors for
186 A. I. Alsalibi et al.

IoT, especially high-accuracy low-power low-cost indoor positioning and healthcare

monitoring sensors [89].

4.7.3 Combined Technology

Almobaideen et al. [90] developed a system designated to tourist people who have
chronic diseases to encourage them to go on a trip with a more enjoyable and safer
as well. The proposed system employed wireless sensors to collect tourist’s health
information and send them via cellular networks to a serving unit exist on the cloud.
The serving unit performs analysis according to the geographical and transportation,
health care, and tourism location databases. The system then offers the best road trip
paths that are supported by healthcare centers.
Catarinucci et al. [91] offered a system that is used ultra-low-power hybrid network
to gather ecological conditions and health information in real time through ultra-low-
power hybrid network that consists of combining 6LoWPAN nodes with UHF RFID
functionalities. The users either local or far used the web services and REST to access
data at the control center.
Chandel et al. [92] employed mobile devices for monitoring in healthcare domain.
Authors introduced an improved sensing algorithm using single inertial measurement
units (IMU) for sensing of the fundamental events, such as fall, stride length, step
count, and immobility.
Elhayatmy et al. [93] introduced a mini review for the IoT depending on the
wireless network of the body area and its architecture, routing, physical layer (PHY),
MAC layer, and security.
Jimenez and Torres [94] implemented an IoT-aware healthcare monitoring system
Using the current IoT technologies and low-cost wireless sensors, the aim is to
maximize the required for specialized personnel and at the same time minimize
the expenditure of health care. Integrating the RFID, ubiquitous technology, and
machine reasoning in the IoT healthcare domain is discussed in [95]. In addition,
authors presented how adaptive care can utilize the broadband communication over
powerline.
Thangaraj et al. [96] provided an explanation for the IoT and healthcare technology
drivers with the information from medical devices, validation of critical data, tech-
nical operation for remote devices, the network architecture, middleware, databases,
and application services.
Al-Majeed et al. [76] proposed a low cost of the analytics device, communication,
medical sensing. An IoT device is developed called “CogSense” to support real-time
care for patients. The CogSense device consists of communication, sensors and data
analytics modules. The developed device includes specific machine-learning engine
to analysis patient data and to suggest appropriate treatment. Table 10 shows the
features/weaknesses of different network and communication techniques.
Internet of Things in Health Care: A Survey 187

Table 10 Comparative evaluation of different networks and communication techniques

Data handling technique (+) Utilizing the way of data handling to improve the performance of
packet communications
(+) Reducing the volume of produced data and bandwidth
requirements
(−) Diverse IoT technologies for a small constrained device designed
for communication across sophisticated network
Energy consumption (+) An efficient energy consumption of IoT devices can avoid
premature death of any device in the network system
(+) Reducing the number of total control messages
(−) An inappropriate selection of transmission power will degrade
the network system performance
Combined technique (+) Multiple technology can be employed to facilitate the mobile
location-based services
(+) Automatic tracking and controlling of the biomedical devices for
patients
(−) Some systems during the runtime of some systems, it’s difficult to
insert the new IoT sensors

4.8 System Design and Architecture

4.8.1 Electrocardiogram

Electrocardiogram (ECG) is the procedure of tracking and save the electrical activity
of the heart by employing the electrodes placed on the specific areas in the patient’s
body. Huge data are expected to be generated from this process. Gupta et al. [97]
stored and displayed these data on a website and made it available for the authorized
personnel. The authors used Raspberry Pi that is a substantive, inexpensive, and has
much scope in future.
Satija et al. [98] proposed a new quality-aware ECG monitoring system (ECG-
SQA) that contains of transmission module, assessment module, and sensing module.
New assessment method has been designed for the lightweight ECG signal quality,
which can enable the android phone, ECG sensors, as well as the cloud server for
ECG controlling framework, while in [99] the authors extracted the amount of entropy
from heart rate to estimate the collision entropy, values of min-entropy, and Shannon
entropy using created mathematical models.

4.8.2 Monitoring

There are many cases in which patients need long-term monitoring such as the
patients with chronic diseases. Hence, the provision of constant monitoring is neces-
sary. This section shows some IoT works that concentrate on this issue. For example,
Anumala and Busetty [100] classify smart devices as health devices like a micro-oven.
They extracted a significant amount of data from the smart devices for monitoring
188 A. I. Alsalibi et al.

the health status of a patient. The working principle of the smart devices is to store
and manage the health information.
Jabbar et al. [102] employed heterogeneous IoT devices by proposing semantic
interoperability model (IoT-SIM) to monitor the current health status of the patients.
The authors used the resource description framework (RDF) that is considered
as a lightweight model for relating things through triples to make it semantically
meaningful. While Azimi et al. [101] adopted a hierarchical computing architec-
ture, namely HiCH, concerning IoT-based health monitoring systems. The proposed
system includes two main points: (i) introduces a suitable computing architecture
for hierarchical partitioning and (ii) creates a management technique that is able to
make autonomous adjustments to the patient status system.
The elderly patients need special treatment, for that both Pinto et al. [103] and Ray
[104] proposed an IoT-based healthcare system for elderly patients where Pinto et al.
[103] introduced an IoT ready solution through monitoring, supplying the techniques
to activate alarms in emergencies, and enrolling vital information of patients. As for
Ray, he proposed a Home Health Hub Internet of Things (3 IoT) framework. 3 IoT is
effective for monitoring the health status of elderly patients.
Zois [105] applied sequential decision making such as observable MDPs
(POMDPs).
and Markov decision processes (MDPs) on the real-time health monitoring and
treat it if necessary.

4.8.3 Mobile Applications

Mobile applications have paved a way for digital revolution in the healthcare industry
by enabling healthcare professionals to quickly cater to medical emergencies and
provide improved patient services.
Alex et al. [106] proposed an intelligence medicine box platform. The box is
connected with the Internet wirelessly using an android application (health IoT),
which gives the patient automatically alarm at the time of taking the medicine. On
other hands, the system sends SMS alerts to the predefined guardian in case there
are any vital signs.
In the [109], a binding protocol was suggested which authorizes the patients to
utilize a general public medical device in the same way like private medical devices
for IoT health care. This protocol deals with mobile phones, where it supplies real-
time transmission function between them through two types: (i) transmission of
streaming data (i.e., continuously transmitted), for example oxygen saturation or
heart rate; (ii) transmission of one-time data (i.e., measuring data are changed every
time), for example weight and blood pressures, and Gelogo et al. [107] discussed
the backdrop of IoT, especially its applications in the u-healthcare area. The authors
planned to make the mobile as an integrated gateway to enable the heterogeneous
devices for u-healthcare convergence.
Utilization of the smartphone fundoscopy to capture super-resolution images is
proposed by Jebadurai and Peter [108]. They enhance the resolution of the captured
Internet of Things in Health Care: A Survey 189

images by using multi-kernel support vector regression (SVR). The proposed hybrid
architecture helps to improve the diagnosis of ophthalmologists by providing them
with high-resolution retinal images, while Saxena et al. [112] developed an IoT
architecture for smart mobile health care using named data networking (NDN). The
proposed architecture solves the problems of the healthcare IoT exchanged data over
the IP-centric Internet.
Lee [111] proposed a healthcare platform IoT-based mobile application through
designing and processing a mobile device protocol, which allows the sharing of a
medical device between multiple users. This protocol divided into: (i) registration
the patients in the medical device; (ii) a protocol used to register a medical device in
the platform. The same author proposed a new service model for smartphones and
wearable devices that are based on distributed processing and connection protocols
between the different devices. This service allows to recognize the user’s function
and processes the measured data such as blood pressure to show it on a healthcare
service platform. In addition, the users can organize the service configurations in their
mobile, handle services, and send the information for the participants in IoT-based
healthcare service platform.
Talpur et al. [113] proposed a system for the chronic cardiac healthcare moni-
toring. The capabilities of the proposed system are: (i) data acquisition of the body
sensors; (ii) analyzing health situation by a smartphone; and (iii) transporting the
bodily data to smartphones by sensors and a wireless body area network through
Bluetooth technology and ZigBee. After that, the bodily data are stored in a secure
database in the external memory of the smartphone. Therefore, the proposed system
allows to all smartphone devices to be utilized in healthcare monitoring.

4.8.4 Sensor Technology

The sensors are small devices that are used to discover data (heat, moisture, light,
motion, etc.) from the physical environment and change it to readable data using an
instrument or an observer.
Neagu et al. [120] focused on enhancing the effectiveness of the used sensor data
in the sensing service. Thus, they proposed a health monitoring-oriented sensing
service scenario (HM-SS) to improve the service quality and accessibility. HM-SS
method highlighted the features of cloud-IoT, for example the easy of the data sharing
and accessibility, scalable data storage and processing resources, disaster recovery
and advanced data governance, whereas the main objective in [117] is to minimize
the consuming power of loT sensor nodes through reducing the designing a digital
block by the hardware architecture system. The design has been executed on FPGA
device to evaluate the functional then using ASIC to implement the silicon.
Miranda et al. [119] studied the common recognition and identification platform
(CRIP) from difference sides, such as the planning, evolution, implementation, as
well as verification of the CRIP. It is worth to mention that the CRIP provides sensor-
based support to determine the patients and health devices. In addition, different tech-
nologies have been used to design and implement the CRIP, which permits seamless
190 A. I. Alsalibi et al.

interaction and combining of users and sensors, for instance Bluetooth for connection
with health devices and face recognition for authentication.
Rasid et al. [121] proposed a global healthcare system characterized in intelligent
and automated monitoring. The contents of the proposed system: (i) Internet service
to provide communication and (ii) IP addresses to enable patients to connect and view
their medical information using portable devices such as mobiles. The 6LoWPAN
sensor nodes have been used for communications as a short range. Thus, it provides
the possibility to introduce healthcare monitoring systems for patient-centric.
The efficiency of sensor data management using IoT has been presented in [114].
To achieve the purpose of this work, the authors used a generic system-level frame-
work, which helps self-serve health monitoring system. The work summarized in
three points: starting in transport the data of sensors into a remote server and then
analyzing these data remotely using a distributed manner. Finally, presenting the
patient status in real time.
Jeong et al. [115] introduced a new method for the execution of IoT health care,
which includes continuity, capability, privacy, reliability, and stability. In gathering
data, it is necessary to have the sensors in the cloud computing environments for
operating and managing the data properly. Thus, the manipulating data are done by
the sensors remotely distributed.
Karamitsios and Orphanoudakis [116] focused on current stacks which was
suggested for personal health devices (PHDs), by utilizing the IEEE 11,073 (main
protocol stack) criterion healthcare implementations and healthcare data collecting.
In this work, the sensor data are collected from patients and healthcare service
providers.
Recently, the home caring services are very important, especially with increasing
the proportion of elderly people who live alone. In addition, it has been noted that
many patients (especially elderly patients) do not remember the time of taking their
medicine. Accordingly, Sohn et al. [122] proposed a self-alarm system for medica-
tion. The system tracks the status of medicine bottles weight. The huge amount of
data generated by sensor devices has been stored in the cloud such as a smart home,
road traffic management, and continuous health monitoring of individuals.
In according to that, Manogaran et al. [118] proposed a new method for IoT to save
and manipulate scalable a big data of the healthcare sensors. The proposed method
includes grouping and choosing (GC) and MetaFog-Redirection (MF-R) for solving
the problem of the huge amount of data from the sensor devices.

4.8.5 Wearable Device

The usage of wearable healthcare devices through the IoT is the best way for long-
term medical care because it allows patients to get their health information. In addi-
tion, for the patients’ health, it is necessary to monitor the environmental criterions
which affect the health. Additionally, it is ease of use for elderly patients. Therefore,
the wearable devices have become the focus of attention of many researchers. Basanta
et al. [124] introduced a new method to improve the goodness of services for the
Internet of Things in Health Care: A Survey 191

elderly healthcare system and integrating various technologies of wearable devices.

The authors collected the data in real time through different wearable devices and
kept in the central database. This data connect patients and doctors at the time of an
emergency. In the [126], the authors focus on the challenge of identifying in real time
the owner of data. This challenge is very important due to the fact that healthcare
facilities support a large number of users who possess various devices, thus leading
to complexities in the streaming process, in other words, forming an N × M data
source heterogeneity.
To address this matter (i.e., the complexities of mapping), the authors developed a
Petri Nets service model to help in creating a transparent path to track outgoing data
from wearable devices, while Lian [125] focuses on the challenges in designing low
power chips, in particular, in designing self-powered wearable wireless sensors that
can be used in healthcare IoT. The author introduced a new system architecture in
continuous-in-time and discrete-in-amplitude (CTDA) domain. The main objective
is to improve power efficiency by reducing the number of data points for trans-
mission with adapting the sampling rate. The objectives of monitoring the patient’s
body with high efficiency and studying the patients risk factors on one hand of
foods and drugs were addressed by Ali et al. [123]. The authors introduced a type-2
fuzzy ontology-aided recommender system by using wearable sensors to locate the
patients’ health condition and then suggested diabetes-specific prescriptions. Mora
et al. [127] proposed an IoT-based distributed framework for controlling biomedical
signals of human when conducting physical activities, such as heart rates of foot-
ballers throughout a football match. The authors used the wearable devices to build
the health application. The proposed framework is available to other mobile envi-
ronments (i.e., extensive data attainment and high processing needs take place). The
high percentage of elders cannot follow the traditional health care. Therefore, Salunke
and Nerkar [128] proposed an IoT-driven healthcare system to manage and record,
for long time, the patient’s physiological information by using wearable sensors.
The proposed system follows, captures, and describes the health trends recorded
from contextual and physiological sensors in order to identify and prevent health
episodes.

4.8.6 Smart Systems

This section introduces the smart healthcare systems utilizing IoT. Starting with
[132], the authors increased the efficiency of the healthcare data communication
in machine-to-machine environments. Granados et al. [131] benefited from merging
data and power into one cable called power over Ethernet that authorizes IoT gateway.
In other words, it transports both energy and cloud connectivity to smart hospital
appliances and medical sensors (smart gateway). The benefit of this method is to
relive the processing overhead and increase the overall system energy performance
and efficiency. IoT-based smart rehabilitation system is introduced by Fan et al.
[130] by proposing an ontology-based automation design methodology (ADM). The
system helps the understanding of the medical resources and symptoms by computers.
192 A. I. Alsalibi et al.

Table 11 Comparative assessment of different system design and architecture techniques

Data handling technique (+) Utilizing the way of data handling to improve the performance of
packet communications
(+) Reducing the volume of produced data and bandwidth
requirements
(−) Diverse IoT technologies for a small constrained device designed
for communication across sophisticated network
Energy consumption (+) An efficient energy consumption of IoT devices can avoid
premature death of any device in the network system
(+) Reducing the number of total control messages
(−) An inappropriate selection of transmission power will degrade
the network system performance
Combined technique (+) Multiple technology can be employed to facilitate the mobile
location-based services
(+) Automatic tracking and controlling of the biomedical devices for
patients
(−) Some systems during the runtime of some systems, it is difficult
to insert the new IoT sensors

In addition, it presents an efficient platform to connect all the resources and

provides instant information interaction. Elanthiraiyan and Babu [129] present a
physical health system and smart medicine using IoT. The system has two advan-
tages: (i) easy allotment of domain knowledge and (ii) rapid creation of restora-
tion system. In other words, the system is outstanding and impressive, as ontology
and IoT have an important role in the method. Table 11 reviews the advan-
tages/disadvantages of diverse system design and architecture techniques. Table
11 reviews the advantages/disadvantages of diverse system design and architecture
techniques.

5 Conclusion and Future Outlook

The IoT has enabled communications with and amid smart objects, hence leading to
the vision of anywhere, anytime, anymedia, anything communications. Healthcare
technologies represent one of the most attractive applications that are enabled by the
IoT. Accordingly, several researchers across the world have recently explored a range
of solutions that would improve the provision of healthcare services by exploiting the
potentials of the IoT. This chapter provides an informative survey of several aspects
of IoT healthcare technologies and classifies them on a number of axes to show up
their similarities and differences. In overall, the outcomes of this survey are likely to
be valuable for researchers, policymakers, and healthcare professionals working in
the fields of IoT and healthcare technologies.
To conclude, a brief summary of future research directions is presented as follows:
Driven Connected Healthcare Henceforth, privacy and security features for network,
Internet of Things in Health Care: A Survey 193

users, data, and applications should be addressed using novel technologies and stan-
dards. As for network security, the IPv6 (Internet Protocol Version 6) is the upcoming
generation Internet protocol that includes security control and addressing informa-
tion to route packets through the Internet. In general, the main issues that need to be
addressed include device capabilities, security, closing the gap between individuals,
sensors production, and safety.
Big data: Future proposed models of big data should focus in providing syntactic
interoperability along with various IoT devices. One of the most serious challenges
of the syntactic interoperability among heterogeneous IoT devices is the security
aspect that should be carefully considered in the in the near future.
Cloud-Centric IoT-based Disease Diagnosis (COND) Healthcare Framework:
A novel COND can be integrated to the previous proposed diagnosis systems to
provide more adaptability. Frequency-COND, Pattern-COND, and Scale-COND can
be enhanced henceforth for effective assessment. The proposed platform can be
deployed in the physical world in order to develop statistical measurements. Further-
more, the proposed system can be evaluated in terms of accuracy with benchmark
evaluation methods that are used by medical professionals.
Focusing on improving quality of pervasive healthcare services like early warning
support to patients that trigger the healthcare service based on patients’ physical status
rather than their feelings. The proposed IoT-based disease monitoring systems could
be benefited from the integration of the data stream management system (DSMS)
technologies to enhance its functions. In addition, context awareness and data stream
mining technologies can also be taken into account to provide more robust pervasive
healthcare services.

References

1. Islam, S. R., Kwak, D., Kabir, M. H., Hossain, M., & Kwak, K. S. (2015). Theinternet of
things for health care: A comprehensive survey. IEEE Access, 3, 678–708.
2. Yang, Z., Zhou, Q., Lei, L., Zheng, K., & Xiang, W. (2016). An IoT-cloud based wearable
ecg monitoring system for smart healthcare. Journal of Medical Systems, 40(12), 286.
3. Pang, Z. (2013). Technologies and architectures of the Internet-of-Things (IoT) for health and
well-being (PhD thesis). KTH Royal Institute of Technology.
4. AlAssaf, N., AlKazemi, B., & Gutub, A. (2003).Applicable light-weight cryptography to
secure medical data in IoT systems. Arabia.
5. Balamurugan, S., Madhukanth, R., Prabhakaran, V. M., & Shanker, R. G. K. (2016).Internet
of health: Applying IoT and big data to manage healthcare systems. International Research
Journal of Engineering and Technology, 310, 732–735.
6. Prasad, V., & Arun Raj Kumar P. (2017). Secretsharing scheme based on cauchy matrix for data
outsourcing. In Proceedings of the 10th International Conference on Security of Information
and Networks (pp. 315–318). ACM.
7. Charalampos Doukas, Ilias Maglogiannis, Vassiliki Koufi, Flora Malamateniou, and George
Vassilacopoulos. Enabling data protection through pki encryption in IoT m-health devices.
In 2012 IEEE 12th International Conference on Bioinformatics & Bioengineering (BIBE)
(pp. 25–29). IEEE.
194 A. I. Alsalibi et al.

8. Elhoseny, M., Ramírez-González, G., Abu-Elnasr, O. M., Shawkat, S. A., Arunkumar, N., &
Farouk, A. (2018). Secure medical data transmission model for IoT-based healthcare systems.
IEEE Access, 6, 20596–20608.
9. Gong, T., Huang, H., Li, P., Zhang, K., & Jiang, H. (2015).A medical healthcare system
for privacy protection based on IoT. In 2015 Seventh International Symposium on Parallel
Architectures, Algorithms and Programming (PAAP) (pp. 217–222). IEEE.
10. Hu, J. X., Chen, C. L., Fan, C. L., & Wang, K. H. (2017).An intelligent and secure health
monitoring scheme using IoT sensor based on cloud computing. Journal of Sensors, 2017.
11. Iqbal, M. A., & Bayoumi, M. (2016).Secure end-to-end key establishment protocol for
resource-constrained healthcare sensors in the context of IoT. In 2016 International Confer-
ence on High Performance Computing & Simulation (HPCS) (pp. 523–530). IEEE.
12. Jung Tae Kim. (2014). Privacy and security issues for healthcare system with embedded rfid
system on internet of things. Advanced Science and Technology Letters, 72, 109–112.
13. Er Kritika, Singh, H. P., Er Singh, N. P., Er Mamta. (2016). Multivariate authentication
and encryption scheme for data privacy in IoT healthcare monitoring. Imperial Journal of
Interdisciplinary Research, 2(8).
14. Lee, J. D., Yoon, T. S., Chung, S. H., & Cha, H. S. (2015). Service-oriented security framework
for remote medical services in the internet of things environment. Healthcare informatics
research, 21(4), 271–282.
15. Luo, E., Bhuiyan, M. Z. A., Wang, G., Rahman, M. A., Wu, J., & Atiquzzaman, M.
(2018).Privacyprotector: Privacy-protected patient data collection in IoT-based healthcare
systems. IEEE Communications Magazine, 56(2), 163–168.
16. Manogaran, G., Varatharajan, R., Lopez, D., Kumar, P. M., Sundarasekar, R., & Thota, C.
(2018). A new architecture of internet of things and big data ecosystem for secured smart
healthcare monitoring and alerting system. Future Generation Computer Systems, 82, 375–
387.
17. Mathur, A., Newe, T., & Rao, M. (2016). Defence against black hole and selective forwarding
attacks for medical wsns in the IoT. Sensors, 16(1), 118.
18. Moosavi, S. R., Gia, T. N., Nigussie, E., Rahmani, A. M., Virtanen, S., Tenhunen, H., & Isoaho,
J. (2015). Sessionresumption-based end-to-end security for healthcare internet-of-things. In
2015 IEEE International Conference on Computer and Information Technology; Ubiquitous
Computing and Communications; Dependable, Autonomic and Secure Computing; Pervasive
Intelligence and Computing (CIT/IUCC/DASC/PICOM) (pp. 581–588). IEEE.
19. Moosavi, S. R., Gia, T. N., Rahmani, A. M., Nigussie, E., Virtanen, S., Isoaho, J., & Tenhunen,
H. (2015).Sea: a secure and efficient authentication and authorization architecture for IoT-
based healthcare using smart gateways. Procedia Computer Science, 52, 452–459.
20. Rghioui, A., L’aarje, A., Elouaai, F., & Bouhorma, M. (2014). The internet of things for health-
care monitoring: security review and proposed solution. In 2014 Third IEEE International
Colloquium in Information Science and Technology (CIST) (pp. 384–389). IEEE.
21. Sharma, S., Chen, K., & Sheth, A. (2018). Toward practical privacy-preserving analytics for
IoT and cloud-based healthcare systems. IEEE Internet Computing, 22(2), 42–51.
22. Han, K.-H., & Bae, W.-S. (2016). Proposing and verifying a security-enhanced protocol for
IoT-based communication for medical devices. Cluster Computing, 19(4), 2335–2341.
23. Kumar, P. M., & Gandhi, U. D. (2020).Enhanced dtls with coapbased authentication scheme
for the internet of things in healthcare application. The Journal of Supercomputing, 1–21.
24. Chibelushi, C., Eardley, A., & Arabo, A. (2013). Identity management in the internet of
things: The role of manets for healthcare applications. Computer Science and Information
Technology, 1(2), 73–81.
25. Gope, P., & Hwang, T. (2016). Bsn-care: A secure IoT-based modern healthcare system using
body sensor network. IEEE Sensors Journal, 16(5), 1368–1376.
26. Habib, K., Torjusen, A., & Leister, W. (2014).A novel authentication framework based on
bio-metric and radio fingerprinting for the IoT in ehealth. In Proceedings of International
Conference on Smart Systems, Devices and Technologies (SMART) (pp. 32–37).
Internet of Things in Health Care: A Survey 195

27. Hou, J.-L., & Yeh, K.-H. (2015). Novel authentication schemes for IoT based healthcare
systems. International Journal of Distributed Sensor Networks, 11(11), 183659.
28. Kang, K., Pang, Z., & Wang, C. (2013). Security and privacy mechanism for health internet
of things. The Journal of China Universities of Posts and Telecommunications, 20, 64–68.
29. Kumar, V., et al. (2015). Ontology based public healthcare system in internet of things (IoT).
Procedia Computer Science, 50, 99–102.
30. Wortman, P. A., Tehranipoor, F., Karimian, N., & Chandy, J. A. (2017). Proposing a modeling
framework for minimizing security vulnerabilities in IoT systems in the healthcare domain.
In 2017 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI)
(pp 185–188). IEEE.
31. Yeh, K.-H. (2016). A secure IoT-based healthcare system with body sensor networks. IEEE
Access, 4, 10288–10299.
32. Gelogo, Y. E., Hwang, H. J., & Kim, H. K. (2015).Internet of things (IoT) framework for
u-healthcare system. International Journal of Smart Home, 9(11), 323–330.
33. Gupta, P., Agrawal, D., Chhabra, J., & Dhir, P. K. (2016).IoT based smart healthcare
kit. In 2016 International Conference on Computational Techniques in Information and
Communication Technologies (ICCTICT) (pp. 237–242). IEEE.
34. Lee, B. M., & Ouyang, J. (2014).Intelligent healthcare service by using collaborations between
IoT personal health devices. Blood Pressure, 10(11).
35. Patra, P., Yadav, K., Naik, J., & Dutta, A. (2015). A true 1 V 1 µW biomedical front end with
reconfigurable ADC for self powered smarter IoT healthcare systems. In TENCON 2015–2015
IEEE Region 10 Conference (pp. 1–4). IEEE.
36. Pescosolido, L., Berta, R., Scalise, L., Revel, G. M., De Gloria, A., & Orlandi, G. (2016). An
IoT-inspired cloud-based web service architecture for e-health applications. In Smart Cities
Conference (ISC2), 2016 IEEE International (pp. 1–4). IEEE.
37. Santos, D. F., Perkusich, A., & Almeida, H. O. (2014). Standard-based and distributed health
information sharing for mhealth IoT systems. In 2014 IEEE 16th International Conference
on e-Health Networking, Applications and Services (Healthcom) (pp. 94–98). IEEE.
38. Singh, R. (2016). A proposal for mobile e-care health service system using IoT for Indian
scenario. Journal of Network Communications and Emerging Technologies (JNCET), 6(1).
39. Ullah, K., Shah, M. A., & Zhang, S. (2016). Effective ways to use internet of things in the field
of medical and smart health care. In 2016 International Conference on Intelligent Systems
Engineering (ICISE) (pp. 372–379). IEEE.
40. Verma, P., & Sood, S. K. (2018). Cloud-centric IoT based disease diagnosis healthcare
framework. Journal of Parallel and Distributed Computing, 116, 27–38.
41. Niranjana, S., & Balamurugan, A. (2015). Intelligent e-health gateway based ubiquitous
healthcare systems in internet of things. International Journal of Scientific Engineering and
Applied Science (IJSEAS), 1(9).
42. Rahmani, A. M., Gia, T. N., Negash, B., Anzanpour, A., Azimi, I., Jiang, M., & Liljeberg,
P. (2018). Exploiting smart e-health gateways at the edge of healthcare internet-of-things: A
fog computing approach. Future Generation Computer Systems, 78, 641–658.
43. Rahmani, A. M., Thanigaivelan, N. K., Gia, T. N., Granados, J., Negash, B., Liljeberg,
P., & Tenhunen, H. (2015). Smart e-health gateway: Bringing intelligence to Internet-of-
Things based ubiquitous healthcare systems. In 2015 12th Annual IEEE on Consumer
Communications and Networking Conference (CCNC) (pp. 826–834). IEEE.
44. Azariadi, D., Tsoutsouras, V., Xydis, S., & Soudris, D. (2016).Ecg signal analysis and
arrhythmia detection on IoT wearable medical devices. In 2016 5th International Conference
on Modern Circuits and Systems Technologies (MOCAST) (pp. 1–4). IEEE.
45. Li, C., Xiangpei, Hu., & Zhang, L. (2017). The IoT-based heart disease monitoring system
for pervasive healthcare service. Procedia Computer Science, 112, 2328–2334.
46. Manisha, M., Neeraja, K., Sindhura, V., & Ramaya, P. (2016).IoT on heart attack detection
and heart rate monitoring. International Journal of Innovation in Engineering and Technology
(IJIET).
196 A. I. Alsalibi et al.

47. Senthilkumar, R., Ponmagal, R. S., & Sujatha, K. (2016). Efficient health care monitoring
and emergency management system using IoT. International Journal of Control Theory and
Applications, 9(4), 137–145.
48. Ukil, A., Bandyoapdhyay, S., Puri, C., & Pal, A. (2016). IoT healthcare analytics: The
importance of anomaly detection. In 2016 IEEE 30th International Conference on Advanced
Information Networking and Applications (AINA) (pp. 994–997). IEEE.
49. Farahani, B., Firouzi, F., Chang, V., Badaroglu, M., Constant, N., & Mankodiya, K. (2018).
Towards fog-driven IoT ehealth: Promises and challenges of IoT in medicine and healthcare.
Future Generation Computer Systems, 78, 659–676.
50. Jara, A. J., Alcolea, A. F., Zamora, M. A., Skarmeta, A. G., & Alsaedy, M. (2010).Drugs
interaction checker based on IoT. In Internet of Things (IOT), 2010 (pp. 1–8). IEEE.
51. Jara, A. J., Zamora, M. A., & Skarmeta, A. F. (2014).Drug identification and interaction
checker based on IoT to minimize adverse drug reactions and improve drug compliance.
Personal and ubiquitous computing, 18(1), 5–17.
52. Konstantinidis, E. I., Bamparopoulos, G., Billis, A., & Bamidis, P. D. (2015). Internetof things
for an age-friendly healthcare. In MIE (pp. 587–591).
53. Rathore, M. M., Ahmad, A., & Paul, A. (2015). The internet of things based medical emergency
management using hadoop ecosystem. In Sensors, 2015 IEEE (pp. 1–4). IEEE.
54. Rathore, M. M., Ahmad, A., Paul, A., Wan, J., & Zhang, D. (2016). Real-time medical
emergency response system: exploiting IoT and big data for public health. Journal of medical
systems, 40(12), 283.
55. Ullah, F., Habib, M. A., Farhan, M., Khalid, S., Durrani, M. Y., & Jabbar, S. (2017). Semantic
interoperability for bigdata in heterogeneous IoT infrastructure for healthcare. Sustainable
Cities and Society, 34, 90–96.
56. Boyi, Xu., Da Li, Xu., Cai, H., Xie, C., Jingyuan, Hu., Fenglin, Bu., et al. (2014). Ubiquitous
data accessing method in IoT-based information system for emergency medical services. IEEE
Transaction Industrial Informatics, 10(2), 1578–1586.
57. Charalampos Doukas and Ilias Maglogiannis. Bringing IoT and cloud computing towards
pervasive healthcare. In 2012 6th International Conference on Innovative Mobile and Internet
Services in Ubiquitous Computing (IMIS) (pp. 922–926). IEEE.
58. Garai, Á., Attila, A., & Péntek, I. (2016). Cognitivetelemedicine IoT technology for dynam-
ically adaptive e-health content management reference framework embedded in cloud archi-
tecture. In 2016 7th IEEE International Conference on Cognitive Infocommunications
(CogInfoCom) (pp. 000187–000192). IEEE.
59. Gupta, P. K., Maharaj, B. T., & Malekian, R. (2017).A novel and secure IoT based cloud
centric architecture to perform predictive analysis of users activities in sustainable health
centres. Multimedia Tools and Applications, 76(18), 18489–18512.
60. Hassan, M. M., Albakr, H. S., & Al-Dossari, H. (2014). Acloud-assisted internet of things
framework for pervasive healthcare in smart city environment. In Proceedings of the 1st
International Workshop on Emerging Multimedia Applications and Services for Smart Cities
(pp. 9–13). ACM.
61. Hassanalieragh, M., Page, A., Soyata, T., Sharma, G., Aktas, M., Mateos, G.,Kantarci,
B., & Andreescu, S. (2015). Health monitoring and management using internet-of-things
(IoT) sensing with cloud-based processing: Opportunities and challenges. In 2015 IEEE
International Conference on Services Computing (SCC) (pp. 285–292). IEEE.
62. Nandyala, C. S., & Kim, H. K. (2016). Green iot agriculture and healthcare application (gaha).
Int J Smart Home, 10(4), 289–300.
63. Plageras, A. P., Psannis, K. E., Ishibashi, Y., & Kim, B. G. (2016). IoT-based surveillance
system for ubiquitous healthcare. In IECON 2016–42nd Annual Conference of the IEEE on
Industrial Electronics Society (pp. 6226–6230). IEEE.
64. Sermakani, V. (2014). Transforming healthcare through internet of things. In Project
Management Practitioners’ Conference.
65. Strielkina, A., Uzun, D., & Kharchenko, V. (2017). Modelling of healthcare IoT using the
queueing theory. In 2017 9th IEEE International Conference on Intelligent Data Acquisition
Internet of Things in Health Care: A Survey 197

and Advanced Computing Systems: Technology and Applications (IDAACS) (Vol. 2, pp. 849–
852). IEEE.
66. Tyagi, S., Agarwal, A., & Maheshwari, P. (2016). A conceptual framework for IoT-based
healthcare system using cloud computing. In 2016 6th International Conference on Cloud
System and Big Data Engineering (Confluence) (pp. 503–507). IEEE.
67. Verma, P., Sood, S. K., & Kalra, S. (2017). Cloud-centric IoT based student healthcare
monitoring frame work. Journal of Ambient Intelligence and Humanized Computing, 1–17.
68. Bibani, O., Mouradian, C., Yangui, S., Glitho, R. H., Gaaloul, W., Hadj-Alouane, N.
B.,Morrow, M., & Polakos, P. (2016). A demo of IoT healthcare application provisioning in
hybrid cloud/fog environment. In 2016 IEEE International Conference on Cloud Computing
Technology and Science (CloudCom) (pp. 472–475). IEEE.
69. Elmisery, A. M., Rho, S., & Botvich, D. (2016). A fog based middleware for automated
compliance with oecd privacy principles in internet of healthcare things. IEEE Access, 4,
8418–8441.
70. Tuan Nguyen Gia, Mingzhe Jiang, Amir-Mohammad Rahmani, Tomi Westerlund, Pasi Lilje-
berg, and Hannu Tenhunen. Fog computing in healthcare internet of things: A case study
on ecg feature extraction. In 2015 IEEE International Conference on Computer and Infor-
mation Technology; Ubiquitous Computing and Communications; Dependable, Autonomic
and Secure Computing; Pervasive Intelligence and Computing (CIT/IUCC/DASC/PICOM)
(pp. 356–363). IEEE.
71. Mahmud, R., Koch, F. L., & Buyya, R. (2018). Cloud-fog interoperabilityin IoT-enabled
healthcare solutions. In Proceedings of the 19th International Conference on Distributed
Computing and Networking (pp. 32). ACM.
72. Sood, S. K., & Mahajan, I. (2017). Wearable IoT sensor based healthcare system for identifying
and controlling Chikungunya virus. Computers in Industry, 91, 33–44.
73. Bhatia, M., & Sood, S. K. (2017). Acomprehensive health assessment framework to facilitate
IoT-assisted smart workouts: A predictive healthcare perspective. Computers in Industry, 92,
50–66.
74. Lee, C. K. M., Cheng, M. N., & Ng, C. K. (2015). IoT-based asset managementsystem for
healthcare-related industries. International Journal of Engineering Business Management, 7,
19.
75. Zhang, B., Wang, X., & Huang, M. (2014). A data replica placement scheme for cloud storage
under healthcare IoT environment. In 2014 11th International Conference on Fuzzy Systems
and Knowledge Discovery (FSKD) (pp. 542–547). IEEE.
76. Al-Majeed, S. S., Al-Mejibli, I. S., & Karam, J. (2015).Home telehealth by internet of things
(IoT). In 2015 IEEE 28th Canadian Conference on Electrical and Computer Engineering
(CCECE) (pp. 609–613). IEEE.
77. Qiu, T., Liu, X., Han, M., Li, M., & Zhang, Y. (2017). Srts: A self-recoverable time
synchronization for sensor networks of healthcare IoT. Computer Networks, 129, 481–492.
78. ´
Khoi, N. M., Saguna, S., Mitra, K., & Åhlund, C. (2015).Irehmo: An efficient IoT-based
remote health monitoring system for smart regions. In 2015 17th International Conference
on E-health Networking, Application & Services (HealthCom) (pp. 563–568). IEEE.
79. Oryema, B., Kim, H. S., Li, W., & Park, J. T. (2017). Design and implementation of an
interoperable messaging system for IoT healthcare services. In 2017 14th IEEE Annual on
Consumer Communications & Networking Conference (CCNC) (pp. 45–52). IEEE.
80. Ugrenovic, D., & Gardasevic, G. (2015). Coap protocol for web-based monitoring in IoT
healthcare applications. In 2015 23rd Telecommunications Forum Telfor (TELFOR) (pp. 79–
82). IEEE.
81. Gia, T. N., Rahmani, A. M., Westerlund, T., Liljeberg, P., & Tenhunen, H. (2015). Faulttolerant
and scalable IoT-based architecture for health monitoring. In Sensors Applications Symposium
(SAS), 2015 IEEE (pp. 1–6). IEEE.
82. Woo, M. W., Lee, J., & Park, K. (2018). A reliable IoT system for personal healthcare devices.
Future Generation Computer Systems, 78, 626–640.
198 A. I. Alsalibi et al.

83. Dhar, S. K., Bhunia, S. S., & Mukherjee, N. (2014). Interference awarescheduling of sensors
in IoT enabled health-care monitoring system. In 2014 Fourth International Conference of
Emerging Applications of Information Technology (EAIT) (pp. 152–157). IEEE.
84. Hindia, M. N., Rahman, T. A., Ojukwu, H., Hanafi, E. B., & Fattouh, A. (2016). Enabling
remote health-caring utilizing IoT concept over lte-femtocell networks. PLoS ONE, 11(5),
e0155077.
85. Chen, Xi., Ma, M., & Liu, A. (2018). Dynamic power management and adaptive packet size
selection for IoT in e-healthcare. Computers & Electrical Engineering, 65, 357–375.
86. Kim, T. Y., Youm, S., Jung, J. J., & Kim, E. J. (2015).Multi-hop wban construction for
healthcare IoT systems. In 2015 International Conference on Platform Technology and Service
(PlatCon) (pp. 27–28). IEEE.
87. Su, H., Wang, Z., & An, S. (2013). Maeb: Routing protocol for IoT healthcare.
88. Wu, T., Wu, F., Redoute, J. M., & Yuce, M. R. (2017). An autonomous wireless body area
network implementation towards IoT connected healthcare applications. IEEE Access, 5,
11413–11422.
89. Chen, X., Rhee, W., & Wang, Z. (2015). Low power sensor design for IoT and mobile
healthcare applications. China Communications, 12(5), 42–54.
90. Almobaideen, W., Krayshan, R., Allan, M., & Saadeh, M. (2017). Internet of things:
Geographical routing based on healthcare centers vicinity for mobile smart tourism desti-
nation. Technological Forecasting and Social Change, 123, 342–350.
91. Catarinucci, L., De Donno, D., Mainetti, L., Palano, L., Patrono, L., Stefanizzi, M. L., &
Tarricone, L. (2015). An IoT-aware architecture for smart healthcare systems. IEEE Internet
of Things Journal, 2(6), 515–526.
92. Chandel, V., Sinharay, A., Ahmed, N., & Ghose, A. (2016). ExploitingIMU sensors for IoT
enabled health monitoring. In Proceedings of the First Workshop on IoT-enabled Healthcare
and Wellness Technologies and Systems (pp. 21–22). ACM.
93. Elhayatmy, G., Dey, N., & Ashour, A. S. (2018). Internet of things based wireless body area
network in healthcare. In Internet of Things and Big Data Analytics Toward Next-Generation
Intelligence (pp. 3–20). Springer.
94. Jimenez, F., & Torres, R. (2015).Building an IoT-aware healthcare monitoring system. In 2015
34th International Conference of the Chilean Computer Science Society (SCCC) (pp. 1–4).
IEEE.
95. Simonov, M., Zich, R., & Mazzitelli, F. (2008). Personalized healthcare communication in
internet of things. In Proc. of URSI GA08, 7.
96. Thangaraj, M., Ponmalar, P. P., & Anuradha, S. (2015). Internet of things (IoT) enabled
smart autonomous hospital management system-a real world health care use case with the
technology drivers. In 2015 IEEE International Conference on Computational Intelligence
and Computing Research (ICCIC) (pp. 1–8). IEEE.
97. Gupta, M. S. D., Patchava, V., & Menezes, V. (2015). Healthcarebased on IoT using raspberry
pi. In 2015 International Conference on Green Computing and Internet of Things (ICGCIoT)
(pp. 796–799). IEEE.
98. Satija, U., Ramkumar, B., & Manikandan, M. S. (2017). Realtime signal quality-aware ECG
telemetry system for IoT-based health care monitoring. IEEE Internet of Things Journal, 4(3),
815–823.
99. Vasyltsov, I., & Lee, S. (2015). Entropy extraction from bio-signals in healthcare IoT. In
Proceedings of the 1st ACM Workshop on IoT Privacy, Trust, and Security (pp. 11–17).
ACM.
100. Anumala, H., & Busetty, S. M. (2015). Distributed devicehealth platform using internet of
things devices. In 2015 IEEE International Conference on Data Science and Data Intensive
Systems (DSDIS) (pp. 525–531). IEEE.
101. Azimi, I., Anzanpour, A., Rahmani, A. M., Pahikkala, T., Levorato, M., Liljeberg, P., & Dutt,
N. (2017).Hich: Hierarchical fog-assisted computing architecture for healthcare IoT. ACM
Transactions on Embedded Computing Systems (TECS), 16(5s), 174.
Internet of Things in Health Care: A Survey 199

102. Jabbar, S., Ullah, F., Khalid, S., Khan, M., & Han, K. (2017). Semanticinteroperability
in heterogeneous IoT infrastructure for healthcare. Wireless Communications and Mobile
Computing, 2017.
103. Pinto, S., Cabral, J., & Gomes, T. (2017). We-care: An IoT-based health care system for elderly
people. In 2017 IEEE International Conference on Industrial Technology (ICIT) (pp. 1378–
1383). IEEE.
104. Ray, P. P. (2014). Home health hub internet of things (h 3 IoT): An architectural framework for
monitoring health of elderly people. In 2014 International Conference on Science Engineering
and Management Research (ICSEMR) (pp. 1–3). IEEE.
105. Zois, D.-S. (2016). Sequential decision-making in healthcare IoT: Realtime health monitoring,
treatments and interventions. In 2016 IEEE 3rd World Forum on Internet of Things (WF-IoT)
(pp. 24–29). IEEE.
106. Alex, G., Varghese, B., Jose, J. G., & Abraham, A. (2016). Amodern health care system using
IoT and android. International Journal on Computer Science and Engineering (IJCSE) ISSN,
8(4).
107. Gelogo, Y. E., Oh, J. W., Park, J. W., & Kim, H. K. (2015).Internet of things (IoT) driven
u-healthcare system architecture. In 2015 8th International Conference on Bio-Science and
Bio-Technology (BSBT) (pp. 24–26). IEEE.
108. Jebadurai, J., & Peter, J. D. (2018).Super-resolution of retinal images using multi-kernel SVR
for IoT healthcare applications. Future Generation Computer Systems, 83, 338–346.
109. Byung Mun Lee. (2015). Dynamic data binding protocol between IoT medical device and IoT
medical service for mobile healthcare. International Journal of Smart Home, 9(6), 141–150.
110. Byung Mun Lee. (2015). Personalized service model for sharing medical devices in IoT
health-platform. Advanced Science and Technology Letters, 99, 180–182.
111. Byung Mun Lee. (2015). Registration protocol for health IoT platform to share the use of
medical devices. International Journal of Bio-Science and Bio-Technology, 7(4), 1–10.
112. Saxena, D., Raychoudhury, V., & SriMahathi, N. (2015). Smarthealthndnot: Named data
network of things for healthcare services. In Mobile- Health@ MobiHoc (pp. 45–50).
113. Talpur, M. S. H., Bhuiyan, M. Z. A., & Wang, G. (2015). Energy-efficient healthcare moni-
toring with smartphones and IoT technologies. International Journal of High Performance
Computing and Networking, 8(2), 186–194.
114. Ahmed, M. U., Björkman, M., & Lindén, M. (2015).A generic system-level framework for
self-serve health monitoring system through internet of things (IoT). In pHealth (pp. 305–307).
115. Jeong, J. S., Han, O., & You, Y. Y. (2016).A design characteristics of smart healthcare system
as the IoT application. Indian Journal of Science and Technology, 9(37).
116. Karamitsios, K., & Orphanoudakis, T. (2017). EfficientIoT data aggregation for connected
health applications. In 2017 IEEE Symposium on Computers and Communications (ISCC)
(pp. 1182–1185). IEEE.
117. Lim, Y., Daas, S. B., Hashim, S. J., Sidek, R. M., Kamsani, N. A., & Rokhani, F. Z.
(2015).Reduced hardware architecture for energy-efficient IoT healthcare sensor nodes. In
2015 IEEE International, Circuits and Systems Symposium (ICSyS) (pp. 90–95). IEEE.
118. Manogaran, G., Lopez, D., Thota, C., Abbas, K. M., Pyne, S., & Sundarasekar, R. (2017).Big
data analytics in healthcare internet of things. In Innovative healthcare systems for the 21st
century (pp. 263–284). Springer.
119. Miranda, J., Cabral, J., Wagner, S. R., Fischer Pedersen, C., Ravelo, B., Memon, M., &
Mathiesen, M. (2016). An open platform for seamless sensor support in healthcare for the
internet of things. Sensors, 16(12), 2089.
120. Neagu, G., Preda, Ş., Stanciu, A., & Florian, V. (2017). A cloud-IoTbased sensing service for
health monitoring. In E-Health and Bio-engineering Conference (EHB), 2017 (pp. 53–56).
IEEE.
121. Rasid, M. F. A., Musa, W. M. W., Kadir, N. A. A., Noor, A. M., Touati, F., Mehmood, W.,
Khriji, L., Al-Busaidi, A., & Mnaouer, A. B. (2014). Embedded gateway services for internet
of things applications in ubiquitous healthcare. In 2014 2nd International Conference on
Information and Communication Technology (ICoICT) (pp. 145–148). IEEE.
200 A. I. Alsalibi et al.

122. Sohn, S. Y., Bae, M., Lee, D. K. R., & Kim, H. (2015). Alarm system for elder patients
medication with IoT-enabled pill bottle. In 2015 International Conference on Information
and Communication Technology Convergence (ICTC) (pp. 59–61). IEEE.
123. Ali, F., Islam, S. R., Kwak, D., Khan, P., Ullah, N., Yoo, S. J., & Kwak, K. S. (2018).Type-
2 fuzzy ontology–aided recommendation systems for IoT–based healthcare. Computer
Communications, 119, 138–155.
124. Basanta, H., Huang, Y. P., & Lee, T. T. (2016). IntuitiveIoT-based h2u healthcare system for
elderly people. In 2016 IEEE 13th International Conference on Networking, Sensing, and
Control (ICNSC) (pp. 1–6). IEEE.
125. Lian, Y. (2015). Challenges in the design of self-powered wearable wireless sensors for
healthcare internet-of-things. In 2015 IEEE 11th International Conference on ASIC (ASICON)
(pp. 1–4). IEEE.
126. Lomotey, R. K., Pry, J., & Sriramoju, S. (2017). Wearable IoT data stream traceability in a
distributed health information system. Pervasive and Mobile Computing, 40, 692–707.
127. Mora, H., Gil, D., Terol, R. M., Azorín, J., & Szymanski, J. (2017). An IoT-based compu-
tational framework for healthcare monitoring in mobile environments. Sensors, 17(10),
2302.
128. Salunke, P., & Nerkar, R. (2017). IoT driven healthcare system for remote monitoring of
patients. Journal for Modern Trends in Science and Technology, 3(06), 100–103.
129. Elanthiraiyan, P., & Babu, D. S. (2015).Smart medicine and physical health system using IoT.
International Journal of Computer Science and Mobile Computing, 4(3), 333–338.
130. Fan, Y. J., Yin, Y. H., Da Xu, L., Zeng, Y., & Wu, F. (2014). IoT-based smart rehabilitation
system. IEEE Transactions on Industrial Informatics, 10(2), 1568–1577.
131. Granados, J., Rahmani, A. M., Nikander, P., Liljeberg, P., & Tenhunen, H. (2014).Towards
energy-efficient healthcare: An internet-of- things architecture using intelligent gateways. In
2014 EAI 4th International Conference on Wireless Mobile Communication and Healthcare
(Mobihealth) (pp. 279–282). IEEE.
132. Natarajan, K., Prasath, B., & Kokila, P. (2016). Smart health care system using internet of
things. Journal of Network Communications and Emerging Technologies (JNCET), 6(3).
133. Keerthana, K., Kiruthikanjali, N., Nandhini, G., Yuvaraj, G., et al. (2017). Secured smart
healthcare monitoring system based on IoT.
134. Yang, G., Xie, Li., Mäntysalo, M., Zhou, X., Pang, Z., Da Li, Xu., Kao-Walter, S., Chen, Q.,
& Zheng, L.-R. (2014). A health-IoT platform based on the integration of intelligent pack-
aging, unobtrusive bio-sensor, and intelligent medicine box. IEEE Transactions on Industrial
Informatics, 10(4), 2180–2191.
135. Alharbe, N., Atkins, A. S., & Akbari, A. S. (2013).Application of zigbee and rfid technolo-
gies in healthcare in conjunction with the internet of things. In Proceedings of International
Conference on Advances in Mobile Computing & Multimedia (pp. 191). ACM.
136. Amato, A., & Coronato, A. (2017). AnIoT-aware architecture for smart healthcare coaching
systems. In 2017 IEEE 31st International Conference on Advanced Information Networking
and Applications (AINA) (pp. 1027–1034). IEEE.
137. Healthcare. (2016). An overview of energy use and energy efficiency opportunities. https://
www.energystar.gov/ia/partners/publications/pubdocs/Healthcare.pdf.
Disease Diagnosis System for IoT-Based
Wearable Body Sensors with Machine
Learning Algorithm

Joseph Bamidele Awotunde , Sakinat Oluwabukonla Folorunso ,

Akash Kumar Bhoi , Paul Olujide Adebayo , and Muhammad Fazal Ijaz

Abstract In recent years, the continuous growth in global infectious disease coupled
with population growth and the associated increase in expectancy lead to the search
for new ways of making the most use of limited resources. Automated disease moni-
toring, diagnosis, prediction, and treatment of patients are not only for fast data but
also to get reliable service at reduced cost and accurate results from medical experts.
The combined wearable devices and Internet of Things (IoT) have been reformed
the medical system. This has minimized the response time in monitoring, diag-
nosis, prediction, and treatment by thrives toward the omnipresence of the health-
care services. However, the integration and design of IoT-based wearable sensors
present many challenges especially in the areas of data exchange, monitoring, and
diagnosing of patients. Therefore, the chapter proposes a framework for IoT-WBN-
based with a machine learning algorithm (ML). The data collected from different
wearable sensors like body temperature, glucose sensors, heartbeat sensors, and chest
were transmitted through IoT devices to the integrated cloud database. To select the
most useful features from the capture data, the ML was used, and the sensor signal
is analyzed using ML for diagnosis of patient data. The proposed framework can be
widely used in a remote area to monitor and diagnose patient health conditions to

J. B. Awotunde (B)
Department of Computer Science, University of Ilorin, Ilorin, Nigeria
e-mail: [email protected]
S. O. Folorunso
Department of Mathematical Science, Olabisi Onabanjo University, Ago-Iwoye, Nigeria
e-mail: [email protected]
A. K. Bhoi
Department of Computer Science and Engineering, Sikkim Manipal Institute of Technology
(SMIT), Sikkim Manipal University (SMU), Majitar, Sikkim 737136, India
P. O. Adebayo
Department of Computer Science, Federal Polytechnic, Nasarawa, Nigeria
M. F. Ijaz
Department of Intelligent Mechatronics Engineering, Sejong University, Seoul 05006, South
Korea
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 201
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_10
202 J. B. Awotunde et al.

reduce, and eliminate medical faults, reduce healthcare costs, minimize pressure on
medical experts, increase productivity, and enhancing patient satisfaction.

Keywords Smart healthcare system · IoT-based devices · Wearable body sensors

network · Healthcare system · Diagnosis · Infectious disease · Machine learning ·
Medical expert

1 Introduction

In the improvement of the healthcare system and the societies globally, tools like
Artificial Intelligence (AI), Internet of Things (IoT), Big Data, 5G Communication,
machine learning (ML) Cloud Infrastructure, and Blockchain remain paramount
technologies. Technologists and engineers would continue to face crucial obstacles
to incorporate and appreciate these optimized solutions and their advantages findings
carry out elaborate regard to risk management, resources, cost, scope, and quality.
Along with mobile device platforms rely on automated-hospital control services that
use automated surgical technologies (for example ultrasound, MRI, endoscopes, elec-
trocardiograms) and computerized patient information systems, healthcare system
technology platforms have been created.
The application of the IoT and wearable body sensors (WBSs) in recent years
through online services providing various dimensionalities in the healthcare system
[1]. This has created a healthy life for millions of people through the integration
of new technologies. The IoT-based devices with related technologies strengthened
the use of smart healthcare applications in real-time and remotely [1, 2]. IoT-based
allowed the integration of a huge number of various devices like fridge, wrist band,
umbrella, and air conditioner, thus removed the barrier of computing devices with
limited storage capacities like smartphone, laptop, and tablet.
The smart healthcare system has transformed the medical system into digital
hospitals, hence, the geographical borders have been overcome by the smart health-
care industry aim at remotely taking care of patients in real-time with high-level
medical services. Various technologies with big data analytics innovations have
transformed the smart healthcare system by creating several services to boost medical
systems. These technologies in conjunction with wearable medical device-based ML
technology moving the medical system from hospital-based therapy to a smart health-
care system. For instance, the huge amount of data collected using wearable sensors
can be used to monitor or track daily patient health information-like blood glucose,
heart rate, electrocardiogram, and nutritional information. Also, the industry is
widening its deployment to telemedicine, mobile wellness, cellular medical facilities,
and targeted therapy.
The IoT healthcare system has become one of the most indispensable parts of
human lives, this has dramatically increased the medical information system that
brings about big data. Healthcare practitioners are already adopting wearable devices
based on the IoT to streamline the diagnosis, monitoring, prediction, and treatment
Disease Diagnosis System for IoT-Based … 203

process. The healthcare system that depends on IoT assists individuals and aid their
vital everyday life activities.
Cloud computing technology has been used to handle the huge amount of data
generated by IoT devices called ‘big data’ and becomes easy to access and use. The
major role-playing by cloud-based applications with IoT-based system can never be
overlooked globally. For secured storage and easy accessibility, cloud computing can
be used in medical applications.
When these two innovations are combined, IoT and Cloud Computing support
each other in an equal manner. The tracking system is built by integrating these two
technologies to accurately track patient records, including at a remote location that
is useful to doctors. The cloud is used to boost the performance of IoT-based systems
to increase their storage and computational capacity, resource usage, and resources.
In recent years, WBS has been an emerging technology that attracts interest from
industrial and academic societies. The WBS with IoT-based applications in medicine
has been for physiological data captures in patients’ lifestyles. This is one of the
prominent and promising technologies that can be used for the realization of real-
time medical diagnosis and monitoring of patients from remote areas. The WBS can
be used as an attachment to the body, and it is a light-weighted wireless network
sensor. This is used to remotely monitor the physiological and other vital organs of
the body like blood glucose, blood pressure, heart rate, and body temperature among
others. Also, wearable devices can be used to process, monitor, and transmit data
to the cloud servers. The base station can be located near the human body. Such a
device will save patients money and time at the hospital.
Machine learning and cognitive algorithms have experienced advances recently,
hence, have been used to solve many complex problems. The ML help in the analysis
of the big data collected using IoT-based devices and stored in the cloud database. The
algorithms used for solving the intelligence aspect of the smart healthcare system for
accurate recognition, diagnosis, and for decision requires human-like intelligence.
The ML outperforms the conventional based methods of decision-making and diag-
nosis. This is useful in a pre-defined task, and the rules are learned from the actual
data. ML uses the secret data structure to make some intelligent decisions in smart
healthcare systems that the physicians can make conclusions. Big data is the core
component of the ML techniques’ high performance. A huge amount of knowledge
is produced by the different number of sensors in the WBS, which makes it possible
to integrate these cognitive skills into the WBS. It will automate the processing
and interpretation of the data. This leads to more efficient and smart solutions for
healthcare professionals that can save lives and time.
For continuous health tracking, wireless sensor networks can be implemented to
enhance the well being of patients, make the healthcare system more effective, and
help respond quickly to crises. The IoT-based healthcare system can be employed to
enhance the well being of patients, make the medical system more effective, and help
respond quickly to emergencies. Hence, it is possible to take advantage of the latest
technical arsenal to build a new wave of smart healthcare cities that would be able
to forecast widespread disease occurrences more accurately, supporting their claim.
It is not quick, apparently, and, as cities become more and more integrated, many
204 J. B. Awotunde et al.

problems will still arise. Nevertheless, the negative effects of the diseases widespread
can give this process the requisite boost. Therefore, this chapter proposed a new IoT-
WBS-based framework with an ML algorithm for disease diagnosis and monitoring
during severe diseases widespread for providing healthier facilities to the people over
the IoT-based platform.
The contributions of the chapter are:
• The general overview of the IoT system applications was presented.
• The applications of WBS in the healthcare system were discussed.
• We propose a framework for IoT-WBS-based with a machine learning algorithm.
• A practical case was used to implement the proposed system.
The followings are the structure of the chapter. Section 2 describes the general
overview of IoT-based applications in the smart healthcare system. Section 3
discusses the applications of wearable body sensors in smart healthcare systems.
Section 4 presents the IoT-wearable body sensors-based framework with a machine
learning algorithm for disease diagnosis. The practical case study of using IoT-based
devices with ML was presented in Sect. 5 for the diagnosis of heart diseases, while
Sect. 6 discusses the results and discussion. The conclusion and future work were
finally presented in Sect. 7.

2 The General Overview of IoT-Based Applications

in Smart Healthcare System

Technologies in combination with the healthcare system are working hands-in-hands

in recent years, and the expansion in the areas of IoT has created new opportunities
for experts in this area. Furthermore, with the adoption of wearable devices/sensors
by people across the globe, there are newly emerged applications for individualized
telemedicine and mobile health. This new emergence of these technologies is a result
of their high availability, simple to personalize, and easy accessibility; thus create
cost-effective on large scale and experts were able to deliver individualized content.
The next generation of mHealth and eHealth services have increasingly attracting
popularity with the introduction of big data analytics and IoT in the healthcare system.
In remote detection, prediction, surveillance, recovery, and therapy, where IoT
technology plays an important part, telemedicine has recently been broadly applied.
With interest in designing smart technologies, such as healthcare tracking systems,
medical diagnosis, prediction and treatment systems, and smart healthcare, IoT has
recently been implemented in the medical sector. Data are obtained from remote
medical devices, such as CT machines and MRI machines, wearable sensors, and
then distributed and reconfigured locally in three dimensions during the telemedicine
process. The IoT-based stuff provides a forum for sensors and machines to connect
inside a smart world and to exchange knowledge across medical channels easily.
Though the new fields emerging speedily, they also have their challenges, particularly
Disease Diagnosis System for IoT-Based … 205

when the goal is healthcare systems with a complicated problem, difficult in energy-
efficient, safe, flexible, suitable, and consistent solutions. By 2020 in the healthcare
system, IoT is projected to rise to a market size of $300B covering medical devices,
systems, applications, and services sectors [3].
This desire for customized e-health care is also likely to be promoted by govern-
ment initiatives. Many database clusters and resources are needed in general to store
big data, and these have become a challenge. It is an important problem to get concrete
trends from big data, such as patient diagnostic information. Nowadays, a variety
of emerging applications for different environments is being developed. Insensitive
systems, sensors are most commonly used for actual or near-future applications. In
recent years, experts have been tried to develop different wearable devices that can
be used for monitoring patient health conditions in remote areas [4, 5]. For instance,
the wearable sensor is used for monitoring physiological people’s health challenges
(e.g., exercises and food habits). Wearable sensors can continuously observe and
store the health data of the patient in a data store during this period [6].
The combined cloud and IoT-based application in terms of works well over ordi-
nary cloud-based applications. The cloud-IoT-based system can be used for emer-
gency and real-time applications like military, banking, and medical systems. In
particular, the cloud-based IoT approach would help provide medical applications
with effective resources for tracking and accessing information from any remote
location. The IoT-based data-centric application can be used to collect data for a
real-time update for a predefined period. Besides, this can be used to records rele-
vant medical data for effective disease diagnosis, and monitoring, especially before
the illness reaches a serious level.
In big data analytics, ML models can be used for intelligent decision-making due
to the huge data generation by the IoT-based devices. The method of implementing
data processing techniques for particular fields requires specifying the data involves
like the velocity, variety, and volume of such data. Normal data analysis modeling
involves the model of the neural network, the model of classification and the process
of clustering, and the implementation of efficient algorithms as well. The IoT devices
can be used to generate various types of data from several sources, hence, it is very
important to describe the features of the data generated for proper data handling.
These help in handling the various characteristics of capture data for scalability,
and velocity, thus help in finding the best model that can provide the best results
in real-time globally without any challenges. These are all known to be one of the
IoT’s big problems. However, in the latest technologies, these are all problems that
generate a large number of possibilities. Such information can be access using the
latest healthcare applications, and the data is securely stored on the cloud server.
In the field of healthcare, IoT devices such as satellites could have diverse tech-
nologies, for instance, the monitoring of various diseases like blood glucose, heart
rate, and endoscopic capsules [7, 8]. The healthcare industry can be more relevant
and operates in real-time in remote areas during disease outbreaks if mobile technolo-
gies, sensors, and actuator are combined [9, 10]. The structure of smart healthcare
devices helps in receiving information offered through web communications systems
for a better medical system [11, 12]. Approximately, 3.7 million biomedical devices
206 J. B. Awotunde et al.

are now in use, and then connected and tracked by various sections of the body
to record clinical remedies [13]. The IoT-based system can be used for monitoring
patients with infectious diseases remotely and for a long time. Also, provide real-
time diagnostic and monitoring to know the status of any patient living in remote
areas, hospitals, and clinics using wearable devices, hence, send the results to the
caregiver of medical experts for further examinations and decision-making. The
wearable devices connected to the human body to monitor physiological signs can
be incorporated or implemented as IoT technologies.
This showed that the IoT-based system operates with a massive deployment
of resources for diagnosis, and monitoring in real-time. The IoT is regarded to
be the large-scale application of machine type communication (MTC) equipment
that performs sensing actuation activities through nominal human intervention. The
number of Internet-connected objects is estimated to surpass the number of persons in
society. That is because IoT is focused on the reasonableness of networking specific
physical gadgets via the connected networks, for instance, the majority of health
gadgets linked to the Internet is growing exponentially every day [14]. Besides,
healthcare services are considered the largest target market for IoT, with the pulse
rate tracking being identified as the most significant benefit. Use essential devices
such as glucose levels pulse rate tracking, the IoT systems used to implement urgent
warning systems, and remote medical surveillance systems accessing data into more
advanced technologies.
IoT-based applications have many features that help in remote monitoring of a
patient, thus begin to gain attention due to scalability, and accessibility in healthcare
sectors [15–17]. The links between IoT-based system and sensors are established
through wireless play in smart healthcare applications. The proper management of
the gateways results in the successful application of IoT-based system. Thus, the
protocol conversion at the sides of the gateways must fault tolerance for efficient
implementation of the system [17].
Ideally, engineers and technicians would avoid warehouses, isolated locations, or
busy places traveling and visiting because they fear being infected with the virus.
Nevertheless, their physical appearance on site is needed because they cannot get
away with manually operating the equipment available. By the use of special sensors
mounted in machinery and enhanced knowledge, real-world overlays, and remote
expert feedback, IoT may help solve their repair problems, conduct machine oper-
ation remotely, and AI may also anticipate when machines need repair (predictive
maintenance) or when they might be faced with a challenge. In this way, phys-
ical visits will be significantly decreased, helping to safeguard workers’ health and
welfare, thus enhancing efficiencies at the level of service of the facilities. When
supermarkets continue imposing limits on sales of goods per customer, future alter-
natives have also been produced by IoT and AI. Smart shelves, smart fridges, video
analytics, and an end-to-end integrated supply chain will help retailers deal with
planning challenges and even reduce the extreme behavior of customers due to
hysteria.
Smart-watches and fitness trackers will be readily available in the not-too-distant
future and people with chronic conditions will be able to record temperature, asthma,
Disease Diagnosis System for IoT-Based … 207

heartbeat without the use of intrusive instruments. For this fact, patients will be able
to opt to transmit real-time and past data to public or private hospitals anytime they
feel unwell so that medical health IT departments will interact with the wearables
and mobile devices of patients. In this way, the treatment of various diseases may
be prepared even more effectively and with minimal resources expended optimally.
Smart connected medical devices, such as smart home ventilators, together with video
and wearables, may support patients monitoring at home, give updates to those in
distress, or even alert when paramedics are required to come and move them to the
hospital.
An IoT-3D printing may be a lifesaver in the face of infectious disease moni-
toring and diagnosis shortages. With an IoT-3D printer that offers critical medical
equipment, for example, replacement valves. This is what an Italian company called
Isinnova does, taking a 3D printer to a Milan hospital and manufacturing incom-
plete valves to be shipped to a hospital in Brescia, Northern Italy. Touch screens
have been the chosen user interface (UI) until recently: among them are tablets,
computers, and even doors. Nonetheless, the fact that infectious disease is more
quickly transmitted from a contaminated surface than by air has made direct contact
sound risky. Many UIs that do not need any physical contact are also usable. Voice
has already triumphed over tactile user interfaces, especially through smart speakers
and digital helpers. Despite people confining themselves inside the building, there
will be increased interest in smart home apps, and the voice apps of smart speakers
(Fig. 1).
These technologies are now being implemented in major cities to reduce various
disease widespread. For example, for simultaneous tracking of a patient at home,
the Shanghai Public Health Clinical Center utilizes physique fever monitors laterally
utilizing facts transfer directly to the nurse post, thus minimizing possible sensitivity
to healthcare workers [18]. Similarly, a device now used for medical interviews with
sensors was installed in Boston to assess the breathing degree and physique fever of
patients.
IoT also helps physicians and healthcare practitioners to offer medical services to
patients by providing them with solutions in crucial and rural locations [19, 20]. This
approach also decreases the overall type of discomfort and increases the productivity
among others, the doctors, workers, nurses. It reduces unnecessary hospital admis-
sions and the overall burden on healthcare organizations by linking older patients
directly with their physicians and thus enabling health information to be transferred
via a protected site.
The introduction of IoT-based system to the disease monitoring and diagnosis
can be used to build a social forum to help individuals access appropriate treatment
at home and to develop a robust repository on disease control for government and
healthcare organizations by diagnosis and healthcare devices could be obtained for
persons with minor symptoms (Preventive disguises, thermometers, medicines, and
personalized disease infection diagnosis). Patients were able to upload their overall
health to the IoT (clinical data storage) site online regularly and exchange their
relevant data to area hospitals, and state and local health offices.
208 J. B. Awotunde et al.

IoT Enable
Ambulance

Nanotechnolog

Ambient Intelligence
IoT-based Environment

Digital Telehealth

Wearable Devices Smart IoT-Based IoT-Based Transportation

Education

Fig. 1 Potential applications of IoT-based in smart healthcare system

The use of laboratory tests can now be combined with big data generated from
sensors to diagnose the health condition of patients by physicians to achieve better
results. Sensor data is also most commonly used to take effective action for the
recommendation of patient well being and care, lifestyle decisions, and early diag-
nosis for patient health improvements. The old methods of storing data are no more
helpful where the volume, speed, and variety of data are increasing in IoT-based
sensor application domains. The creation of an appropriate storage system for the
storage and processing of voluminous data needs this issue [21].
The accelerated growth of mobile phones, wireless networking, radar systems,
and entrenched technologies has been contributed to medical information schemes
entering the age of omnipresent computers. This has been contributed to a steady
increase in technologies and apps (i.e., wearable gadgets) for the delivery of inno-
vative, omnipresent health systems [22], which are being increasingly implemented
worldwide [23]. On the other hand, the IoT [24] offers computing resources and Web
Disease Diagnosis System for IoT-Based … 209

connectivity to ordinary objects that can access, interact, and archive data from the
real world and most IoT applications are very widespread [25].

3 The Applications of Wearable Body Sensors in Smart

Healthcare System

The smart healthcare system has been proven to be a system that can help healthcare
professionals to save lives and times by an automatic collection of data and real-time
processing. The WBS has been widely used in various fields for several applica-
tions. The system can greatly help to reduce cost and save time spent in treating
elderly patients and reduced the workload of the medical experts. The mechanism by
Hommersom et al. [26] will strengthen the relationship between patient and doctor
and enable the patients themselves to self-manage chronic diseases. The embedded
system with wireless devices is user-friendly, use in remote areas, and very secure
compare to conventional services.
The IoT-based systems save the stress of queue at the hospitals especially patients
that need to be monitor regularly like blood pressure or patient glucose level. Also,
there will be more time and space for the physicians and experts that deal with the
diagnosis, monitoring, and treatment of a patient, thus have enough time to attend
to critical needs in the system. Also, ML help the smart healthcare system to be
more efficient and cost-effective by provided more accurate results, and the hazard
that associated with conventional systems are reduced [27, 28]. In IoT-WBS-based
systems, the ML methods are used to increase processing at various levels [29, 30].
It can be used either to improve the efficiency of the network itself or to gain insight
and take the appropriate actions from the data collected.
In IoT-WBSs, ML has found a big application. The integration of such cognitive
skills into the IoT-WBS-based system would improve the healthcare systems, and
offer a solution to many human life issues [31]. The devices collect a large amount of
data using wearable devices and sensors attached to the human body. Such sensors
track and wirelessly by sending the capture data to the cloud servers for proper
keep. ML algorithms will be fueled by the high volume of data generated to provide
accurate results from diagnosis, monitoring, and forecasting of diseases for decision-
making. There are still critical problems to be addressed using an IoT-Based system
with ML models for a better result. ML models are very paramount in IoT-based
system for training and evaluation of collected data for proper decision-making.
Technological advances of low-cost, insightful, compact, and insubstantial
medical sensor networks have been accepted for positioning themselves ideally on
the human psyche communication wirelessly [32]. This helps to create a network for
long-term tracking of various major chronic physical diseases, delivering real-time
input to users and medical workers [32]. The WSN technologies are regarded as one
of the leading scientific fields for enhancing the quality of healthcare sectors [43].
Utilizing digital innovations like wearable sensor networks and the IoT movement are
210 J. B. Awotunde et al.

bringing improvement in diseases guided surveillance of patients [9]. Together with

the IoT powered mobile devices, the processing strategies are supporting evolving
and encouraging real-time healthcare systems in isolated areas. Meanwhile, elec-
tronic and wireless connections have been thoroughly changed the medical sectors
to become smart healthcare applications.
The implementation of these devices with other various technologies funda-
mentally the conventional medical system with high sustainable, intelligent, and
omnipresent medical cities and homes for the elderly in remote areas [34]. The WBS
is an improvisatory and prospective candidate for stepping up medical innovation
and education to improve the healthcare platform further. The IoT-based system has
transformed our lifestyles in various ways with superior technologies in the areas of
transportation, retails, entertainment, and healthcare system with given emergency
services control. A multidisciplinary concept of ambient intelligence has been estab-
lished to resolve the challenges we face in our daily lives by combining several
healthcare technologies [35].
These have become a successful technique that lets for a wide range of appli-
cations. New technology has enabled WBS to be useful for many purposes in the
production and integration of IoT-based devices and other related devices [36]. The
system can be used in various fields of medical industries like surveillance, and
other fields like in engineering for heavy machinery, seismic monitoring, control
of the environment, and agriculture. Telehealth tracking is among the most hopeful
applications of WSN [37]. The sensor network used a memory chip, a multi-hop
source of power with a multi-node network provided by a spectrum analyzer [38].
The nodes in the same proximity will communicate with each other via routing
responsibilities [37]. The ad-hoc used to make them fault-tolerant, robust, fault-
tolerant with larger coverage areas when compared with normal sensor nodes that
are meticulously built and stated in one location [39]. Using a private system that can
extensively monitor and track patient health globally has helped the healthcare sector.
These strategies will also alleviate strain and stress on healthcare personnel, eradi-
cate medical faults, reduce the effectiveness of workload and medical professionals,
reduce lengthy healthcare costs and increase client experience [40, 41].
Also, as displayed in Fig. 2, BSNs consists of various medical and wearable
devices to communicate and monitor physiological signals like temperature, pulse,
and blood pressure wirelessly. Wearable sensors can collect and record data on one’s
diagnosable disorder and moving responses in real-time.
Lithium battery smart devices are needed to efficiently and conveniently transfer
data between the WBAN and personal servers, and their power consumption during
transmission should be relatively low. Smartphones with modern technologies like 5G
technology and GPRS are examples of IoT devices. These help in keeping caregivers,
medical experts, and patients updated with the recent health status of the end-users.
This, when applied to personalized medicine, is called healthcare monitoring wear-
able networks [42, 43]. The WBSN allows for the integration of low-power, remotely
controlled, intelligent, omnipresent sensor nodes to track physiological proximity.
Disease Diagnosis System for IoT-Based … 211

Smart
Glass

Global Blood
Positioning Sugar

Global
Positioning Smart
Belt

Heart Rate
Proximity

Electromyography Body
Temperature

Blood Airflow
Pressure (Breathing)
Sleeping
Monitoring Smart
Shoes

Smart
Socks

Fig. 2 Worthwhile sensor devices for collection and capture of data

4 IoT-Wearable Body Sensors-Based Framework

with Machine Learning Algorithm for Disease Diagnosis

The diagnostic system architecture based on the IoT-WBS was used to capture data
for the creation of smart personalized medical systems with the integration of various
IoT-based sensors. Data are not necessarily available instantly in most situations, so it
is difficult to analyze or incorporate a broad variety of data for diagnosis. A big issue
that needs to be discussed urgently is the collection of data using IoT-based device.
Subsequently, a series of gene expression of any diseases promising to examine
the effectiveness of artificial intelligence with IoT-based devices. It would also be
necessary to critically investigate whether there was any other context information
that could boost the model’s performance.
Also, to evaluate the performance of the disease characteristics selected from
each biomarker, further studies are needed. This chapter proposes the IoT-WBS-
based diagnostic system architecture, which applies a set of connected wearable
technology to monitor and monitor the health condition of a person. The capture
data will be transferred directly to the cloud server using the wireless gateways
to avoid using smartphone memory as storage facilities and processing. Figure 3
displayed the framework for the proposed system.
212 J. B. Awotunde et al.

Fig. 3 The proposed architecture IoT-WBS-based for disease diagnosis and monitoring

The proposed IoT-WBS-based disease diagnosis system has a four-level design

to capture, store, and process enormous devices/sensors information. The first layer
is used to capture and gather information IoT-based device for data analysis for
monitoring physiological body signs for disease diagnosis. In any situation, a patient
can have several wearable sensors, and very possible for a patient to have more than
one wearable sensor (WSs), but experts determine the total number of WSs and this
also depends on the required information to capture. The second layer is the gateway
between the data capture and the third layer (Cloud-IoT subsystem) and the IoT
devices link to the Internet and transfer the obtained data from the first layer to the
cloud storage database. Likewise, the third layer employs Apache to build up the
ML-based diagnosis for the disease.
The ML model generates a real-time diagnosis, and monitoring events, and
produce the required reports needed by experts [27–30]. This provides a high
computing ability to enhance the IoT devices that have a lower computing skill.
With the different ML models that report disease suspicious events, the doctors and
experts can promptly make a decision and advise accordingly. The third layer has
two main functions (i) Generate real-time models to diagnose, and monitor unde-
sired events or signs related to many diseases such as hypertension, dengue, chronic
cholestasis, drug reaction, heartbeat, and measuring a reduced amount of insulin in
Disease Diagnosis System for IoT-Based … 213

a patient body and (ii) help to determine the positions of IoT-based devices in the 3D
environment and the total number of IoT-based devices. The last layer is the diag-
nosis alert generation, the result of this layer is a description of the circumstances of
patients sent to doctor and expert. Also, the layer responds to signs of irregularity
when it is identified, e.g., submitting a request for assistance (e.g., demand for an
assistant care provider) or an urgent request (e.g., call for an ambulance) when an
urgent condition is detected.

5 The Application of Machine Learning for the Diagnosis

of Heart Diseases as Case Study

In western culture, cardiac arrest is a crippling disease. Various conditions can be

used to defined the cardiovascular disease (CVD) this conditions affecting the body’s
heart and blood vessels, such as atherosclerosis caused in the arteries by plaque
accumulation. A block plate can be caused by plaque develop and form clots. Either
a heart attack or stroke can be cuased by a smooth flow of blood in the artery [44].
The side effects of CVD are in various forms-like prolapsing of the mitral valve,
pulmonary stenosis, dilated cardiomyopathy, congenital heart disease, arrhythmia,
coronary artery disease among others.
The normal age of death between 1990 and 2015 has been decreased by the main
factor of the CVD, and in Africa, heart disease has cause highest rate of death and
sudden deaths globally has result of this disease [45]. Effective detection of such
deadly diseases would also significantly help to minimize death rates globally. The
prediction of who next to develop CVD as been a major research challenge globally
[46]. Various parameters like smoking habits, age, drug has been used as varibles
to predict who next to improved the CVD, but none have yet seemed to provide an
accurate prediction models [47]. The rise in the number of infestious and chronic
illness can be attributed to a global population increase [48]. The exponential increase
in volume, pace, and diversity of disease and health conditions data has therefore
called for smarter and more robust healthcare systems for effective data storage and
control to fulfill the standards of huge datasets [48, 49].
To achieve accuracy, efficiency, management, identification of important features
to for prediction in COVID-19 outbreak, computational intelligence (CI) methods
is very necessary on the medical data [50]. The importance of the IoT technique in
providing real-time tracking, and diagnosis in healthcare system has been proved
by various authors [48, 49]. When compare the CVD with other communicable
diseases, CVD is major global disease due to the increase of aging, urbanization,
and globalization [51]. In 2005, 30% of global deaths estimated to be 17 million
people suffered and died as a result of CVD illness, with 7.2 and 5.7 million deaths
due to heart attacks and stroke, respectively; 80% of these deaths were reported by
developing countries, and the WHO estimated 24 million deaths can still be recorded
if the trend is not stop on time by 2033 [52, 53].
214 J. B. Awotunde et al.

5.1 The Heart Disease Dataset Characteristics

This heart disease dataset contains 14 attributes with 303 instances [54]. The dataset
description is depicted in Table 1.

Table 1 The details features of the CVD dataset

S/N Features Description Data type Data values
1 Age Age in years Numeric
2 Sex Numeric 0 = Male
1 = Female
3 Cp Chest pain type Numeric 1: Typical angina;
2: Atypical angina;
3: Non-anginal pain;
4: Asymptomatic
4 Treetops Resting blood pressure (in mm Numeric
Hg on admission to the hospital)
5 Chol Serum cholesterol in mg/dl Numeric
6 Fbs Fasting blood sugar >120 mg/dl Numeric 1 = True;
0 = False
7 Resting Resting electrocardiographic Numeric 0: Normal;
results 1: Having ST-T wave
abnormality;
2: Showing probable or definite
left ventricular hypertrophy
by Estes’ criteria
8 Thalach Maximum heart rate achieved Numeric
9 Exang Exercise induced angina Numeric 1 = Yes
0 = No
10 Oldpeak Oldpeak = ST depression Numeric
induced by exercise relative to
rest
11 Slope The slope of the peak exercise Numeric 1: Upsloping;
ST segment 2: Flat;
3: Downsloping
12 Ca Number of major vessels (0–3) Numeric
colored by flourosopy
13 Thal Thal: 0 = normal; 1 = fixed Numeric 0; 1; 2;
defect; 2 = reversable defect
14 Target The predicted attribute Numeric 0 = No heart disease
1 = Heart disease
Disease Diagnosis System for IoT-Based … 215

5.2 Performance Evaluation Metrics

The true negative (TN), true positive (TN), false negative (FN), and false positive
(FP) were used has performance evaluation in this chapter. Table 2 depicted the
explanations of the performance evaluations.
Where:
TN = the positively diagnosed diabetes by the model, the healthy people.
TP = the positively diagnosed diabetic mellitus patient in the dataset used.
FN = the negatively diagnosed individuals diagnosed with diabetes mellitus
healthy patient.
FP = the false positively diabetic patient but not diagnosed by the model.

6 Results and Discussion

Six classifier models was used on the training set dataset, and the results of the
preliminary performance evaluation using all 13 input parameters are presented in
this section. The results of experiments are better performed with the Extra Tree
algorithm. This means that the performance of the algorithm is good at recording
accuracy, with a recall and accuracy value of 88%, 86% and 87%, respectively. This
outcome is favorable in heart diseases with previously implemented algorithms.
87% of the test data are correctly identified by the classifier, and the sensitivity
is balanced between the two target class groups. In terms of accuracy, extra tree
effectively registered 88%, demonstrating its ability to reliably provide accurate
results. In addition, a ROC-AUC value of 94% reported by the classifier indicates
that it has a good dissimilarity measure. This means it has a 94% chance of reliably
determining from patient data between the existence and avoidance of heart disease.

Table 2 Performance evaluation metrics

Measures Definition Formula
Sensitivity test Evaluate the ability to diagnose diabetes illness TP/(TP + FN)
appropriately
Specificity test Evaluate the desire to stop labeling diabetes illness TN/(TN + FP)
ordinary stuff in diagnostic exams
PPV Positively predicated value TP/(TP + FP)
NPV Negatively predicated value TN/(TN + FN)
Accuracy Evaluate the amount of diabetes detectors accurately TN+TP
TN+FP+FN+TP × 100
216 J. B. Awotunde et al.

6.1 The Precision-Recall Curve (PRC)

The PRC displays the tradeoff between precision and recall for different probabilities
threshold. For the PRC values, the closer to 1, the better the classification performance
of the model. A high PRC value signifies both high recall and high precision, where
high precision depicts a low-false-positive rate, and high recall depicts a low-false-
negative rate. High scores for both precision and recall indicate that the learner
is returning accurate outcomes (high precision), as well as returning a majority of
all positive outcomes (high recall). Figure 4 shows that extra trees gave a superior
performance with a PRC value of 0.940 and 0.937 for No heart and heart disease
classes, respectively.

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
XGBoost LGBM Extra Tree Decision k-NN SVM Random
Tree Forest

Accuracy Precision Recall F1-Score ROC RPC

Fig. 4 Performance evaluation of the machine learning algorithms

Table 3 Performance evaluation metrics

Models Accuracy Precision Recall F1-score ROC_AUC RPC
XGBoost 80 0.80 0.80 0.80 0.91 0.894
LGBM 79 0.79 0.78 0.78 0.89 0.885
Extra tree 87 0.88 0.86 0.87 0.94 0.935
Decision tree 80 0.81 0.80 0.80 0.80 0.744
k-NN 64 0.64 0.63 0.63 0.63 0.597
SVM 57 0.57 0.56 0.55 0.67 0.689
Random forest 84 0.84 0.83 0.83 0.91 0.903
Disease Diagnosis System for IoT-Based … 217

Table 3 presents the performance evaluation of seven diverse learners based on

the assessment of precision, recall, F1-Score, and accuracy metrics. This outcome
is based on the test set which is 20% of the dataset. It is observed that Extra Tree
achieved the maximum classification report values of above 80% across all evaluation
measures. The assessment reported 87%, 0.88, 0.86, and 0.87 values for accuracy,
precision, recall, and F1-Score, respectively. Its performance value is more consistent
than with other models across all metrics. The least performing learner was SVM
achieving 57% accuracy. This indicates that the model was guessing ad could barely
classify correctly.

6.2 Confusion Matrix

This section further analysis the assessment of all models using the confusion matrix
for the test set as shown by Fig. 5. It is observed by the outcome presented in
Fig. 5 representing the confusion matrix for extra trees gave the best recall for each
class with minimal error. Two classes were analyzed where ‘0’ represents ‘No heart
disease’ and ‘1’ represents ‘Heart disease’. For XGBoost, out of 28 test instances of
No heart disease, 21(75%) instances were correctly classified as No heart disease,
7(25%) instances were incorrectly classified as with heart disease. For the class
heart disease, 5(15.16%) instances were misclassified as No heart disease, while 28
instances (84.85%) were correctly classified as heart disease class. Comparing all
models on Heart disease classification, extra trees performed best with a recall rate
of 0.86. For the model, out of 28 instances of No heart disease, 22(78.57%) were
correctly classified as No heart disease, while 6(21.43%) instances were misclassi-
fied as heart disease. For heart disease class, 31(93.94%) instances were correctly
classified as heart disease while 2(6.01%) instances were misclassified as No heart
disease. For most models, it could be observed that the majority of the misclassifi-
cation was No heart disease was misclassified as heart disease. Hence, our objective
has been achieved by building a classification model for heart disease detection.

7 Conclusion and Future Research Directions

The smart healthcare system has the capacity of producing and communicating huge
quantities of data produced from IoT-based devices and sensors. The sizes and quan-
tities of networking equipment are more than ever and the volume of data grows even
more than ever before. For example, IoT-based devices and sensors can be used to
capture and collect patient physiological features and check metrics for patient diag-
nosis and monitoring. The strong association came from mental fitness data which
were closely related, including the vital signs of the body. The function of the body
is not accurately described by itself or a small portion of the data. For example, data
on human temperature or blood pressure that a patient senses as elevated cannot
218 J. B. Awotunde et al.

XGBoost Confusion Matrix

LGBM Confusion Matrix Extra Tree Confusion Matrix

DT Confusion Matrix
K-NN Confusion Matrix SVM Confusion Matrix

RF Confusion Matrix

Fig. 5 Confusion matrix for the machine learning algorithms

ascertain the illness, nor does it decide what type of illness a patient has. Its cause
is perhaps acute infection, diabetes, malaria, fever, injury to the tissue, hepatitis,
leukemia. Nevertheless, if the white blood cells in urine are found to be strong in
the interim, you can conclude that it suffers from kidney failure in the first place.
Inaccessible healthiness discovery principal analytic facilities Rural health facilities
are severely limited in several developed countries, so it is important to incorporate
rural medical infrastructure using advanced detection technologies and to create a
‘large health initiative’ appropriate for rural areas. While it is possible to monitor
the development of all types of growing acute, chronic sickness inmates. Bettering
care, antecedent detection, antecedent intervention, and early recovery will advance
the value of the life of people with serious illness and decrease the risk of sickness.
Disease Diagnosis System for IoT-Based … 219

The chapter, therefore, proposed an IoT-WBN-based system with a machine learning

algorithm (ML) to diagnose capture data from the patient body. The data collected
from different wearable sensors like body temperature, glucose sensors, heartbeat
sensors, and chest were transmitted through wirelessly interconnected devices to the
integrated cloud database. The ML was used to extract features from the patient
capture data, and the sensor signal is analyzed using ML for diagnosis of patient
data. Heart disease was used as a case study to prove the applicability of the proposed
system. From the seven ML algorithms used, the extra tree performs best. The algo-
rithm performance is good at recording precision, with a recall and accuracy value of
88%, 86%, and 87%, respectively. This result is competitive with previously applied
algorithms in heart diseases. The classifier was able to correctly classify 87% of the
test data and the sensitivity is balanced between the two categories of the target class.
In future work, more effective security algorithms-like Spike lightweight algorithms
will be implemented to secure the IoT-based devices. It is because data privacy and
security are very pressing priorities for IoT-based diagnosis smart healthcare systems.

References

1. Kumar, P. M., Lokesh, S., Varatharajan, R., Babu, G. C., & Parthasarathy, P. (2018). Cloud and
IoT based disease prediction and diagnosis system for healthcare using Fuzzy neural classifier.
Future Generation Computer Systems, 86, 527–534.
2. Yuehong, Y. I. N., Zeng, Y., Chen, X., & Fan, Y. (2016). The internet of things in healthcare:
An overview. Journal of Industrial Information Integration, 1, 3–13.
3. Firouzi, F., Rahmani, A. M., Mankodiya, K., Badaroglu, M., Merrett, G. V., Wong, P., &
Farahani, B. (2018). Internet-of-Things and big data for smarter healthcare: From device to
architecture, applications, and analytics.
4. Peterson, T., Rice, J., & Valane, J. (2005). Solar tracker (p. 476). ECE: Final Project in Cornell
University.
5. Paradiso, R., Loriga, G., & Taccini, N. (2005). A wearable health care system based on knitted
integrated sensors. IEEE Transactions on Information Technology in Biomedicine, 9(3), 337–
344.
6. Lorincz, K., Malan, D. J., Fulford-Jones, T. R., Nawoj, A., Clavel, A., Shnayder, V., Mainland,
G., Welsh, M., & Moulton, S. (2004). Sensor networks for emergency response: Challenges
and opportunities. IEEE pervasive Computing, 3(4), 16–23.
7. Pramanik, P. K. D., Upadhyaya, B. K., Pal, S., & Pal, T. (2019). Internet of things, smart
sensors, and pervasive systems: Enabling connected and pervasive healthcare. In Healthcare
Data Analytics and Management (pp. 1–58). Academic Press.
8. Srivastava, G., Parizi, R. M., & Dehghantanha, A. (2020). The future of blockchain tech-
nology in healthcare internet of things security. In Blockchain Cybersecurity, Trust and Privacy
(pp. 161–184). Springer, Cham.
9. Oladipo, I. D., Babatunde, A. O., Awotunde, J. B., & Abdulraheem, M. (2021). An improved
hybridization in the diagnosis of diabetes mellitus using selected computational intelligence.
Communications in Computer and Information Science, 1350, 272–285.
10. Darwish, A., Ismail Sayed, G., & Ella Hassanien, A. (2019). The impact of implantable sensors
in biomedical technology on the future of healthcare systems. Intelligent Pervasive Computing
Systems for Smarter Healthcare, 67–89.
11. Joyia, G. J., Liaqat, R. M., Farooq, A., & Rehman, S. (2017). Internet of Medical Things
(IOMT): Applications, benefits, and future challenges in the healthcare domain. The Journal
of Communication, 12(4), 240–247.
220 J. B. Awotunde et al.

12. Adeniyi, E. A., Ogundokun, R. O., & Awotunde, J. B. (2021). IoMT-based wearable body
sensors network healthcare monitoring system. In IoT in Healthcare and Ambient Assisted
Living (pp. 103–121). Springer, Singapore.
13. Qadri, Y. A., Nauman, A., Zikria, Y. B., Vasilakos, A. V., & Kim, S. W. (2020). The future
of healthcare internet of things: A survey of emerging technologies. IEEE Communications
Surveys & Tutorials, 22(2), 1121–1167.
14. Alharthi, N., & Gutub, A. (2017). Data visualization to explore improving decision-making
within Hajj services. Scientific Modeling and Research, 2(1), 9–18.
15. Sebestyen, G., Hangan, A., Oniga, S., & Gál, Z. (2014, May). eHealth solutions in the context
of the internet of things. In 2014 IEEE International Conference on Automation, Quality, and
Testing, Robotics (pp. 1–6). IEEE.
16. Rahmani, A. M., Thanigaivelan, N. K., Gia, T. N., Granados, J., Negash, B., Liljeberg,
P., & Tenhunen, H. (2015, January). Smart e-health gateway: Bringing intelligence to
internet-of-things based ubiquitous healthcare systems. In 2015 12th Annual IEEE Consumer
Communications and Networking Conference (CCNC) (pp. 826–834). IEEE.
17. Woo, M. W., Lee, J., & Park, K. (2018). A reliable IoT system for personal healthcare devices.
Future Generation Computer Systems, 78, 626–640.
18. Ogundokun, R. O., & Awotunde, J. B. (2020). Machine learning prediction for COVID-19
pandemic in India. medRxiv.
19. Jan, M. A., Usman, M., He, X., & Rehman, A. U. (2018). SAMS: A seamless and authorized
multimedia streaming framework for WMSN-based IoMT. IEEE Internet of Things Journal,
6(2), 1576–1583.
20. Qureshi, F., & Krishnan, S. (2018). Wearable hardware design for the internet of medical things
(IoMT). Sensors, 18(11), 3812.
21. Ng, J. W., Lo, B. P., Wells, O., Sloman, M., Peters, N., Darzi, A., Toumazou, C., & Yang, G. Z.
(2004, September). Ubiquitous monitoring environment for wearable and implantable sensors
(UbiMon). In International Conference on Ubiquitous Computing (Ubicomp).
22. Connelly, K., Mayora, O., Favela, J., Jacobs, M., Matic, A., Nugent, C., & Wagner, S. (2017).
The future of pervasive health. IEEE Pervasive Computing, 16(1), 16–20.
23. Kutia, S., Chauhdary, S. H., Iwendi, C., Liu, L., Yong, W., & Bashir, A. K. (2019). Socio-
technological factors affecting user’s adoption of eHealth functionalities: A case study of
China and Ukraine eHealth systems. IEEE Access, 7, 90777–90788.
24. Minerva, R., Biru, A., & Rotondi, D. (2015). Toward a definition of the internet of things (IoT).
IEEE Internet Initiative, 1(1), 1–86.
25. Zahoor, S., & Mir, R. N. (2018). Resource management in pervasive internet of things: A
survey. Journal of King Saud University-Computer and Information Sciences.
26. Hommersom, A., Lucas, P. J., Velikova, M., Dal, G., Bastos, J., Rodriguez, J., Germs, M.,
& Schwietert, H. (2013, October). Moshca-my mobile and smart health care assistant. In
2013 IEEE 15th International Conference on e-Health Networking, Applications and Services
(Healthcom 2013) (pp. 188–192). IEEE.
27. Ayo, F. E., Awotunde, J. B., Ogundokun, R. O., Folorunso, S. O., & Adekunle, A. O. (2020). A
decision support system for multi-target disease diagnosis: A bioinformatics approach. Heliyon,
6(3), e03657.
28. Ayo, F. E., Ogundokun, R. O., Awotunde, J. B., Adebiyi, M. O., & Adeniyi, A. E. (2020, July).
Severe acne skin disease: A fuzzy-based method for diagnosis. Lecture Notes in Computer
Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in
Bioinformatics), 2020, 12254 LNCS, pp. 320–334.
29. Oladele, T. O., Ogundokun, R. O., Awotunde, J. B., Adebiyi, M. O., & Adeniyi, J. K. (2020,
July). Diagmal: A malaria coactive neuro-fuzzy expert system. Lecture Notes in Computer
Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in
Bioinformatics), 2020, 12254 LNCS, pp. 428–441.
30. Awotunde, J. B., Jimoh, R. G., Oladipo, I. D., & Abdulraheem, M. (2021). Prediction of
malaria fever using long-short-term memory and big data. Communications in Computer and
Information Science, 1350, 41–53.
Disease Diagnosis System for IoT-Based … 221

31. Oniani, S., Marques, G., Barnovi, S., Pires, I. M., & Bhoi, A. K. (2020). Artificial Intelligence for
the Internet of Things and Enhanced Medical Systems. Studies in Computational Intelligence,
2021(903), 43–59.
32. Gravina, R., Alinia, P., Ghasemzadeh, H., & Fortino, G. (2017). Multi-sensor fusion in body
sensor networks: State-of-the-art and research challenges. Information Fusion, 35, 68–80.
33. Yadav, A., Kumar Singh, V., Kumar Bhoi, A., Marques, G., Garcia-Zapirain, B., & de la Torre
Díez, I. (2020). Wireless body area networks: UWB wearable textile antenna for telemedicine
and mobile health systems. Micromachines, 11(6), 558.
34. Sodhro, A. H., Zongwei, L., Pirbhulal, S., Sangaiah, A. K., Lohano, S., & Sodhro, G. H. (2020).
Power-management strategies for medical information transmission in wireless body sensor
networks. IEEE Consumer Electronics Magazine, 9(2), 47–51.
35. Popoola, S. I., Popoola, O. A., Oluwaranti, A. I., Atayero, A. A., Badejo, J. A., & Misra, S. (2017,
October). A cloud-based intelligent toll collection system for smart cities. In International
Conference on Next-generation Computing Technologies (pp. 653–663). Springer, Singapore.
36. Awotunde, J. B., Adeniyi, A. E., Ogundokun, R. O., Ajamu, G. J., & Adebayo, P. O.
(2021). MIoT-based big data analytics architecture, opportunities and challenges for enhanced
telemedicine systems. Studies in Fuzziness and Soft Computing, 410, 199–220.
37. Samuel, V., Adewumi, A., Dada, B., Omoregbe, N., Misra, S., & Odusami, M. (2019).
Design and development of a cloud-based electronic medical records (EMR) system. In Data,
Engineering, and Applications (pp. 25–31). Springer, Singapore.
38. Venugopal, K. R., & Kumaraswamy, M. (2020). An introduction to QoS in wireless sensor
networks. In QoS Routing Algorithms for Wireless Sensor Networks (pp. 1–21). Springer,
Singapore.
39. Fortino, G., Galzarano, S., Gravina, R., & Li, W. (2015). A framework for collaborative
computing and multi-sensor data fusion in body sensor networks. Information Fusion, 22,
50–70.
40. Kumar Behera, R., Kumar Rath, S., Misra, S., Damaševičius, R., & Maskeliūnas, R. (2019).
Distributed centrality analysis of social network data using MapReduce. Algorithms, 12(8),
161.
41. Ajayi, P., Omoregbe, N., Misra, S., & Adeloye, D. (2017). Evaluation of a cloud-based health
information system. In Innovation and Interdisciplinary Solutions for Underserved Areas
(pp. 165–176). Springer, Cham.
42. Panigrahy, S. K., Dash, B. P., Korra, S. B., Turuk, A. K., & Jena, S. K. (2019). Comparative
study of ECG-based key agreement schemes in wireless body sensor networks. Recent Findings
in Intelligent Computing Techniques (pp. 151–161). Springer, Singapore.
43. Velez, F. J., Chávez-Santiago, R., Borges, L. M., Barroca, N., Balasingham, I., & Derogarian,
F. (2019). Scenarios and applications for wearable technologies and WBSNs with energy
harvesting. Wearable Technologies and Wireless Body Sensor Networks for Healthcare, 11, 31.
44. Smith, J. R., Joyner, M. J., Curry, T. B., Borlaug, B. A., Keller-Ross, M. L., Van Iterson, E. H.,
& Olson, T. P. (2020). Locomotor muscle group III/IV afferents constrain stroke volume and
contribute to exercise intolerance in human heart failure. The Journal of Physiology, 598(23),
5379–5390.
45. Roth, G. A., Johnson, C., Abajobir, A., Abd-Allah, F., Abera, S. F., Abyu, G., Ahmed, M.,
Aksut, B., & Alla, F. (2017). Global, regional, and national burden of cardiovascular diseases
for 10 causes, 1990 to 2015. Journal of the American College of Cardiology, 70(1), 1–25.
46. Björnson, E., Borén, J., & Mardinoglu, A. (2016). Personalized cardiovascular disease predic-
tion and treatment—A review of existing strategies and novel systems medicine tools. Frontiers
in Physiology, 7, 2.
47. Yasnitsky, L. N., Dumler, A. A., Poleshchuk, A. N., Bogdanov, C. V., & Cherepanov, F. M.
(2015). Artificial neural networks for obtaining new medical knowledge: Diagnostics and
prediction of cardiovascular disease progression. Biology and Medicine (Aligarh), 7(2), BM.
48. Alamri, A. (2019, January). Big data with integrated cloud computing for prediction of health
conditions. In 2019 International Conference on Platform Technology and Service (PlatCon)
(pp. 1–6). IEEE.
222 J. B. Awotunde et al.

49. Bhatia, M., & Sood, S. K. (2017). A comprehensive health assessment framework to facilitate
IoT-assisted smart workouts: A predictive healthcare perspective. Computers in Industry, 92,
50–66.
50. Kalantari, A., Kamsin, A., Shamshirband, S., Gani, A., Alinejad-Rokny, H., & Chronopoulos,
A. T. (2018). Computational intelligence approaches for classification of medical data: State-
of-the-art, future challenges, and research directions. Neurocomputing, 276, 2–22.
51. De Rosa, R., Palmerini, T., De Servi, S., Belmonte, M., Crimi, G., Cornara, S., Maffeo, D.,
Toso, A., & Bartorelli, A. (2018). High on-treatment platelet reactivity and outcome in elderly
with non-ST-segment elevation acute coronary syndrome-Insight from the GEPRESS study.
International Journal of Cardiology, 259, 20–25.
52. Aggarwal, A., Srivastava, S., & Velmurugan, M. (2016). Newer perspectives of coronary artery
disease in young. World journal of cardiology, 8(12), 728.
53. Mendis, S., Puska, P., Norrving, B., & World Health Organization. (2011). Global atlas on
cardiovascular disease prevention and control. World Health Organization.
54. Detrano, R., Janosi, A., Steinbrunn, W., Pfisterer, M., Schmid, J., Sandhu, S., Guppy, K., Lee,
S., & Froelicher, V. (1989). International application of a new probability algorithm for the
diagnosis of coronary artery disease. American Journal of Cardiology, 64, 304–310.
Integration of Machine Learning and IoT
in Healthcare Domain

Ananya Chattopadhyay, Sushruta Mishra, and Alfonso González-Briones

Abstract Recent advancement in technology has been fruitful in the unification

of Internet of things and machine learning in numerous domains. While IoT is
involved in aggregation of data and resources, data analysis, expansion of data,
learning and taking decision action on input data is taken care of by machine learning
approach. The combination of these two technologies can be of great help in clinical
sector in generation of a receptive and interrelated environment, thereby providing
various services to healthcare staffs and patients. IoT devices are based on developing
smart applications for medical usability like wearable modules, smart capsules, and
sensory-based units to assist medical personnel in gathering data. Models based on
machine learning are used to analyze and detect several variations in health status
of a patient, suggest diagnosis methods and alternatives, thereby improving patient’s
health outcome. This chapter discusses the role of machine learning as well as IoT in
healthcare domain. Feasible and ongoing vital applications of both machine learning
and IoT are highlighted. A framework for medical IoT along with possible solu-
tions for medical IoT is presented. Future trends related to the integration of these
two approaches are pointed out. Later an intelligent and smart prototype model for
disease identification is discussed.

A. Chattopadhyay · S. Mishra (B)

School of Computer Engineering, KIIT Deemed To Be University, Bhubaneswar, India
A. Chattopadhyay
e-mail: [email protected]
A. González-Briones
Research Group on Agent-Based, Social and Interdisciplinary Applications (GRASIA),
Complutense University of Madrid, 28040 Madrid, Spain
BISITE Research Group, University of Salamanca, Calle Espejo s/n. Edificio Multiusos I+D+i,
37007 Salamanca, Spain
Air Institute, IoT Digital Innovation Hub, Calle Segunda 4, 37188 Salamanca, Spain
A. González-Briones
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 223
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_11
224 A. Chattopadhyay et al.

Keywords Internet of things (IoT) · Machine learning · Outbreak prediction ·

Disease diagnosis · Classification accuracy

1 Healthcare Viewpoint

Health care constitutes an important part of life. Health care primarily is the suste-
nance and enhancement of good health conditions by the diagnosis of various types
of diseases and preventing their spread. There are three key points in determining if a
healthcare solution is feasible for the society or not quality of health care offered, its
value or cost, and the necessary outcome. Diagnostic approaches like CT, MRI, PET,
and SPECT detect any ruptures or abnormalities beneath the skin layer. Around the
globe, healthcare specialists and stakeholders focus on innovative ways to deliver
this target. Emergency health issues like heart attack or epilepsy can be detected
today even before they occur in a patient’s body. As a result of the population explo-
sion and unpredictable spread of communicable and chronic diseases have stated
depending on current day’s healthcare systems increasing the demand of doctors
and medical workers and medical resources like hospital beds and oxygen masks.
Health care also contributes to be the top earner of revenue nations which delivering
promising results and facilities in disease diagnosis. Enormous time and resources
are lost in the world’s leading healthcare systems due to misdiagnosis. Additional
tests, delayed results in treatment plans create havoc in smooth flow of health care
of an individuals and groups which lead to wastage of healthcare resources at a large
scale worldwide. Thus, the healthcare specialists should come up with a solution
which would considerably reduce the strain on current human-operated healthcare
systems and also the maintain quality and effectiveness of the health care offered
at its required level. This in turn would reduce cost of hospital administration and
supply of medical items considerably.

1.1 Machine Learning in Health Care

In order to deal with colossal amounts of data faster and accurate, medical profes-
sionals or doctors take aid from AI and ML concepts. Applications of machine
learning in health care involve suggesting suitable outcomes based on the analysis
of thousands of different data points. It also helps to provide timely risk scores,
precise resource allocation, etc. This has amplified public health and safety world-
wide. Machine learning is a field which is leveled out of artificial intelligence (AI).
Artificial intelligence involves creation of machines which are better and clever. In
machine learning the sole motive is to gain from patterns and experiences, but not
being altered expressively. The data is entered to the general algorithm, instead of
writing code, and logic is built based on the entered data. The big packets of data are
analyzed with ML algorithms. This simplifies the jobs of data scientists. Since the
process is automated, it gets equal relevance and recognition as fields like big data
Integration of Machine Learning and IoT in Healthcare Domain 225

and cloud computing. Other applications of ML include Web searches, placement

of ads, and also filtering of spam. Most of the machine learning algorithms which
old and traditional are designed for completely fed data into the memory. A number
of enormous and highly complex data-intensive fields such as medicine, astronomy,
biology, and so on have adopted machine learning techniques to provide possible
mining solutions for the information carried by the data common benefits offered by
machine learning in healthcare domain are noted as follows:
• Collection of data
• Identification of diseases and their diagnosis
• Manufacture and discovery of drugs and medicines
• Medical imaging
• Assessment of behavioral modification using ML
• Well-maintained health records.

1.2 IoT in Health Care

IoT is well-developed and integrated within multiple organizations, including the

healthcare fields, thus coming a long road. We are moving toward a completely
new age of patient health monitoring, and management with digital health solutions
coming in the picture. Global Data believes that the IoT will expedite this process
of technology developments with respect to application of medical devices, thus
improving the outcomes with better analysis of patient’s health condition. As IOT is
kept implementing in health care, it is expected that the productivity of the data and
also its analysis will increase drastically. The Internet of Healthcare Things (IoHT) is
a concept used in the medical area that is based on Internet accessible devices having
unique identification. The devices also facilitate communicate among each other for
proper execution of medical procedures, thus enabling remote localization and actual-
time information extraction. It is also possible to manage resources automatically.
Most importantly, the IoHT solutions are very flexible and personalized. Not only
does this enable care of better quality in less time, but also promises safety of the
patients. The size or specialization of the system does not matter at all. They can be
adapted to everywhere if the facility is well-modified connected and integrated with
primary technologies and software.
Managing medical facilities is more efficient with uninterrupted access to medical
equipment’s, required data and patient’s medical information. Some common utilities
of IoT in health care are listed here.
• Instantaneous notification of problems, by the patient/device
• Continuous assessment/monitoring based on health parameters
• Reminders given automatically regarding medicines, etc
• Enhanced safety by remote medical care and consultations
• Detection of dangerous health issues prior to their occurrence
• The transfer and analysis of data automatically.
226 A. Chattopadhyay et al.

2 Renowned Machine Learning Application in the Field

of Health Care

Since the healthcare system business of all people, so finding a way around is really
necessary. And, it is not an easy task to keep the all details in the mind as mental
and physical performances of human beings have limitations. IoT and AI concepts
are really helpful in carrying out healthcare operations which are humanely difficult.
Regardless in what fields the technologies are applied, IoT unified with AI are strong
driving forces involving transformation digitally.

2.1 Identifying Disease and Diagnosis

ML models are capable of predicting disease susceptibility, thus helping in early diag-
nosis of diseases. With population explosion and increased life expectancy decreasing
the mortality rate, more pressure is being put to healthcare systems due to which the
systems not being well equipped for the day-to-day diagnosis due to shortage of
resources. Feebris (a UK-based technology start-up) uses algorithms which are AI
based for the accurate detection of conditions of respiratory system that are complex
to identify. It is a technique that involves connecting to medical sensors. It can be
used by individuals for detection of early respiratory issues, thus avoiding respi-
ratory complications and untimely hospitalizations. The Artificial Intelligence and
Computer Science Laboratory of MIT is known to have created a new deep learning-
based prediction model. This model is believed to detect any sign of breast-related
cancer up to five years before its occurrence in the body thus being an absolute game
changer in the field of medical science. The model was implemented on mammo-
grams, and the follow-up data of patients were collected. Various patterns were
identified that would not even be observed or analyzed by doctors. In a predictive or
pre-diagnosis-based discovery, the results obtained via the model can be relied upon.
Process illustrated in Fig. 1.

2.2 Machine Learning in Radiology

Researchers have been lately working to implement ML and AI in the field of radi-
ology. In order to accelerate the process of detection using machine learning algo-
rithms, it provides software for the radiologist. The medical image obtained as a
result is used for the detection of abnormalities in the body. The algorithm mostly
used is supervised machine learning algorithm. The process of identifying struc-
tures in an image is called segmentation, and ML techniques are used for the same.
Primarily, the graph cut segmentation method is used mostly for segmentation. For
the analysis for radiology text reports, natural language processing is used. Thus,
Integration of Machine Learning and IoT in Healthcare Domain 227

Fig. 1 A disease diagnosis workflow using machine learning

machine learning can improve patient healthcare services in the field of radiology as
demonstrated in Fig. 2.

2.3 Clinical Trial and Research

A set of queries that require answers to obtain the efficiency and overall working of
biomedical or pharmaceutical procedures comprises clinical trials. The sole purpose
228 A. Chattopadhyay et al.

Fig. 2 Machine learning application in radiology

of this trial is to focus on the development and implementation of treatments. These

clinical trials on normal basis involve a lot of money and time. An ML-equipped
system surely can help in providing actual-time monitoring and robust and reliable
service making a significant impact in the trials. The significant benefit is that it can
be monitored remotely and at the same time providing a safe ambience for patient
trials. Also, the efficiency of the clinical trial is enhanced by the supervised machine
learning in this field (Fig. 3).

2.4 Outbreak Prediction

Around the world AI and machine leaning-based technologies have a firm grip in the
prediction of epidemics. Vast data is acquired from social media updates, by people
in real time, satellites, Web sites, etc., are accessed by scientists and researchers
for work in their fields. The collation of this information thus obtained is done by
the artificial neural networks smoothly predict everything from epidemic outbreaks
to infectious and chronic diseases. In the third-world developing countries lacking
adequate medical resources, infrastructure and facilities prediction of these disease
outbreaks come with a lot of help. The ProMED-mail is a good example in this
respect. It is a reporting platform (Internet-based) which monitors diseases, which
can emerge as well as evolve over a certain time period, thus providing disease
outbreak reports in actual time.
Integration of Machine Learning and IoT in Healthcare Domain 229

Fig. 3 Clinical trial approach using ML

Fig. 4 Outbreak prediction based on date and location

3 Internet of Things (IoT) Applications in Clinical Domain

3.1 Depression Monitoring Apple Watch App

In November 2017, at pharma and biotech conference (CNS Summit), the results
from the study were presented and the people found a good amount of assent with
application. The people used the application in daily basis for monitoring their cogni-
tive memory and mood shifts. The day-wise assessments corresponded with objective
230 A. Chattopadhyay et al.

cognition tests and outcomes reported by patients, thus demonstrating the applica-
tion to be robust and reliable. A collaboration was done by Takeda Pharmaceuticals
U.S.A. and Cognition Kit Limited in the year 2017, for exploring the use of an Apple
Watch application which can measure cognitive health of a group of patients. Their
intention was to monitor and assess people with major depressive disorder (MDD)
and a perfect use of wearable technology in the mental health field.
The Apple Watch application thus offers healthcare professionals and patients
insight and conclusion into the mental health conditions, just like smart medical
devices that collect patient data. This application has clearly demonstrated that wear-
able technology certainly has the potential to assess depression and anxiety conditions
in an individual’s daily life. Thus, it is a smooth way of enable good care.

3.2 Coagulation Testing

Roche, in the year 2016, created a device to check how rapidly clotting of blood
occurs in patients. The equipment constituted of a Bluetooth-enabled coagulation
system. This Bluetooth-enabled coagulation system is the first of its kind, with self-
testing shown. This comes as a help to patients. When results can be inferred by
healthcare providers with very few clinical visits, it definitely is a great benefit for
patients. The results within the therapeutic range indicate that the patient has lowered
the risk of bleeding or heart attack. Some notifiable features of this device are:
• The patients can add comments to their results.
• The application also reminds the patients to get themselves tested.
• In relation to the range of the target, the results are flagged by the device. Process
demonstrated in Fig. 5.

Fig. 5 Device working for coagulation testing in patients

Integration of Machine Learning and IoT in Healthcare Domain 231

3.3 Medical Information Distribution

One of the most challenging works, the medical field has to carry out is to distribute
accurate and up to date information to the patients about their health issues. Not only
do IoT devices improve health care of individuals but also facilitate the professional
practice of medical workers. A remarkable innovation of IoT in the healthcare field
is the distribution of medical information.
IoT facilitates intrusive care into the social space like home or office, thus enabling
systems to take health care out of facilities like hospitals. Individuals thus can to cater
to their own health conditions. Consequently, this has paved the path for smooth flow
of health care, resulting in fewer accidents particularly from miscommunication, thus
increasing patient satisfaction and preventive care in a large scale as shown in Fig. 2.
Also healthcare providers (medical workers and doctors) can deliver better care to
patients (Fig. 6).

Fig. 6 Medical information distribution via IOT applications in the healthcare field
232 A. Chattopadhyay et al.

3.4 Emergency Care

One of the problems that the emergency support systems have always countered is
the limitation of the medical resources and disconnection with the facilities of the
base. IoT applications having advanced automation and analytics provide a number
of effective solutions to these problems. The doctors or medical workers can get easy
access to the patient profiles and can execute proper diagnosis and care to the patients
in due time, way before their arrival to the clinic or hospital. Thus if any emergency
condition takes place, the necessary steps can be implemented or analyzed from a far
distance. On a large scale, associated losses are thus reduced in this way consequently
enhancing emergency health care (Fig. 7).
As the days go by, Internet of things and machine learning together will have a
significant role in health care as both technologies are currently in a rapid state of
development.

Fig. 7 Emergency support with IOT

Integration of Machine Learning and IoT in Healthcare Domain 233

4 A General Architecture for IoMT Systems

Field sensor networks which can be attached to the human body are important
constituents of remote monitoring solutions. Fog/edge computing architectures are
used, thus leading the way to protective, responsive, and preventive systems. In
majority of cases, fog nodes act as local servers. The work is collection and processing
of health-related information to swiftly respond to the requirements. For more than
ten years, the scientific community is involved in investigating and discovering effec-
tive solutions which can monitor the health condition of the patients with the vision
of providing appropriate reports to doctors and remotely. The early methods include
PC-based systems for monitoring; for example, high heart rate and filtering of critical
situations of patients were solely done by data derived from ECG and accelerome-
ters. Clinicians were advised based on that. The successive implementations included
single-board computers and microcontrollers which involved in gathering physiolog-
ical data of patients from sensors that are specialized including—electrocardiograph,
electronic-health sensor system. In a PC-based environment, this was further to be
processed and analyzed. IoT technologies have thus lead to several reliable and smart
solutions in both software platforms and network architecture fields and domains.
With such solutions, health care can be addressed at different levels: chronic
disease supervision, childcare and old age solutions, epidemiology, private health-
care, computer monitored algorithm-based CPS, occupational therapy related to
health and wellness, thus leading to great ease of systems.
Systems that address health and wellness-related problems can also be divided
into two types: static remote monitoring—assumes hospital or house-based patients
and dynamic form of monitoring—considers the patient as a moving unit (not in a
closed space like a home or a hospital) and monitors it.
A basic solution involving a multilevel architecture is shown in Fig. 8 which
consists of three levels:
• EDGE—Preprocessing is performed by devices like cell phones, electronic smart-
watches, compact embedded systems, and compact gateways. They also perform
data validations at lower levels and subsequent collection from wireless body
sensor networks (WBSN).
• FOG—Data is collected from field sensor networks and edge devices by PCs
servers/gateways. Processing at local level or storage is done.
• CLOUD—It involves cloud computing services for high-level computing tasks.
Cloud services also help in remote data storage.
It is not mandatory to implement all the three levels. For instance, in the static
monitoring problems, fog devices can directly collect data from sensors and with
support from the cloud services if required. Similarly, in some conditions of dynamic
monitoring there would be direct interaction of edge devices with cloud services,
where implementation of fog level could not be done.
234 A. Chattopadhyay et al.

Fig. 8 Health care addressed using a multilevel architecture (fog, edge, and cloud systems)

5 Various Extensive Studies Conducted

In the healthcare field, IOT-based devices are changing the lifestyles of people. The
devices have great contribution in monitoring, analysis, diagnosis, of a variety of
health conditions [1–11]. A review of the extensive use of IoT-based devices is
presented in this section [12]. A medical sensor which can be attached to the body
and consisting of vast uses and applications has been proposed by some technicians.
Some have also proposed a system developed by fuzzy logic and wireless sensor
network (WSN) which gives the interface [13]. With advanced microelectromechan-
ical systems (MEMS) concept, the wireless sensor network can make a body sensor
network (BSN) which can monitor unusual changes in the health parameters of the
patients. The clinical data will be collected from the patients, and storage is done
in the cloud database. The accuracy is improved by transferring the data to the
fuzzy logic controller [14]. In another Zigbee technology-based unique healthcare
monitoring system, the clinical data of patients obtained from the sensors of various
parameters can be monitored remotely. The standard IEEE 802.15.4 Zigbee protocol
can be used for the measurement and monitoring of different medical parameters
like ECG, the rate of the heart, functioning of lungs (spirometer), and temperature
[15]. Information is obtained by the means of radio frequency, and the data gets
displayed on devices like mobiles computers. Today personalized diagnosis of the
Integration of Machine Learning and IoT in Healthcare Domain 235

patient can be performed by just a sensor which is wearable and attached to the
body. A process of tele monitoring can also be used with the help of wireless body
area network (WBAN) which is capable of providing home-based health care [16].
In the WBAN sensor, the physiological data is collected, i.e., EEG, EKG, and BP,
and then the monitoring and analysis are done by the application. A health moni-
toring system can also be developed by using progress in embedded computing
technologies and wireless sensor network (WSN) [17]. The patients’ physiological
parameters are continuously monitored and analyzed in the system [18]. Platforms
that can monitor intake of food send nutritional information regarding intake of food
have also been developed. IoT-based healthcare devices can also identify correlations
between possibilities of getting certain diseases. Intercommunicated sensors can be
used for monitoring the activities of people dealing with obesity to assist them get a
proper health by maintaining a proper diet plan [19]. Lee and Ouyang [20] involve
the presentation of a study that deals with identifications of correlations between the
possibilities of getting certain diseases in the body and the uses clinical devices in the
context of the IoT. By contrast, [21] deals with presentation of a platform that uses
intercommunicated sensors for monitoring activities of obese children. A collabo-
ration protocol can help to transmit the risk notifications to the devices like smart
phones [22]. This can be done with the help of a new service application algorithm
which can be added in the devices of patients dealing with obesity conditions, high
blood sugar level, and heart issues [23]. Wearable IoT (WIoT) can give different path-
ways regarding clinical and operative methods in terms of applications, functions,
and design. An application for treatment of obesity conditions can be developed in a
mobile through machine-to-machine (M2M) exchange of information, to achieve a
healthy diet and monitor weight [24]. In [25], a web page and a mobile application
are proposed for causing reduction of weight through machine-to-machine (M2M)
exchange of information exchange or communication, in which a specific proportion
of weights was used to achieve a healthy diet. Physiological data can be obtained
from devices linked to triaxial accelerometers and a heart rate monitor, and physical
activity can be detected [26]. The performance of the classifiers was evaluated with
respect to the patients’ physical activities. Additionally, [27] defined a new concept
of IoT-learning, with which a health application was developed using a combination
of the IoT and architecture supported by the IoT. In addition, [28] described future
trends in wireless communication with a focus on 5G networks, in which the benefits
for the IoT and e-health are notable. Further, [29] presented a study focusing on the
use of IoT in e-health, along with future guidelines and the evolution of the IoT in
the field of health. Various studies are thus conducted across the world focusing on
the use of IoT in e-health [30, 31]. Trends in wireless communication with emphasis
on 5G networks in the future for benefits of IoT applications and implementations
in the field of e-health are notable.
236 A. Chattopadhyay et al.

6 Review of IoMT Monitoring Solutions

A lot of contributions based on IoT architectures have been suggested and proposed
for smart health monitoring in less than ten years.

6.1 Physiological Analysis

Protective approaches and systems have been developed based on different parame-
ters which are physiological health indicators and critical accidental circumstances
have been assessed [32]. The heart-beating rate and rate of motion of people are moni-
tored by a WBSN user within their homes. Early detection of dip in values that lead
to tachycardia or bradycardia is indicated. Whenever rapid changes in values which
were measured occur, the edge node through the Internet helps to send an immediate
alert to members of the family or doctors via cell phone. Similarly, Villarrubia et al
a system has been proposed where an ECG data analysis is done for cardiovascular
monitoring at home. By a common TV interface, the patients perform interaction
with the system at ease. Cloud technology provided by Bluemix enables physio-
logical data storage consequently allowing access remotely to medical practitioners
as well as result visualization and analysis through IOT system of IBM Watson.
Fever can also be diagnosed by using self-embedded monitoring system where the
patient’s temperatures are recorded. In a real-time ECG telemetry based on IOT on
an android smart phone, the quality assessment algorithms are directly implemented,
thus allowing evaluation in a real time. Under different physical activities, the effec-
tiveness of the method is demonstrated by the authors. The use of field sensors for
gathering of contextual data for multilevel recognition tasks can be leveraged by
static monitoring. Video, any movement, and audio signals are collected with the
help of environment sensing devices, an optitrack camera, and sensors containing
smartwatches along with specially modified wearables for gathering of various phys-
iological data. Actually, it is a fog system to cloud system architecture in cooperation,
where a home gateway does the tasks involving the preprocessing of data, localiza-
tion of indoor activities, and related recognition algorithms. The data is stored by a
private cloud data for remote accessing [33].
In gathering of physiological parameters and analysis from sensors attached to
human body, recent proposals start diving into the use of edge ML systems. In
order to cater to the necessities of compression of data, extraction of features and
grouping and also comparison of the precision and correctness of some traditional
classifiers (Naive Bayes, random forests, and classification/regression trees, k-nearest
neighbors,), researchers have individuated the main functions at the edge side of
ML [29]. An approach for classification of deviations in the ECG signals is the
proposal Azimi et al. A variant of the MAPE-K model is implemented by hierarchical
computing architecture for healthcare (HiCH) (solution name) as introduced by IBM.
The computations are divided throughout the three systems, namely edge, fog, and
Integration of Machine Learning and IoT in Healthcare Domain 237

cloud. The proposed model comprises of main four computing parts: the monitor
element that creates a connection between the sensors and the edge. Here prior
processing of the data, its aggregation, and its storage is done. Next is the analyze
element that has cloud machines to carry out heavy computation tasks. The third is
the plan element which is at the edge system. The analyze component periodically
updates it. The whole system becomes highly available by execution of model and
providing a decision which is localized. In both the cases, the methodology has been
verified. The first: partitioning of a linear machine learning method (linear SVM)
and second: deploying of a deep learning (nonlinear) algorithm.

6.2 IoMT Solutions in Rehabilitation Systems

After an operation or a surgery in order to make sure that the patient body is free from
any infection or any other anomaly IoMT solutions are quite effective and protective.
By just the monitoring of the temperature and walking pattern health condition of
the residual limb in the lower limb amputee can be assessed. A combined set of edge
devices which includes a mobile phone with android setup captures and feeds the
data to a fog station which based on various parameters gives the prediction based
on ML [30]. In the same way, estimation of human limb movements can be carried
out by simply two accelerometers (low power) attached to the fore arm.
A fog system architecture can be used to create a system that can monitor the
speech of an individual which can be effectively used for patients with Parkinson’s
disease. Audio signals are detected by a microphone or a cellphone and sent to a fog
node for acoustic feature extraction. Classification will occur in the cloud system.

6.3 Assessing of Diet Intake and Skin Pathology

By implementing deep learning techniques in mobiles, we can carry out DL inference

directly. Without the support of the cloud system, classifications of various forms of
lesions in skin can be performed through a CNN model running in a smart phone.
Assessment of diet can also be done by visual food recognition dietary assessment.
The image is processing and segmented and then executed on device while the CNN
classification takes place in the cloud.
238 A. Chattopadhyay et al.

6.4 Treatments Pertaining to the Spread of Epidemics

and Their Diagnosis

In diagnosis of epidemics, IoMT systems are effective solutions. In these contexts,

actual-time operation, mobility, and effective fusion of data from different environ-
mental sensors, location sensors or bio-sensors are a concern. The virus infections
can be treated or diagnosed in an initial state, and rapid recovery can be enabled by
such IoMT systems. The outbreak of Chikungunya can be prevented, and it can be
diagnosed as discussed [34]. The patient’s symptoms and environmental conditions
nearby can be analyzed by a fog system implementation to understand the presence
of the virus and its diagnosis. The infected and risk-prone zones can also be inferred
to the user through alerts. Zika virus infection can also be prevented disease by an
MCC paradigm. The sole motive is to treat the infected people representing them
in Google maps and also places which densely breed in mosquitoes. This can assist
the pandemic health care to control the zones which are prone to the spread of the
disease. In the above cases, a fog-based system is needed to analyze the data and
processes it while cloud services are involved in storing the data, its deep analysis,
and its access remotely. The COVID-19 pandemic has raised a major concern for
evaluation by screening devices, in untouchable conditions, for primary symptoms
like fever, headache and cyanosis. The temperature of the forehead can be estimated
by a camera which is infrared, while cyanosis can be determined from lips in the
visible spectrum just by running DL algorithms at the edge system. The multi-modal
artificial intelligence approaches on smart phone embedded sensors readings can
also be used to predict COVID-19.

6.5 Diagnosis and Treatment of Diabetes

In the treatment of diabetes, many IoMT solutions are appearing. The blood glucose
monitors, insulin pens, insulin pumps, and closed-loop artificial pancreas systems
are portable and wearable devices that can communicate with the smart phones
wirelessly. The smart phones can be considered as edge node systems which can
do basic analysis without the support of the cloud. Prediction of the onset of type-2
diabetes can be done by a J48Graft decision tree classifier which can be directly
implemented on a smart phone [35–37].

7 Trends and Discussions About Applications

The above-mentioned applications have justified that various AI and ML-based tech-
niques are highly able in decision making, detection of anomalies, monitoring of
predictive risk, and so on. Due to the limitation of resources on low-power devices
Integration of Machine Learning and IoT in Healthcare Domain 239

in the edge system, in most of the cases deep analysis functions are carried out in the
cloud system. However, the cooperation among edge-based, fog-based, and cloud-
based systems functionally can be actually observed. A large number of applications
based on machine learning/deep learning techniques such as predicting or recog-
nizing activities which are multi-modal, monitoring prior the incident (detection of
fall in value and preventing the condition to occur), medical automation in disease
treatment and care in emergency situations require the deduction to be carried out
as close to the sensor nodes as possible. This is because it lowers the latency and
reduces consumption of energy and the requirement of real-time operation is met.
Fog system architectures which have GPU enhanced servers locally seem to be the
widely embraced solution for dealing with machine learning deduction tasks with
good demand in the static monitoring systems. To address dynamic monitoring prob-
lems in which a computer node having a small size and low power, other strategies by
which the functionality can be optimized come into play. As on one side, edge stream-
computing methods between various edge system peers which focus at parallelism
the deduction process are a tangible answer while on the other side, deep learning
lightweight models which are embraced for carrying out deduction processes on
devices which are embedded also gives good results.

8 A Smart Predictive Framework for Disease Risk Factors

Detection

A significant part of the healthcare costs round the globe includes diagnosis of
chronic diseases which need lifelong treatment of patient. In the forecast and diag-
nosing of these diseases, various models of predictive framework are often applied
in the present times [38–42]. Machine learning models serve to enhance the quality
of medical-based data, thus reducing variations in patient rates. This in turn saves
medical costs. Therefore, compared to all other conventional methods, these models
are often used to for investigation and analysis. Therefore, a decision model which can
prove beneficial for diagnosis of long-term diseases (chronic) and forecast the future
outcomes of patients is important to design and develop. Also, automated data collec-
tion using bod sensors can be achieved with little effort with smart IoT frameworks
and also on a timely basis. An integrated machine learning-based IoT framework
design is presented in this section that can facilitate clinical experts in effective diag-
nosis. According to the figure below, the patient’s body is embedded with different
sensors like BP and ECG temperature sensors. The sensors are responsible for the
recording of various health parameters readings. For further analysis and processing,
sensory readings are collected, aggregated values are made and finally transmitted
to the microcontroller unit. Using appropriate standardization techniques, the data
is further preprocessed if any ambiguity or flaw is detected in the required health
parameters. The inconsistencies of the unstructured data fed are removes here, and
the data is normalized. Lastly, the resultant output of this is directed into the machine
240 A. Chattopadhyay et al.

Fig. 9 Predictive framework for detection of chronic diseases

learning unit. Using a training set of data here, a suitable machine learning algorithm
is trained. The algorithm can successfully predict unknown test data objects and clas-
sify into the specific class label, once the algorithm gets trained. Thus, the degree
of disease risk disorders of a patient can be predicted. Then subsequent results of
this implementation can be validated and evaluated with the help of certain suitable
evaluation metrics. The predictive framework is shown in Fig. 9.
In the aforementioned system, simultaneous monitoring of multi-parameters inte-
grated with single chip onto wearable devices. Multiple sensors work together to
obtain required data eliminating noise disturbances. By reducing less human inter-
ventions, it maintains its accuracy. For remotely located heart-prone patients having
remote locations, this is effective to measure, body temperature, heart rate, blood
pressure, and many other parameters with the use of biomedical sensors.
With the approach of machine learning, the trained data will be sensitized and
analyzed the understanding heart efficiency of the patient. The transfer the infor-
mation between the doctor and patient will be enabled by two major interfaces.
Underlying health conditions in real time can be detected by this framework, and an
alarm is generated in the form of SMS, email, etc., based on the threshold values
which the doctor had set and also for a family member who is who is registered.
The healthcare system thus enriches feeling of care or attention for the patient, thus
adding health value. This invention considerably reduces the cost, and one does not
Integration of Machine Learning and IoT in Healthcare Domain 241

95
90

Classification Accuracy
85
80
75
70
Naive Decision
Rate (%) RNN KNN SVM
Bayes Tree
Diabetes 92.5 87.4 84.9 90.9 78.4
Heart 90.4 87.9 82.2 92.5 82.7
Kidney 91.3 90.5 86.7 89.5 85.9
Hepatitis 90.4 91.7 88.8 90.4 89.3
Lung 91.1 92.5 87.4 84.9 90.5
Brain 87.7 89.2 90.5 86.7 89.5

Fig. 10 Classification accuracy rate

have to purchase different devices at different prices. This being a onetime invest-
ment, hospitalization costs get reduced to a great extent. The accuracy rate of various
classifier algorithms using the discussed model is depicted below (Fig. 10).

9 Summary

Better information regarding patient care can thus be inferred to doctors using these
advanced analytics. Easy accessibility to the blood pressure readings and other vital
information of patients is routine and expected by doctors and clinicians. Based on the
latest reading of blood pressure, the laboratory test results, gender, family history,
and latest clinical trial data, a patient’s risk of getting for heart attack, coronary
artery disease or failure of the kidney can be assessed. Clinicians require advanced
information so that they are successful in making better decisions about patient
diagnosis and treatment. The should also while understand the possible outcomes
in regard to the solutions. In the domain of health care, through machine learning
large datasets can be processed for diagnosis which is far from the reaches of human
capability. It has been aiding doctors and clinicians in planning diagnosis methods
and providing care by transforming the analysis of the data to clinical insights. This
has in turn leaded to cheap and effective health care, increased patient satisfaction,
and better outcomes.
The best machine learning tool health care is the brain of the doctor. If we consider
that IoT is as the digital nervous system (receiver of signals), then it would be
justifying to say ML as its medulla oblongata. Without implementing ML, it would
be a difficult job for IOT applied smart devices to make smart and clever decisions
in real time, without implementation of machine learning.
Autoworkers once feared that robotics would lead ultimately curb their daily
working and jobs. In the same way, clinicians and doctors may fear that they might be
rendered obsolete by the onset of machine learning, but the contribution of medicine
is irreplaceable. However, the caring and compassionate relationship of healthcare
242 A. Chattopadhyay et al.

professionals is always valued by the patients. Machine learning, or any other futur-
istic technologies supposedly, will not interfere. The technologies will become clin-
ical tools that can be implemented to enhance health care. In order to assist the
pathologist, an ML algorithm with a proper diagnosis is really worth.
If results are acquired in a short period of time, with a similar accuracy and correct-
ness, then, undoubtedly, this is beneficial in health care and satisfaction. Health care
needs to implement ML as a real-world tool in the present rather than visualizing it
as a futuristic concept. It is important to find cases in which the capability of machine
learning provides value from an application which is technologically specific like
Google and Stanford. Thus, machine learning and predictive algorithms can be incor-
porated into clinical and diseases diagnostic practices by implementing and following
a step-by-step pathway.

References

1. Roy, G., Bhoi, A. K., & Bhaumik, S. (2021). A comparative approach for MI-based EEG
signals classification using energy, Power and Entropy. IRBM.
2. Nayak, S. R., Sivakumar, S., Bhoi, A. K., Chae, G. S., & Mallick, P. K. (2021). Mixed-
mode database miner classifier: Parallel computation of graphical processing unit mining.
The International Journal of Electrical Engineering & Education, 0020720920988494.
3. Pramanik, M., Pradhan, R., Nandy, P., Bhoi, A. K., & Barsocchi, P. (2021). Machine learning
methods with decision forests for Parkinson’s detection. Applied Sciences, 11(2), 581.
4. Panigrahi, R., Pramanik, M., Chakraborty, U. K., & Bhoi, A. K. (2020). Survivability prediction
of patients suffering hepatocellular carcinoma using diverse classifier ensemble of grafted
decision tree. International Journal of Computer Applications in Technology, 64(4), 349–360.
5. Bhatt, T. V., Patel, R. K., Chitara, H. B., Marques, G., & Bhoi, A. K. (2020). Fuzzy logic
system for diabetic eye morbidity prediction. International Journal of Computer Applications
in Technology, 64(4), 339–348.
6. Yadav, A., Kumar Singh, V., Kumar Bhoi, A., Marques, G., Garcia-Zapirain, B., & de la Torre
Díez, I. (2020). Wireless body area networks: UWB wearable textile antenna for telemedicine
and mobile health systems. Micromachines, 11(6), 558.
7. Marques, G., Bhoi, A. K., Albuquerque, V. H. C. & de, K. S., H. (Eds.) (2021). IoT in healthcare
and ambient assisted living. Springer.
8. Bhoi, A. K., Mallick, P. K., Liu, C. M., & Balas, V. E (Eds.) (2021). Bio-inspired
neurocomputing. Springer.
9. Bhoi, A. K., Sherpa, K. S., & Khandelwal, B. (2018). Arrhythmia and ischemia classification
and clustering using QRS-ST-T (QT) analysis of electrocardiogram. Cluster Computing, 21(1),
1033–1044.
10. Bhoi, A. K., & Sherpa, K. S. (2016). Statistical analysis of QRS-complex to evaluate the
QR versus RS interval alteration during ischemia. Journal of Medical Imaging and Health
Informatics, 6(1), 210–214.
11. Marques, G., Miranda, N., Kumar Bhoi, A., Garcia-Zapirain, B., Hamrioui, S., & de la Torre
Díez, I. (2020). Internet of things and enhanced living environments: Measuring and mapping
air quality using cyber-physical systems and mobile computing technologies. Sensors, 20(3),
720.
12. Mosenia, A., Sur-Kolay, S., Raghunathan, A., & Jha, N. (2017). Wearable medical sensor-based
system design IEEE transactions on MultiScale computing systems 2017 May 20.
13. Patil, P., & Mohsin, S. (2013). Fuzzy logic based health care system using wireless body area
network. International Journal of Computer Applications, 1, 80(12)
Integration of Machine Learning and IoT in Healthcare Domain 243

14. Madhyan, E., & Kadam, M. (2014). A unique health care monitoring system using sensors
and ZigBee technology. International Journal of Advanced Research in Computer Science and
Software Engineering, 4(6).
15. Balasubramanian, A., Wang, J., & Prabhakaran, B. (2016 ). Discovering multidimensional
motifs in physiological signals for personalized healthcare. IEEE Journal of Selected Topics
in Signal Processing, 10(5), 832–841.
16. Kim, Y., Lee, S., & Lee, S. (2016). Coexistence of ZigBee-based WBAN and WiFi for health
tele monitoring systems. IEEE journal of biomedical and health informatics, 20(1), 222–230.
17. Kakde S., et al. (2015). Implementation of health-care monitoring system using raspberry pi.
IEEE ICCSP 2015 Conference, 2, 1083–1086.
18. Vazquez-Briseno, M., Navarro-Cota, C., Nieto-Hipólito, J., Jiménez-García, E., & Sanchez-
Lopez, J. (2012). A proposal for using the internet of things concept to increase children’s health
awareness. In Proceedings of the CONIELECOMP 2012, 22nd International Conference on
Electrical Communications and Computers, Puebla, Mexico, 27–29 February 2012; pp. 168–
172.
19. Vilallonga, R., Lecube, A.; Fort, J. M., Boleko, M. A., Hidalgo, M., & Armengol, M. (2013).
Internet of Things and bariatric surgery follow-up: Comparative study of standard and IoT
follow-up. Minimally Invasive Therapy and Allied Technologies, 22, 304–311. [PubMed]
20. Lee, B. M., & Ouyang, J. (2013). Application protocol adapted to health awareness for smart
healthcare service. Advanced Science and Technology Letters, 43, 101–104.
21. Zaragozá, I., Guixeres, J., Alcañiz, M., Cebolla, A., Saiz, J., & Álvarez, J. (2013). Ubiquitous
monitoring and assessment of childhood obesity. Personal and Ubiquitous Computing, 17,
1147–1157.
22. Lee, B. M., & Ouyang, J. (2014). Intelligent healthcare service by using collaborations between
IoT personal health devices. International Journal of Bio-Science and Bio-Technology, 6, 155–
164.
23. Hiremath, S., Yang, G., & Mankodiya, K. (2014). Wearable internet of things: Concept, archi-
tectural components and promises for person-centered healthcare. In Proceedings of the 4th
International Conference on Wireless Mobile Communication and Healthcare—Transforming
Healthcare Through Innovations in Mobile and Wireless Technologies (MOBIHEALTH),
Athens, Greece, 3–5 November, pp. 304–307.
24. Alloghani, M., Hussain, A., AI-Jumeily, D., Fergus, P., Abuelmatti, O., & Hamden, H. (2016).
A Mobile Health Monitoring Application for Obesity Management and Control Using the
Internet-of-Things. In Proceedings of the 2016 Sixth International Conference on Digital
Information Processing and Communications (ICDIPC), Beirut, Lebanon, 21–23 April 2016,
pp. 19–24.
25. Wibisono, G., & Astawa, I. G. B. (2016). Designing machine-to-machine (M2M) prototype
system for weight loss program for obesity and overweight patients. In Proceedings of the
2016 7th International Conference on Intelligent Systems, Modelling and Simulation (ISMS),
Bangkok, Thailand, 25–27 January 2016, pp. 138–143.
26. Dobbins, C., Rawassizadeh, R., & Momeni, E. (2016). Detecting physical activity within life
logs towards preventing obesity and aiding ambient assisted living. Neurocomputing, 230,
1–23.
27. Shin, S.-A., Lee, N.-Y., & Park, J.-H. (2017). Empirical study of the IoT-learning for obese
patients that require personal training. In J. J. Park, Y. Pan, G. Yi, V. Loia (Eds.) Advances in
Computer Science and Ubiquitous Computing (Vol. 421, pp. 1005–1012). Singapore: Springer.
28. Camara-Brito, J. M. (2016). Trends in wireless communications towards 5G networks—The
influence of e-health and IoT applications. In Proceedings of the International Multidisciplinary
Conference on Computer and Energy Science (SpliTech), Split, Croatia, 13–15 July 2016;
pp. 1–7.
29. Mishra, S., Tripathy, H. K., & Mishra, B. K. (2018). Implementation of biologically motivated
optimisation approach for tumour categorisation. International Journal of Computer Aided
Engineering and Technology, 10(3), 244–256.
244 A. Chattopadhyay et al.

30. Mishra, S., Chaudhury, P., Mishra, B. K., & Tripathy, H. K. (2016, March). An implemen-
tation of feature ranking using machine learning techniques for diabetes disease prediction.
In Proceedings of the Second International Conference on Information and Communication
Technology for Competitive Strategies, pp. 1–3.
31. Ifrim, C., Pintilie, A.-M., Apostol, E., Dobre, C., & Pop, F. (2017). The art of advanced
healthcare applications in big data and IoT systems. In C. X. Mavromoustakis, G. Mastorakis,
C. Dobre, (Eds.) Advances in mobile cloud computing and big data in the 5G Era (Vol. 22,
pp. 133–149). Berlin/Heidelberg, Germany: Springer.
32. Mishra, S., Mallick, P. K., Jena, L., & Chae, G. S. (2020). Optimization of skewed data using
sampling-based pre-processing approach. Frontiers in Public Health, 8, 274. https://doi.org/
10.3389/fpubh.2020.00274
33. Ray, C., Tripathy, H. K., & Mishra S. (2019). A review on facial expression based behavioral
analysis using computational technique for autistic disorder patients. In: M. Singh, P. Gupta,
V. Tyagi, J. Flusser, T. Ören, R. Kashyap (Eds.) Advances in Computing and Data Sciences.
ICACDS 2019. Communications in Computer and Information Science (Vol. 1046). Singapore:
Springer https://doi.org/10.1007/978-981-13-9942-8_43
34. Sahoo, S., Mishra, S., Mishra, B. K. K., & Mishra, M. (2018). Analysis and implementation
of artificial bee colony optimization in constrained optimization problems. In Handbook of
Research on Modeling, Analysis, and Application of Nature-Inspired Metaheuristic Algorithms
(pp. 413–432). IGI Global.
35. Mishra, S., Tripathy, H. K., Mishra, B. K., & Mohapatra, S. K. (2018). A succinct analysis of
applications and services provided by IoT. In Big Data Management and the Internet of Things
for Improved Health Systems (pp. 142–162). IGI Global.
36. Mishra, S., Mallick, P. K., Tripathy, H. K., Bhoi, A. K., & González-Briones, A. (2020). Perfor-
mance evaluation of a proposed machine learning model for chronic disease datasets using an
integrated attribute evaluator and an improved decision tree classifier. Applied Sciences, 10(22),
8137.
37. Mishra, S., Mallick, P. K., Tripathy, H. K., Jena, L., & Chae, G.-S. (2021). Stacked KNN with
hard voting predictive approach to assist hiring process in IT organizations. The International
Journal of Electrical Engineering & Education. https://doi.org/10.1177/0020720921989015
38. Mishra, S., Mishra, B. K., Tripathy, H. K., & Dutta, A. (2020). Analysis of the role and scope
of big data analytics with IoT in health care domain. In Handbook of Data Science Approaches
for Biomedical Engineering (pp. 1–23). Academic Press.
39. Mishra, S., Dash, A., & Mishra, B. K. (2020). An insight of Internet of things applications in
pharmaceutical domain. In Emergence of Pharmaceutical Industry Growth with Industrial IoT
Approach (pp. 245–273). Academic Press.
40. Mishra, S., Tripathy, H. K., Mishra, B. K., & Sahoo, S. (2018). Usage and analysis of big data
in E-health domain. In Big Data Management and the Internet of Things for Improved Health
Systems (pp. 230–242). IGI Global.
41. Mishra, S., Tripathy, H. K., Mallick, P. K., Bhoi, A. K., & Barsocchi, P. (2020). EAGA-MLP—
An enhanced and adaptive hybrid classification model for diabetes diagnosis. Sensors, 20(14),
4036.
42. Mallick, P. K., Mishra, S., & Chae, G. S. (2020). Digital media news categorization using
Bernoulli document model for web content convergence. Personal and Ubiquitous Computing.
https://doi.org/10.1007/s00779-020-01461-9
Managing Interstitial Lung Diseases
with Computer-Aided Visualization

Adriana Trus, culescu, Laura Broască, Versavia Maria Ancus, a,

Diana Manolescu, Emanuela Tudorache, and Cristian Oancea

Abstract Pulmonary interstitial pathology and subsequent diagnosis represent

a significant component of pulmonary medical practice [97.9/100000 cases].
Complexity in differential diagnosis is added due to the individual pathological
entities’ low prevalence, very similar imaging aspects and individual variations.
Current approaches almost always require a multi-disciplinary approach, on a case-
by-case basis, with a team of high expertise and experienced medical practitioners.
Moreover, human diagnosis error rate is unsuitably high (~20–30%) (Trusculescu
et al. in European Radiology 30:6285–6292, 2020), with severe consequences for
the patient’s evolution and prognosis. High-resolution computer imaging together
with biopsy form the diagnosis foundation, yet often the biopsy is absent, creating
the need for accurate diagnosis based solely on visualization. Recent years have

A. Trus, culescu · E. Tudorache · C. Oancea

Department of Pneumology, Center for Research and Innovation in Precision Medicine of
Respiratory Diseases (CRIPMRD), “Victor Babes, ” University of Medicine and Pharmacy, Eftimie
Murgu Sq. no. 2, 300041 Timis, oara, RO, Romania
e-mail: [email protected]
E. Tudorache
e-mail: [email protected]
C. Oancea
e-mail: [email protected]
L. Broască · V. M. Ancus, a
Department of Computer and Information Technology—Automation and Computers Faculty,
“Politehnica” University of Timisoara, Vasile Pârvan Blvd. No. 2, 300223 Timis, oara, RO,
Romania
e-mail: [email protected]
V. M. Ancus, a
e-mail: [email protected]
D. Manolescu (B)
Department of Radiology, Center for Research and Innovation in Precision Medicine of
Respiratory Diseases (CRIPMRD), “Victor Babes, ” University of Medicine and Pharmacy, Eftimie
Murgu, Sq. no. 2, 300041 Timis, oara, RO, Romania
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 245
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_12
246 A. Trus, culescu et al.

proven that a mixture of computer enhancements and medical expertise are a syner-
gistic and precise approach. Computer techniques span two categories: learning and
discovery, each working either supervised or unsupervised; however, their output
is largely data-centric. A more novel tactic is using visual aids in which data is
processed and abstracted in the background, providing helpful hints and concen-
trated information to allow autonomous human conclusions. The valid and flex-
ible modeling of diffuse interstitial lung diseases (DILD) is especially important in
idiopathic pulmonary fibrosis (IPF), in which the accepted computer tomography
diagnosis criteria allow extremely diverse individual variations. Current advances
strictly structure the criteria, yet the four categories: typical, probable, indeterminate
and most consistent with non-UIP (usual interstitial pneumonia) span a vast pattern
array. Since IPF has such a dire prognosis (76, 8% mortality rate at 3 years), the
need for good pattern recognition and discovery is imperative to allow proper patient
management.

Keywords Diffuse interstitial lung diseases · Idiopathic pulmonary fibrosis ·

Computer-aided diagnosis · Visualization techniques · Pattern recognition

Abbreviations

AEP Acute eosinophilic pneumonia

AIP Acute interstitial pneumonia
BAL Bronchoalveolar lavage
CAD Computer-aided diagnosis
CEP Chronic eosinophilic pneumonia
COP Cryptogenic organizing pneumonia
CTD Connective tissue disease
DILD Diffuse interstitial lung disease
DIP Desquamative interstitial pneumonia
EBB Endobronchial biopsy
EBUS–TBNA Endobronchial ultrasound with transbronchial needle aspiration
HP Hypersensitivity pneumonia
HRCT High-resolution computer tomography
HSP Hypersensitivity pneumonia
GERD Gastroesophageal reflux disease
IT Information technology
IIP Interstitial idiopathic pneumonia
ILD Interstitial lung disease
IPF Idiopathic pulmonary fibrosis
LAM Lymphangioleiomyomatosis
LIP Lymphoid interstitial pneumonia
MDD Multi-disciplinary discussion
NSIP Idiopathic non-specific interstitial pneumonia
Managing ILD with Computer-Aided Visualization 247

PAH Pulmonary arterial hypertension

PLCH Pulmonary langerhans cell histiocytosis
PPFE Pleuroparenchymal fibroelastosis
RBILD Respiratory bronchiolitis interstitial lung disease
SLB Surgical lung biopsy
SPL Secondary pulmonary lobule
TBLB Transbronchial lung biopsy
TBLC Transbronchial lung cryobiopsy
UIIP Unclassifiable idiopathic interstitial pneumonia
UIP Usual interstitial pneumonia
VEGF-D Vascular endothelial growth factor D

1 Introduction

Pulmonary interstitial pathology is significantly modulated by the numerous and

diverse disease set. Although individual disorders present low prevalence, at the
threshold for rare diseases, together they amass a sizable component of pulmonary
medical practice [2].
Diffuse interstitial lung diseases (DILDs or simply ILDs), characterized by diffuse
damage to the lung parenchyma, with varying degrees of inflammation, fibrosis
and architectural distortion, are a heterogeneous group comprising more than 200
disorders. Pathological processes, although initiated at the pulmonary interstitium,
frequently surpass it, expanding to include the alveolar space, the small airways,
the vessels and/or the pleura. DILDs often present similar signs and symptoms,
yet different evolution patterns, therefore generating difficulties in determining the
correct diagnostic and treatment, more often than not requiring a multi-disciplinary
approach [3].
Data-centric decisions are the forte of computer-based applications, making it only
natural to use them as aides. Computer-aided diagnosis (CAD) is often employed
in DILD management, at both diagnosis and treatment stages. However, CAD has
several facets that need careful consideration: selecting the proper technique for the
medical needs and conveying the answers in a manner accessible to persons without
information technology (IT) background.
The purpose of this paper is to provide a guide in choosing such algorithms for
the specific field of ILD management, with an emphasis on visual-based complex
network method. This paper is organized following the normal procedures of diag-
nosis and treatment with Sect. 2 over viewing ILD classification, highlighting the
differential diagnosis elements segueing into treatment, and Sect. 3 follows computer
algorithms types used in the process.
248 A. Trus, culescu et al.

2 ILD Diagnosis

Histopathologically, all ILDs display different inflammation and fibrosis degrees

characterized by high variability among ILD diseases subtypes and individuals
afflicted with the same disease.
When the inflammatory pattern predominates, the histology is that of organizing
pneumonia (OP) or non-specific interstitial pneumonia (NSIP) that responds well to
corticosteroid anti-inflammatory treatment, while in the fibrosis pattern, the presen-
tation is interstitial pneumonia (UIP) displaying fibroblastic foci as well as a lesser
inflammation response and reacting well to anti-fibrotic medication [4].
A recent study [5] on almost 2000 subjects identified the four most common
subtypes of ILDs as idiopathic pulmonary fibrosis (IPF, 20.3%), followed by intersti-
tial pneumonia with autoimmune features (IPAF, 17.9%), connective tissue disease-
associated ILD (CTD-ILD, 18.3%) and unclassifiable idiopathic interstitial pneu-
monia (UIIP, 14.7%). IPAF is a provisional category of ILD, with subtle character-
istics for an autoimmune etiology, which do not meet the classification criteria for a
specific CTD, therefore being allocated for the time being the subgroup of idiopathic
interstitial pneumonias (IIP).
Idiopathic pulmonary fibrosis (IPF) is set apart by its high frequency and partic-
ularly severe prognosis. IPF mortality is high, exceeding most neoplasms, with an
average survival time of 3–5 years from diagnosis [6]. Due to the effective anti-fibrotic
medication remarkable progress, the correct IPF diagnosis in the initial stages makes
a considerable difference in patient’s outcome [7].
Figure 1 illustrates an ILD classification based on [8], presented as a Sankey
diagram. The node width and flow are proportional with the relative incidence
frequency for each disorder class. Understandably, there are variations, correlated
with population ethnicity, age, socioeconomic status, genetic imprints, geographical
location, etc., therefore the values used are indicative of the general proportions and
do not represent exact figures.
There are four main classes of DILD: those of known cause, mainly a consequence
of connective tissue disorders [9]; those that have a granulomatous manifestation
like sarcoidosis; those that have an unknown cause—idiopathic and the final group
comprising the “leftovers,” mostly rare diseases with an extremely low incidence.
The most prevalent are the idiopathic interstitial pneumonias (IIP) with IPF
and IPAF as common manifestations. A notable difference is that IPF has a
chronic/fibrotic manifestation, while IPAF a predominantly inflammatory one,
together amassing almost 40% from total cases [5], while IIP in general reach almost
80%, confirming a Pareto-type distribution.
Differential diagnosis of ILD must take into consideration atypical pneumonia
(including pneumocystis species), lymphangitic spread of cancer and congestive
heart failure that might present similar chest radiographic aspects; however, various
paraclinical exams can confirm/infirm those hypothesis [10].
Managing ILD with Computer-Aided Visualization 249

Fig. 1 ILD classification

250 A. Trus, culescu et al.

Precise ILD diagnosis is critical [11, 12], because treatment benefiting one DILD
might grievously harm another (e.g., steroid dispensation in IPF is detrimental, due
to the underlying mechanism of fibroblast activation and proliferation) [13]
In practice, there is a clear sequential algorithm to diagnose DILD, presented in
[14]. Initial clinical evaluation includes history and physical examination, pulmonary
function test (PFT), connective tissue serology and most often than not, high-
resolution computer tomography (HRCT). If there are associated conditions that
warrant a specific diagnosis, a biopsy might not be necessary; otherwise, the chest
HRCT pattern will lead a multi-disciplinary discussion (MDD) that might suggest a
lung biopsy and a new MDD post-biopsy result. The HRCT and histopathological
patterns define the diagnosis.
A recent development is the molecular markers use, such as galectin (Gal) 1,
3, 9 and /or human epididymis protein 4 (HE4) that might offer valorous insights
for diagnosis, prognosis and therapy response [15]. Correlation between markers
and various HRCT patterns can immensely help the diagnosis process [16] and also
present therapeutic opportunities, for example by inhibiting galectins in pulmonary
fibrosis.
ILD diagnosis is a developing field, even if one considers the current practice
status. In many cases where biopsy might have been considered mandatory, HRCT
can now supplement its presence. For example, in some patients with IPF, the HRCT
appearance may be characteristic enough to render biopsy unnecessary(when UIP
criteria are met) [17–19]. In the fewer remaining instances where a radiological
diagnosis is not possible, HRCT is still very useful, but as a guide for the best surgical
biopsy site. There are even approaches that favor HRCT for prognostic evaluation
and disease staging [20–23].

2.1 HRCT Patterns

When diagnosing ILD, the lung tissue intrinsic nature, manifesting in particular
photon interaction properties combined with the need to integrate spatial data, makes
high-resolution computed tomography (HRCT) the preferred method in diagnosis.
Since the data consists of grayscale sequential images, differentiation is given by
textural patterns and their distribution.
Typical ILD patterns in (HR) CT images are reticulation (RE), honeycombing
(HC), ground-glass opacity (GGO), consolidation (CD), micro-nodules (MN),
emphysema (EM) or combinations of the above. However, finding the patterns is no
easy tasks as they can overlap, merge or might be missing some elements, rendering
them inconclusive.
Seen in Fig. 2, linear reticulations are the most frequent findings, while GGO
(Fig. 2b) is although considered a favorable prognostic finding, they are less common.
HC patterns (Fig. 2c) are a very inauspicious prognostic finding as is indicative of
end-stage disease.
Managing ILD with Computer-Aided Visualization 251

Fig. 2 HRCT patterns

The HRCT differential diagnosis of DILD takes into account the predominant
pattern, the ancillary findings and their distribution. In 2018, diagnostic categories
of UIP based on computed tomography patterns were revised by the Fleischer Society
[24]. A summary is presented below correlated with the histopathology form [11].
A special mention should be the one on ILD location. The distribution pattern is
an accepted diagnosis criterion [8, 25], assessing two dimensionally: in the cranio-
caudal plane and relative to the secondary pulmonary lobule (SPL) and interstitial
distribution. Table 1 showcases the main findings.
A complete diagnosis would include both the definitive HRCT distribution and
localization as well as the histopathological findings. In the much too frequent
instance of non-ideal cases in which the HRCT indicates one category and the biopsy
another, [14, 26] offers a solution for IPF that can be extended for all UIP, consid-
ering either definite finding as proof-positive and initiating treatment based on it.[26]
suggests using statistical data like age and sex to enhance the probability of positive
diagnosis.
252 A. Trus, culescu et al.

Table 1 Location characteristic for ILDs patterns

Distribution of abnormalities Craniocaudal plane
Upper lobe Basal Diffuse
Relation with Peripheral IPF COP
SPL (sub-pleural) NSIP
Central Alveolar
proteinosis
Perilymphatic Sarcoidosis Carcinomatous
lymphangitis
miliary
tuberculosis
Centrolobular Silicosis Asbestosis HSP

2.1.1 UIP Criteria

An ideal UIP diagnosis (Fig. 3) presents the following elements:

• CT distribution:
– Heterogeneous CT-often shown distribution.
– Basal (occasionally diffuse) and sub-pleural CT-shown localization
• CT features:
– Honeycombing
– Reticular pattern with or without peripheral bronchiectasis or bronchiolectasis.
– Absences of features to suggest an alternative diagnosis.
• Marked fibrosis/architectural distortion and patchy involvement of parenchyma
and fibroblast foci in the histopathological findings. Fig. 3e.
Figure 3 presents a typical UIP pattern, with Fig. 3a, b displaying a reticular
pattern transverse CT section, and Fig. 3c, d showcasing a honeycombing pattern
with sub-pleural and basal predominance. Figure 3e is a histopathological sample
demonstrating usual interstitial pneumonia (UIP) with marked fibrosis and fibroblast
foci.

2.1.2 Probable UIP Criteria

Probable UIP diagnosis (Fig. 4) presents the following elements:

• CT distribution:
– Distribution is often heterogeneous.
– Basaland sub-pleural predominant (Fig. 4a, b)
• CT features:
Managing ILD with Computer-Aided Visualization 253

Fig. 3 Usual interstitial pneumonia patterns

– Reticular pattern with (Fig. 4b–d) or without peripheral bronchiectasis or

bronchiolectasis(Fig. 4a, c)
– Honeycombing is absent (Fig. 3c, d)
– May have mild GGO (Fig. 2b)
• Some histological features from UIP criteria are present but to an extent that
precludes a definite diagnosis of UIP/IPF and absence of features to suggest an
alternative diagnosis.
Figure 4 shows in bilateral lung coronal reconstruction the presence of a reticular
pattern, without (Fig. 4a, c) or with (Fig. 4b, d) peripheral bronchiolectasis and
sub-pleural and basal predominance. Also, in Fig. 4c mild ground-glass opacities in
254 A. Trus, culescu et al.

Fig. 4 Probable usual interstitial pneumonia (UIP) pattern: (a–c) Transverse computed tomography
(CT) section, (b) Sagittal view of the right lower lobe, (d) Coronal reconstruction

the sub-pleural areas of right lower lungs could be observed, associating with the
reticular pattern, however with the marked absence of honeycombing.

2.1.3 Indeterminate UIP

Indeterminate UIP diagnosis (Fig. 5) presents the following elements:

• CT distribution:
– Basal and sub-pleural predominant (Fig. 4a, b)
• CT features:
– Subtle reticulation may have mild GGO or distortion (Fig. 5a, b) (known as
the “early UIP pattern”)
– CT features and/or distribution of lung fibrosis that do not suggest any specific
etiology (Fig. 5c, d) (the notorious “truly indeterminate for UIP”)
• Fibrosis with or without architectural distortion, with features favoring either a
pattern other than UIP or features favoring UIP secondary to another cause.
Managing ILD with Computer-Aided Visualization 255

Fig. 5 Indeterminate pattern for usual interstitial pneumonia (UIP) in transverse computed
tomography (CT) section

In Fig. 5a,b, the early UIP pattern shows subtle reticulation and mild distortion,
in the sub-pleural areas. Meanwhile, Fig. 5c,d illustrates mild to marked ground-
glass opacity, asymmetrical distribution between both lungs, and no sub-pleural
predominance—a truly indeterminate for UIP pattern.

2.1.4 Alternative Diagnosis

Alternative diagnosis most consistent with non-UIP diagnosis presents the following
elements:
• CT distribution:
– Peribronchovascular
– Perilymphatic
– Upper or mid-lung.
• CT features: Cysts; marked mosaic attenuation; predominant GGO; profuse
micro-nodules; centrolobular nodules; nodules; consolidation.
256 A. Trus, culescu et al.

Fig. 6 Patterns indicative of an alternative diagnosis

• Features of other histological patterns of IIPs (e.g., absence of fibroblast foci or

loose fibrosis) in all biopsies. Histological findings are indicative of other diseases.
Figure 6 presents typical non-UIP cases as follows: Fig. 6a—coronal CT sections
emphasizes the cystic pattern, with bilateral distribution in the upper lobes (histi-
ocytosis x); Fig. 6b—axial CT view with crazy paving pattern typical for alveolar
proteinosis; Fig. 6c—transverse CT section, illustrating consolidation pattern with
atoll sign-organizing pneumonia; Fig. 6d—transverse CT illustrating micro-nodules
with perilymphatic distribution most often found in sarcoidosis.

2.2 ILD Diagnosis Summary

The sheer number and variability of ILD require, despite the exhaustive literature
formulated diagnostic criteria, a multi-disciplinary discussion in order to remove
perception and knowledge bias. However, due to the fact that such actions are
conducted by humans, inherently flawed, the overall diagnosis precision is only
around 70% for good technical teams [1].
Managing ILD with Computer-Aided Visualization 257

A noted single point of failure is that the assessment focuses on an image’s gray
tones and geometrical structures. Such judgments are in fact a pattern matching
problem or, seen from a different perspective, a pattern learning and matching
problem.
Computer-aided diagnosis (CAD) seems a perfect choice to augment or even
supplement the multi-disciplinary discussion, since computers can excel at pattern
learning and matching.

2.3 ILD Treatment Algorithms

DILD treatment focuses on inflammation management and lung function conser-

vation, and it begins even before confirmation, at the mere ILD suspicion. Obvi-
ously, once confirmed, the disease severity will determine the approach aggressive-
ness. As a general directive, if a clear cause is established, the treatment will try
to remove that cause, however that is not often the case in practice, delegating the
doctor at prescribing symptomatic treatment aimed at controlling subsequent cough
and dyspnea, while improving or at least keeping constant patient’s life quality. ILD
complications are also considered in the treatment schema, as gastroesophageal reflux
disease (GERD), pulmonary arterial hypertension (PAH), lung cancer, sleep apnea
syndrome and depression can potentiate some negative aspects and need specific
treatment on their own.

2.3.1 Pharmacological Treatment

Pharmacological treatment focuses on two main components: one that tries to slow
and stabilize disease progression and the other that tries to reverse ILD effects on
the lung.
Oral corticosteroid medication is usually the “first responder” in ILD treatment,
with prednisone or a similar derivate drug as most common choice. Together with
certain immunosuppressive medication (such as mycophenolate mofetil, azathio-
prine, cyclophosphamide, rituximab), ILD progression is slowed significantly or even
stopped. However, the side effects need constant monitoring and might decrease life
quality.
Alternatively, anti-fibrotic medication, like pirfenidone and nintedanib, is focused
on the lung tissue, more specifically at limiting the scarring amount present,
conserving lung function with major indication in IPF. Conversely, the side effects
are even more severe than the previous options and to make this choice good
communication doctor–patient is required [27].
In [28], an overview of the clinical practice guidelines and statements for ILD has
been made. American Thoracic Society (ATS) European Respiratory Society (ERS)
and the Fleischner Society remain pillars in the issuance of the guidelines. If first the
258 A. Trus, culescu et al.

guidelines targeted IPF, the range of the new guidelines extends past management
of IPF to include treatment recommendations for other ILD.

2.3.2 Non-pharmacological Treatment

Treatment options include as follows:

• Long-term oxygen therapy must be prescribed to those who have severe chronic
resting room air hypoxemia [29]
• Pulmonary rehabilitation—with the aim of improving daily functioning.
There is a wide array of responses to the treatment, some ILD may not respond
at all, while others may respond instantly. Even the success standard is somehow
flawed: if the measurements (symptoms, physiologic and X-ray findings) taken
show stabilization, then the treatment is considered correct. Nevertheless, prac-
tice shows that this plateau is only temporary. Patient status decline can be due
to ILD natural progression, comorbidities or therapeutical side effect. Osteoporosis,
infection and muscle weakness are common treatment side effects that can trigger
condition deterioration.
Progressive fibrosing-interstitial lung disease phenotype should always and
constantly assess regarding pulmonary transplant [30].

3 Computer-Aided Techniques

In any computer application, one must take into account that the computer is basically
just a machine that does as it is programmed to, therefore selecting the right program
is crucial. To select the program, one must define its inputs, desired outputs and the
way to generate them.
In ILD diagnosis, the inputs are the ones available in the multi-disciplinary discus-
sion: patient-generated information (the initial clinical evaluation, high-resolution
computer tomography (HRCT), biopsy) and literature or practice-generated infor-
mation (ILD classification, diagnosis criteria, statistical and historical data). The
output is (ideally) the diagnosis for the current patient or (at least) a helpful hint for
the clinician.
Early computer applications in medical diagnosis started in the 1990s, having at
first a precision below or at most au pair with medical practitioners, but evolving
together with computer science to a much better accuracy [31].
Despite the much-touted advantages of using such applications, there is an ethical
discussion which needs to be at least mentioned here, regarding the implications of
using computers for a second or even first opinion, how much one should rely on them
and if their engineer creators are responsible for their outputs in time and to what
extent. Any medical act carries liability, but if the computer performs it, who takes
responsibility for it? This examination is beyond the purpose of this paper (which
Managing ILD with Computer-Aided Visualization 259

is to highlight ILD management-specific algorithms), yet before implementing any

such algorithms, their improvement in performance and reliability [32] should be
weighted by their legal implications, further debated in [33].
Returning to the technical aspects of ILD management computer-aided applica-
tions, two categories can be defined: The first analyzes data and generates a conclu-
sion based on it, while the second conjugates complex data sets and presents its
rendering that allows the humans to generate a conclusion.
The first category uses the computer as a second opinion maker, focusing on
learning and reasoning applications, while the latter category uses the computer only
as an enhancer, creating new ways to visualize data that allow the humans to process
them better.
Figure 7 presents an ILD-specific algorithm classification, splitting them among
three axes: conclusion maker (computer / human), purpose (discovery, learning
and reasoning) and dependence on human supervision (supervised, unsupervised,
reinforced).
The categories in Fig. 7 provide a quick guide in selecting an algorithm class.
For example, if we are looking to implement a diagnosis model that recognizes

Fig. 7 ILD-specific algorithms

260 A. Trus, culescu et al.

existing patterns, learning algorithms are the natural choice, because the algorithms
learn from the offered data. However, if a new diagnosis parameter is sought, the
discovery section provides the algorithms. The reasoning algorithms are used when
we need a consensus between inputs and rules, sometimes allowing for uncertainty or
using statistical inferences. Therefore, these algorithms are helpful in implementing
a diagnosis or treatment algorithm, consisting of clear rules that quantify the inputs.
As their name suggests, supervised machine learning consists of training a model
by feeding it a set of input data together with the expected corresponding output
values (which are known beforehand). The algorithm then generates a suitable model
(formula) which fits the input data and can be used for the analyzing new input data
[34].
Two of the most common supervised learning purposes are linear regression and
classification.
With unsupervised machine learning, there is no training set which tells the algo-
rithm how it should generate results, but rather the algorithm itself is responsible
for finding common denominators among data. They are mostly used for clustering
purposes, anomaly detection and neural networks. Some of the most popular unsu-
pervised ML algorithms are as follows: K-means clustering, hierarchical clustering,
DBSCAN clustering and hidden Markov model.
Reinforcement learning (RL) algorithms are based on a trial-and-error approach.
The learner is not being told what to do, but rather learns what are good actions and
what are unfavorable actions based on the rewards or penalties it gets according to
its actions. Thus, it will always choose the moves which allow it to maximize the
rewards [35].
These types of algorithms can be used in medical diagnosis in combination with
medical imaging in cases where doctors might be dealing with a prolonged therapy
process. Reinforcement learning algorithms have multiple applicability directions
such as ILD, dynamic treatment regimes, automated medical diagnosis or more
general domains [36, 37].
The following sub-sections will focus on the most implemented ILD-specific
applications from Fig. 7.

3.1 Regression

Regression models focus on determining the relationships between input and output
values, where the input values represent a number of features characterizing the
studied data. The output values consist of continuous values. Some of the most
widely used regression models are linear regression, logistic regression, polynomial
regression and support vector machine (SVM).
SVM can also be used as unsupervised ML, yet ILD-specific applications are yet
to be implemented.
The main difference between linear regression and logistic regression consists
in the purpose they are being used for. Linear regression is used for determining
Managing ILD with Computer-Aided Visualization 261

continuous values dependent on the input data. The goal of this algorithm is to fit a
straight line so as to best estimate the training data. Logistic regression is used for
classification purposes, which means labeling input data as one of a number of two
categories, with a certain probability or confidence level.
In the medical field, both algorithms have been successfully used for the detection
of a number of illnesses. With the vast number of markers and indicators gathered
with the help of medical hardware and software, it can be difficult to make relevant
correlations between all these aspects. But if all these indicators are to be consid-
ered as interdependent features, then a regression algorithm can improve the data
interpretation and correlation.

3.1.1 Linear Regression

Linear regression models have been widely used in the detection of interstitial lung
disease (ILD). It has been shown that such models can take into consideration multiple
features (some of them directly measured through lung function testing, some of
them calculated or observed through medical imaging) such as pulmonary artery (PA)
diameter, mean pulmonary arterial pressure (mPAP), HRCT determined ILD severity,
forced vital capacity (FVC), transfer factor of the lungs for carbon monoxide (TLCO)
and age, sex and body surface area (BSA). Through the use of these algorithms,
factors such as PA diameter have proven to play an important role in the diagnostic
process, therefore discovering s a new correlation to ILD [38, 39].

3.1.2 Logistic Regression

Logistic regression models, on the other hand, can be used to determine the prob-
ability of certain factors having an influence on a certain disease. In the case of
elderly pneumonia patients for example, assessing the probability of it being fatal
might include features like mean age, pulmonary severity index, associated diseases
(diabetes mellitus, high blood pressure) and QTc interval prolongation [40].
These regression algorithms are most often used in combination with medical
imaging techniques which should enforce a more accurate diagnosis. For instance,
there are algorithms for breast cancer risk estimation which have been devel-
oped based on logistic regression, artificial neural networks and medical imaging
(mammographies) [41, 42].
Another regression algorithm, random forest, was used to focus on the regions
of pulmonary from distinct topographic regions, by analyzing and simultaneously
extracting data from the three sections (upper, medial and lower), significantly
increasing the probabilistic degree of correct detection of ILD lesions [43]
262 A. Trus, culescu et al.

3.2 Hidden Markov Models

Hidden Markov models are often used in the decision-making process and are based
on the assumption that a patient is at a point in time in one of a finite number of
health states [44, 45]. These states bear the name of Markov states.
K-means clustering, hierarchical clustering and hidden Markov model algo-
rithms have all been quite often used in the disease progression evaluation, such as
chronic obstructive pulmonary disease, interstitial lung disease (ILD), tuberculosis
surveillance, Huntington’s disease and diabetes mellitus [46–49].

3.3 Neural Networks

Neural networks are algorithms that aim to reproduce the way human neurons work,
by implementing a complex architecture that starts with an image as an input that is
split into multiple layers (convolutional, fully connected) arranged in a specific order.
These algorithms were successfully used for lung nodule detection and classification
[50, 51], tuberculosis lesions classification [52], lesions detection and classification
[53] and parenchymal pulmonary disease classification [54].
Usual ILD patterns easily achieve an 85.5% classification performance [55], with
other algorithms like [56] having near-perfect performance for specific cases.
The number of HRCT specific applications is limited, e.g., [57, 58] and even
though improvements are used, other implementations, e.g., [59, 60] offer a far
better performance.
The most difficult problem for ILD neural networks is not, however, the algo-
rithm, but the training set [61, 62]. Large, balanced HRCT imaging data sets present
lesions, otherwise overlooked but they are not public and not as substantial as needed
[54, 57, 58, 63].

3.4 Complex Networks

The human capacity to keep track of all the available data volume starts to fall
behind. Considering that each research institute studies and produces their own data
and, respectively, their own conclusions, no single researcher can be completely up-
to-date, nor can they verify information accuracy and validity by hand. This is where
a need for large storage media and fast processing power arises [64].
One of the medical areas on which software could help with is information storage
using a desired data format. Although standard notions do have a few common
Managing ILD with Computer-Aided Visualization 263

medical representation format [65], there is always the meta-data, and the not-so-
formal information which gets stored in a more human language, which is charac-
teristic to each institute or researcher. In this aspect, every entity is free to use their
own custom format which fits their needs.
On the other hand, whenever one wishes to start modeling data, based on existing
medical databases, in most cases, the researcher needs to “cure” the data, or pass it
through a “harmonization” process which either restructures the data, or skims out
the unnecessary details which could slow down any subsequent data analysis process
[66].
Another approach to medical science using IT tools refers to processing power.
Given there is such a vast amount of information at hand, modeling it would take a
very long time without any additional help. This is where processing power could
intervene on different aspects: modeling data according to data-type-specific models
[67], generating prediction algorithms based on statistics and probabilities [68],
programmatically exploring different scenarios based on data confidence level, etc.
In terms of data modeling and visual representation, there have been numerous
approaches to it.
Complex networks have become an increasingly popular means of visualizing and
analyzing healthcare data [69]. This comes as an alternative method to the all-too
knows bar charts, graphs, pie charts and squares schemas, and the explanation for
this lies in a number of extra traits which this type of representation has to offer
[70, 71].
But a few emerging intrinsic characteristics make network science a very good
tool in the hand of medical research, due to the data nature: the interconnectedness
and the connection dynamism.
The connections between data entities are not only static, but represent transfor-
mation relationships, or better yet, convey the influence that nodes assert over one
another. The patients flow in a hospital is an example of such data which could be
represented as a complex network [72]. Metabolic processes and elements are also
a good candidate to be used as database for a network representation.
The more information such a graph reflects, the more the person looking at it will
encompass in one glance, rather than corroborating multiple spreadsheets, or looking
at plain numbers or statistics.
There is a wide range of network visual tools and algorithms to choose from
when it comes to data representation. The most popular visualization tools used in
the domain of bioinformatics are as follows: Gephi, Cytoscape, NetworkX (Python
library), iGraph (R and Python) or Pajek. The advantage of using such tools is that
not only do they offer a network visual representation but they also implement a
number of different visualization algorithms for the user to choose from: Force Atlas
2, Fruchterman-Reingold, Kamada Kawai, OpenOrd, Circular layout, etc. [73, 74].
In the end, this variety of options translates into different perspectives of the same
network. This means that the user does not have to worry about how to arrange
elements over a blank canvas, but leverages the ready-made algorithms for this
purpose and then observes the end result in order to gain a better understanding
of the data.
264 A. Trus, culescu et al.

Network visualization tools are split into two categories depending on their main
focus. The first category of tools is that of user-friendly tools based on a visually
appealing user interface and quite an intuitive way of using. This includes Gephi,
Pajek and Cytoscape. At the opposite end, the second category is mainly focused
on efficiency and delivering fast results with a scalable architecture. However, this
former category requires more technical knowledge from the end user, who might
need specialized assistance in order to be able to fully exploit the advantages of these
tools. Visual interface lacks almost completely in this case, yet this is the cost of
speed-oriented and heavily customizable network visualization tools like NetworkX
or iGraph.

3.5 Layout Algorithm Selection

Depending on the need and the desired output, there are a few types of network
visualization algorithms: force-directed algorithms, circular layout, arc layout and
adjacency matrix. However, not all of them are as suitable to all kinds (and sizes)
of medical data. While matrix layout looks almost too mathematical and far too
abstract in order to see interactions clearly, arc layout and simple circular layout
become too crowded in the case of large networks. Thus, force-directed layouts come
across as adjustable, scalable and—most importantly—meaningful from a medical
perspective. However, what lacks in this case is a sort of ordering between nodes so
that the user can easily determine visually the node rankings or a hierarchy [75].
We can consider that choosing the most appropriate network layout is based on
four factors:
• The characteristics of the network—the available data traits.
• The desired aspects one wishes to emphasize—what the user wants to obtain by
using a certain layout.
• The layout algorithm specifications—what the algorithm claims that it does best.
• Previous studies or performance reviews.
In terms of network characteristics and data traits, fortunately medical data has
multiple facets to it. If we consider a metabolic network which consists of genes
and proteins, it is safe to say that most such networks are dense; they have a lot
of edges and nodes. The importance of each element might be expressed in such
a network through: node size (might be proportional to its degree—aka number of
connections), node color—representing a certain cluster or a function/role it fulfills
within the network, proximity to other nodes or clusters, edge width, etc.
The desired outcome, depending on what the researcher needs, might be a very
good clustering—so as to delimitate groups of nodes with a high similarity degree,
might be delimitation of odd ones, or even discovering patterns otherwise invisible
without a proper visual representation (Fig. 8).
While the first two aspects depend solely on the user’s judgment, the third and the
fourth—algorithm specifications and review documentation—are based on available
Managing ILD with Computer-Aided Visualization 265

Fig. 8 Bio diseasome network rendered with a hybrid layout algorithm

documentation offered by their creators, but also on the experience of others (user
feedback) who have tried, tested and reviewed them but also shared their results.
There are quite a few studies on the performance and efficiency of visualization
tools and layout algorithms, and the data sets are often quite variated, including
medical data [76].
In reality, when it comes to layout selection, there is a trial-and-error process,
where experimenting with layouts may convey exceptional results depending on the
specifics of the network. Cluster separation, node distancing or proximity, different
node colors and sizes, edge length and weight—these are all visual cues that indicate
and underline the connectivity between the elements of a network (Fig. 9).
Beyond the qualities of each type of layout, the discussion also centers on the
efficiency and speed of such algorithms. Given the amount of data which needs
processing and displaying, there have been case studies showing that beyond a certain
order of magnitude and volume, some of them cannot cope with such a workload.
It may take hours to generate a suitable layout provided that the user detains the
required processing power for this kind of tasks, otherwise the whole attempt might
end in the application crashing before the end of the rendering process. Metabolic
networks for example, or even genetic pathways classify as a type of dense networks
consisting of a large number of nodes and a great number of edges as well.
At this point, there is a clear differentiation between them and opting for user-
friendly tools takes its toll.
Although there have been studies on the performance of such tools [76, 77], from
a performance perspective, when it comes to people in the medical field using them,
there does not seem to be a clear predilection toward a certain algorithm, some studies
do not even mention the type of algorithm used for the generation of a certain network.
At best, it may mention the clustering algorithm used for community detection.
266 A. Trus, culescu et al.

Fig. 9 DNA network of genes with respiratory function—Force Atlas 2 layout

This means that once the researcher has found a suitable visualization algorithm,
they do not invest more time into a more in-depth analysis of multiple such tools,
but rather try to work out the best possible outcome with the chosen one (Fig. 10).

Fig. 10 Using multiple layout algorithms throughout the same research material. Reproduced with
permission from: [78]
Managing ILD with Computer-Aided Visualization 267

Recent studies on the COVID-19 disease show that researchers may also use
multiple tools (and algorithms) in order to generate multiple points of view for the
same network, but the accent is always on the medical gain [78]. Which is, indeed,
the main purpose of such studies, yet in these cases, the importance and utility of
the infrastructure used to obtain the results remain unquantified, and as such, the
knowledge and experience do not get transmitted to the consumers of such studies.

4 Conclusion

Managing ILD is a difficult task that needs to juggle: the disease variety, low indi-
vidual prevalence, high mortality rates, relatively low diagnosis accuracy, highly
individualized treatment plan, high costs [79] and complex computer algorithms to
help with it all.
The computer can offer two assistance types: second opinion or information visu-
alization and aggregation. Selecting assistance, one needs to select the purpose:
learning, discovery or reasoning, and the way this is achieved: through various
predictive analytics or visualization techniques.

References

1. Trusculescu, A. A., Manolescu, D., Tudorache, E., & Oancea, C. (2020). Deep learning in
interstitial lung disease—how long until daily practice. European Radiology, 30(11), 6285–
6292. https://doi.org/10.1007/s00330-020-06986-4
2. Collard Harold R and King Jr Talmadge E, Diffuse Lung Disease. A Practical Approach, Robert
P. Baughman and Roland M. du Bois, Eds., 2012.
3. Du Bois, R. M., Egan, J. J., & Costabel, U. (2007). Diffuse parenchymal lung disease. In C. T.
Bolliger (Ed.), Progress in respiratory research (vol. 36). Karger.
4. Mikolasch Theresia, A., Garthwaite Helen, S., & Porter Joanna, C. (2017). Update in diagnosis
and management of interstitial lung disease. Clinical Medicine (London, England), 17(2),
146–153. https://doi.org/10.7861/clinmedicine.17-2-146
5. Guo, B., Wang, L., Xia, S., & Mao, M. (2020). The interstitial lung disease spectrum under a
uniform diagnostic algorithm: a retrospective study of 1,945 individuals. Journal of Thoracic
Disease, 12. https://doi.org/10.21037/jtd-19-4021.
6. Ley, B., et al. (2015). Unified baseline and longitudinal mortality prediction in idiopathic
pulmonary fibrosis. European Respiratory Journal, 1374–1381. https://doi.org/10.1183/090
31936.00146314.
7. Nakamura, H., & Aoshiba, K., (Eds.) (2016). Idiopathic pulmonary fibrosis. Advances in
diagnostic tools and disease management. Springer. https://doi.org/10.1007/978-4-431-555
82-7.
8. Desai, S. R., & Galvin, J. R. (2015). Plain film and HRCT diagnosis of interstitial lung disease. In
J. Hodler, R. A. Kubik-Huch, & G. K. von Schulthess (Eds.), Diseases of the chest, breast, heart
and vessels 2019–2022: Diagnostic and interventional imaging [Internet] (pp. 2019–2020).
Springer.
9. Kondoh, Y., Johkoh, T., Fukuoka, J., et al. (2015). Broader criteria of undifferentiated connective
tissue disease in idiopathic interstitial pneumonias. Respiratory Medicine, 109(3), 389–396.
268 A. Trus, culescu et al.

10. Fan, L. L., Deterding, R. R., & Langston, C. (2004). Pediatric interstitial lung disease revisited.
Pediatr Pulmonol, 38, 369–378.
11. Collard, H. R., Egan, J. J., Raghu, G., et al. (2011). An official ATS/ERS/JRS/ALAT statement:
Idiopathic pulmonary fibrosis: Evidence-based guidelines for diagnosis and management.
American Journal of Respiratory and Critical care Medicine, 183(6), 788–824.
12. Calandriello, L., Sverzellati, N., Walsh, S. L., et al. (2016) Interobserver agreement for the
ATS/ERS/JRS/ALAT criteria for a UIP pattern on CT. THORAX, 71, 45–51.
13. Nicholson, A. G., et al. (2004). Inter-observer variation between pathologists in diffuse
parenchymal lung disease. THORAX, 59, 500–5. https://doi.org/10.1136/thx.2003.011734.
14. Raghu, G., Remy-Jardin, M., Myers, J. L., et al. (2018). Diagnosis of idiopathic pulmonary
fibrosis. An official ATS/ERS/JRS/ALAT clinical practice guideline. American Journal of
Respiratory and Critical Care Medicine, 198(5), e44–e68. https://doi.org/10.1164/rccm.201
807-1255ST
15. d’Alessandro M. et al. (2020). Galactin-1, 3 and 9: Potential biomarkers in idiopathic pulmonary
fibrosis and other interstitial lung diseases. Respiratory Physiology & Neurobiology. https://
doi.org/10.1016/j.resp.2020.103546.
16. Masuo, M., Nukui, Y., Tateishi, T., Kishino, M., Tateishi, U., Morota, K., Ohbo, K., Miyazaki,
Y., & Nishiyama, N. (2020). Human epididymis protein 4 is a new biomarker to predict the
prognosis of progressive fibrosing interstitial lung disease. Respiratory Investigation. https://
doi.org/10.1016/j.resinv.2020.08.002.
17. Society American Thoracic and Society European Respiratory. (2000). Idiopathic pulmonary
fibrosis: diagnosis and treatment. International consensus statement. American Thoracic
Society (ATS), and the European Respiratory Society (ERS). American Journal of Respiratory
and Critical Care Medicine, 161(21), 646–664. https://doi.org/10.1164/ajrccm.161.2.ats3-00
18. Hunninghake, G. W., Zimmerman, M. B., Schwartz, D. A., et al. (2001). Utility of a lung
biopsy for the diagnosis of idiopathic pulmonary fibrosis. American Journal of Respiratory
and Critical Care Medicine, 164(2), 193–196. https://doi.org/10.1164/ajrccm.164.2.2101090
19. Wells, A., et al. (1998). Clinical usefulness of high resolution computed tomography in
cryptogenic fibrosing alveolitis. Thorax, 53, 1080–1087.
20. Edey, A. D., Devaraj, A. A., Barke, R. P., et al. (2011). Fibrotic idiopathic interstitial
pneumonias: HRCT findings that predict mortality. European Radiology, 21, 1586–1593.
21. Goh, N. S., Desai, S. R., Veeraraghavan, S., et al. (2008). Interstitial lung disease in systemic
sclerosis: a simple staging system. American Journal of Respiratory and Critical Care
Medicine, 177, 1248–1254.
22. Wells, A. U., Sverzellati, N., Walsh, S. L., et al. (2014). An integrated clinicoradiological staging
system for pulmonary sarcoidosis: a case-cohort study. The Lancet Respiratory Medicine, 2,
123–30.
23. Wells, A. U., & Antoniou, K. M. (2014). The prognostic value of the GAP model in chronic
interstitial lung disease: The quest for a staging system. Chest, 145, 672–674.
24. Lynch, D. A., et al. (2018). Diagnostic criteria for idiopathic pulmonary fibrosis: A Fleischner
society white paper. The Lancet Respiratory Medicine, 6(2), 138–152.
25. Societatea Romană de Pneumologie and Grupul de lucru pentru Pneumopatii Interst, iale difuze
s, i Sarcoidoză, “Ghid de diagnostic si tratament al pneumopatiilor interstitiale difuze, 2015.
26. An American Thoracic Society Pocket Publication. Guidelines for the Diagnosis and
Management of Idiopathic Pulmonary Fibrosis: UPDATE 2019, 2019.
27. Maher, T. M., & Strek, M. E. (2019). Antifibrotic therapy for idiopathic pulmonary fibrosis:
time to treat. Respiratory Research. https://doi.org/10.1186/s12931-019-1161-4.
28. Montesi Sydney, B. (2020). Practice guidelines for interstitial lung diseases: Widening the
reach. Lung India. https://doi.org/10.4103/lungindia.lungindia_487_20.
29. Jacob. S. S. et al. (2020). Home oxygen therapy for adults with chronic lung disease. An official
American thoracic society clinical practice guideline. American Journal of Respiratory and
Critical Care Medicine, 202(10). https://doi.org/10.1164/rccm.202009-3608ST.
30. De Oliveira, N. C., et al. (2012). Lung transplant for interstitial lung disease: Outcomes for
single versus bilateral lung transplantation. Interactive Cardiovascular and Thoracic Surgery,
14(3), 236–267. https://doi.org/10.1093/icvts/ivr085
Managing ILD with Computer-Aided Visualization 269

31. Suzuki, K. (2017). Overview of deep learning in medical imaging. Radiological Physics and
Technology, 10(3), 257–273.
32. Armstrong, N., & Hilton, P. (2014). Doing diagnosis: Whether and how clinicians use a
diagnostic tool of uncertain clinical utility. Social Science and Medicine, 120, 208–214.
33. Grote, T., & Berens, P. (2019). On the ethics of algorithmic decision-making in healthcare.
Medical ethics, 46(3).
34. Wernick, M. N., Yang, Y., Brankov, J. G., Yourganov, G., & Strother, S. C. (2010). Machine
learning in medical imaging. IEEE Signal Processing Magazine, 27(4), 25–38.
35. Netto, S. M. B., Leite, V. R. C., Silva, A. C., de Paiva, A. C., & de Almeida Neto, A. (2008).
Application on reinforcement learning for diagnosis based on medical image. In Reinforcement
learning (pp. 379–398). IntechOpen.
36. Ker, J., Wang, L., Rao, J., & Lim, T. (2017). Deep learning applications in medical image
analysis. IEEE Access, 6, 9375–9389.
37. Yu, C., Liu, J., & Nemati, S. (2019). Reinforcement learning in healthcare: A survey. arXiv
e-prints.
38. Chin, M., et al. (2018). Pulmonary artery size in interstitial lung disease and pulmonary hyper-
tension: Association with interstitial lung disease severity and diagnostic utility. Frontiers in
Cardiovascular Medicine, 5, 53.
39. Hamzah, M. F. M., Kasim, R. M., Yunus, A., Rijal, O. M., & Noor, N. M. (2017). Detection of
interstitial lung disease using correlation and regression methods on texture measure, In 2017
IEEE International Conference on Imaging, Vision & Pattern Recognition (icIVPR), pp. 1–4.
40. Taooka, Y., Takezawa, G., Ohe, M., Sutani, A., & Isobe, T. (2014). Multiple logistic regression
analysis of risk factors in elderly pneumonia patients: QTc interval prolongation as a prognostic
factor. Multidisciplinary Respiratory Medicine, 9(1).
41. Ayer, T., et al. (2010). Comparison of logistic regression and artificial neural network models
in breast cancer risk estimation. Informatics in Radiology, 30(1), 13–22.
42. Jiang, F., Jiang, Y., Zhi, H., et al. (2017). Artificial intelligence in healthcare: past, present and
future. Stroke and vascular neurology, 2(4), 230–243.
43. Liu, C., Zhao, R., Xie, W., & Pang, M. (2020). Pathological lung segmentation based on
random forest combined with deep model and multi-scale superpixels. Springer. https://doi.
org/10.1007/s11063-020-10330-8.
44. Sonnenberg, F. A., & Beck, J. R. (1993). Markov models in medical decision making: A
practical guide. Medical Decision Making, 13(4), 322–338.
45. Kawamoto, R., et al. (2006). Application of hidden markov models and hidden semi-markov
models to financial time series. Studies in Health Technology and Informatics, 192, 491–495.
46. Pikoula, M., et al. (2019). Identifying clinically important COPD sub-types using data-
driven approaches in primary care population based electronic health records. BMC Medical
Informatics and Decision Making, 19(1), 86.
47. Powell, G. A., Verma, A., Luo, Y., Stephens, D., & Buckeridge, D. (2019). Modeling chronic
obstructive pulmonary disease progression using continuous-time hidden markov models.
Studies in Health Technology and Informatics, 264, 920–924.
48. Kwon, B. C. et al. (2020). DPVis: Visual analytics with hidden markov models for disease
progression pathways. IEEE transactions on visualization and computer graphics.
49. Caron, M. (2020). Pulmonary function tests as surrogate markers of interstitial lung disease
onset in systemic sclerosis. McGill University.
50. Ciompi, F., de Hoop, B., van Riel, S. J., et al. (2015). Automatic classification of pulmonary peri-
fissural nodules in computed tomography using an ensemble of 2D views and a convolutional
neural network out-of-the-box. Medical Image Analysis, 26(1), 195–202.
51. Tu, X., Xie, M., Gao, J., et al. (2017). Automatic categorization and scoring of solid, part-solid
and non-solid pulmonary nodules in CT images with convolutional neural network. Scientific
reports, 7(1), 8533.
52. Lopes, U. K., & Valiati, J. F. (2017). Pre-trained convolutional neural networks as feature
extractors for tuberculosis detection. Computers in Biology and Medicine, 89, 135–143.
270 A. Trus, culescu et al.

53. “Cicero, M., Bilbily, A., Colak, E., et al. (2017). Training and validating a deep convolutional
neural network for computer-aided detection and classification of abnormalities on frontal chest
radiographs. Investigative Radiology, 52(5), 281–287.
54. Walsh, S. L. F., Calandriello, L., Silva, M., et al. (2018). Deep learning for classifying fibrotic
lung disease on high-resolution computed tomography: A case-cohort study. The Lancet
Respiratory Medicine. Published online Sept 16. https://doi.org/10.1016/S2213-2600(18)3
55. Anthimopoulos, M., Christodoulidis, S., Ebner, L., et al. (2016). Lung pattern classification
for interstitial lung diseases using a deep convolutional neural network. IEEE Transactions on
Medical Imaging, 35(5), 1207–1216
56. Gao, M., Bagci, U., Lu, L., et al. (2018). Holistic classification of CT attenuation patterns
for interstitial lung diseases via deep convolutional neural networks. Computer Methods in
Biomechanics and Biomedical Engineering: Imaging & Visualization, 6(1), 1–6.
57. Li, Q., et al. (2014). Medical image classification with convolutional neural network. In 13th
International Conference on Control Automation Robotics & Vision (ICARCV), pp. 844–848.
58. Li, Q., Cai, W., & Feng, D. D. (2013). Lung image patch classification with automatic feature
learning. In 35th Annual International Conference of the IEEE Engineering in Medicine and
Biology Society (EMBC), pp. 6079–6082.
59. “Christodoulidis, S., Anthimopoulos, M., & Ebner, L., et al. (2017). Multi-source transfer
learning with convolutional neural networks for lung pattern analysis. IEEE Journal of
Biomedical and Health Informatics, 21(1), 76–84.
60. “Kim, G. B., Jung, K. H., Lee, Y., et al. (2018). Comparison of shallow and deep learning
methods on classifying the regional pattern of diffuse lung disease. Journal of digital imaging,
31(4), 415–424.
61. “Soffer, S., Ben-Cohen, A., Shimon, O., et al. (2019). Convolutional neural networks for
radiologic images: A radiologist’s guide. Radiology, 290(3), 590–606.
62. “LeCun, Y., Bengio, Y., Hinton, G. (2015). Deep Learning. Nature 521(7553), 436–44.
63. Simonyan, K., & Zisserman, A. (2014). Very deep convolutional networks for large-scale image
recognition. Computer Vision and Pattern Recognition
64. Dash, S., Shakyawar, S. K., Sharma, M., & Kaushik, S. (2019). Big data in healthcare:
Management, analysis and future prospects. Journal of Big Data, 6
65. Weeks, J., & Pardee, R. (2019). Learning to share health care data: A brief timeline of influential
common data models and distributed health data networks in U.S. health care research. The
Journal for Electronic Health Data and Methods, 7(1), 4.
66. Ancusa, V., & Broasca, L. (2015). A method to pinpoint undiscovered links in genetic and
protein networks. Studies in Health Technology and Informatics, 771–775.
67. Danese, M. D., Halperin, M., Duryea, J., & Duryea, R. (2019). The generalized data model for
clinical research. BMC Medical Informatics and Decision Making, 19. (Online). https://doi.
org/10.1186/s12911-019-0837-5
68. Li, J., Chen, Q., & Liu, B. (2016). Classification and disease probability prediction via
machine learning programming based on multi-GPU cluster MapReduce system. The Journal
of Supercomputing, 73, 1782–1809.
69. Kitsak, M., et al. (2010). Identification of influential spreaders in complex networks. Nature
Physics, 6, 888–893.
70. Vulliard, L., & Menche, J. (2020). Complex networks in health and disease. Systems Medicine,
26–33.
71. Goh, K.-I., et al. (2007). The human disease network. PNAS, 104(21), 8685–8690.
72. Bean, D. M., Stringer, C., Beeknoo, N., Teo, J., & Dobson, R. J. (2017). Network analysis of
patient flow in two UK acute care hospitals identifies key sub-networks for A&E performance.
PLoS ONE, 12(10)
73. Jacomy, M., Venturini, T., Heymann, S., & Bastian, M. (2014). ForceAtlas2, a continuous graph
layout algorithm for handy network visualization designed for the gephi software. PLoS ONE,
9(6).
74. NetworkX developers. (2014). NetworkX network analysis in python. (Online). https://net
workx.org/documentation/stable//reference/drawing.html
Managing ILD with Computer-Aided Visualization 271

75. McGuffin, M. J. (2012). Simple algorithms for network visualization: A tutorial. Tsinghua
Science and Technology, 17(4), 383–398.
76. Broasca, L., Ancusa, V. M., & Ciocarlie, H. (2019). A qualitative analysis on force directed
network visualization tools in the context of large complex networks. In 23rd International
Conference on System Theory, Control and Computing (ICSTCC).
77. Broască, L., Ancuşa, V., & Ciocârlie, H. (2016). Bioinformatics visualisation tools: An
unbalanced picture. Studies in Health Technology and Informatics, 760–764.
78. Das, A. B. (2020). Lung disease network reveals the impact of comorbidity on SARS-CoV-2
infection. bioRxiv. https://doi.org/10.1101/2020.05.13.092577
79. Raimundo, K., Chang, E., Broder, M. S., et al. (2016). Clinical and economic burden of
idiopathic pulmonary fibrosis: a retrospective cohort study. BMC Pulmonary Medicine, 16,
2.
Use of Machine Learning Algorithms
to Identify Sleep Phases Starting
from ECG Signals

Giovanna Sannino and Ivanoe De Falco

Abstract The absence of the rapid eye movement (REM) phase during sleep can
have negative consequences, as, e.g., anxiety, increase in appetite, irritability, while,
on the other hand, it can help in improving some kinds of depression. The goal of
the research described in this chapter consists in the identification of the different
sleep phases a subject is experiencing by using heart rate variability (HRV) values.
These are computed starting from the signals gathered from electrocardiogram (ECG)
sensors placed on the subject. To this aim, the publicly available Sleep Heart Health
Study (SHHS) data set is taken into account, which contains both types of informa-
tion. Several machine learning classification algorithms are tested on this data set,
and their performance is compared in terms of F1-score value, as SHHS is highly
unbalanced. Once the most suitable classification algorithm is found, it can be firstly
trained offline on the problem and then used online in an IoT-based fully automated e-
health system. In this latter, sensors gather, in real time, ECG signals from a sleeping
subject, send them to a device where data is processed, HRV values are computed,
sleep phase identification takes place, and medical personnel, close or not to the
subject, are immediately informed of the subject’s sleeping phases.

Keywords Electrocardiogram · Heart rate variability · Sleep phases ·

Classification · Machine learning

1 Introduction

The Internet of Things (IoT) [1] is a set of technologies that, in the coming years,
will involve all aspects of our daily life, but it will also be revolutionary in key
fields, like medicine and health care. The HealthCare Internet of Things is a vast and

G. Sannino (B) · I. De Falco

Institute for High-Performance Computing and Networking (ICAR), National Research Council
of Italy (CNR), Via Pietro Castellino, 111, 80131 Naples, Italy
e-mail: [email protected]
I. De Falco
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 273
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_13
274 G. Sannino and I. De Falco

complex sector that concerns the methods of managing health data, their flow, and
their processing [2]. Numerous are papers related to the use of healthcare data for
the classification or the detection of possible dangerous situations, as in [3–9].
The practical consequence of the use of IoT in health care will be the enor-
mous usability of health data, a zeroing of the processing times of these data, and
extreme efficiency in health management. All these advantages could in many cases
be achieved by using artificial intelligence methods, be they machine or deep learning
techniques.
Within this field, AI techniques can support the definition of new treatment plans
and the development of smart models and approaches to the analysis of patient data,
such as, for example, the detection and monitoring of sleep stages [10].
Sleep is of significant importance in human mental and physical health. It is an
important mechanism for restructuring and rebalancing the metabolic and humoral
systems of human subjects [11]. Significant alterations in sleep patterns determine
profound effects on essential aspects of life such as the metabolism [12], like diabetes
or obesity, psychic health, like stress, irritability, and depressive disorders [13], the
immune system [14], and the cardiovascular system [15], like hypertension. Addi-
tionally, an altered sleep could be a possible index of mild and/or severe pathologies
of neurodegenerative diseases [16].
In sleep studies aiming at clinical diagnosis and treatment of sleep disturbances,
the fundamental step consists in identifying the sleep stages of a subject.
The American National Sleep Foundation recommends continuous sleep periods
of at least 7–9 h for adults [17], but as many as one-third of the populations studied (the
USA, UK, and France) sleep less than 6 continuous hours a night and 7% even less
than 5 h (https://www.sleepassociation.org/about-sleep/sleep-statistics/). Although
real sleep (recorded instrumentally) is not easily correlated with subjective sleep
(detected by questionnaires), however in a large sample of adult subjects, it was
found that 43% sleep less than 6 h a night, as documented in the CARDIA study
[18]. Globally, more than 50 epidemiological studies on sleep disorders have been
conducted, the comparison of which is difficult due to the differences in the criteria
used.
In medical practice, the classification of sleep stages is commonly based on
three instrumentally measurable neurophysiological variables: electroencephalog-
raphy (EEG), electro-oculography (EOG), and electromyography (EMG). These
signals are collected during the well-known polysomnography (PSG) test, which
is a multiparametric examination to study sleep and to diagnose sleep diseases.
The standardized correlation criteria of three instrumental variables, EEG, EOG,
and EMG, were codified in 1968 by Rectschaffen and Kales and were revised in
2007 by the American Academy of Sleep Medicines.
In general, the sleep–wake cycle roughly recognizes three stages: wakefulness,
sleep without rapid eye movements (non-rapid eye movement, NREM), and rapid
eye movement (REM) stage. NREM stage is divided into three other phases: N1
sleep, N2 sleep, and N3 sleep, which characterize what is called deep sleep.
Use of Machine Learning Algorithms to Identify … 275

Fig. 1 Overview of the polysomnography (PSG) test. Source National Heart Lung and Blood
Institute (NIH). The figure is taken from Wikimedia Commons—Sleep Studies—https://commons.
wikimedia.org/wiki/File:Sleep_studies.jpg. Last Accessed: February 10, 2021

As mentioned, the normal practice to monitor and detect the sleep stages is the
PSG test, which, unfortunately, is a very invasive examination, as can also be seen
in Fig. 1.
To make the sleep stage detection an easy and non-invasive examination, we have
conducted a research study consisting in the classification of the different sleep phases
of a subject by only using the heart rate variability (HRV) values that can be calculated
starting from the signals gathered from a single and wearable electrocardiogram
(ECG) sensor placed on the subject.
This approach can be effortlessly adapted into the various scenarios of IoT health
care and mobile IoT health care, sleep health monitoring in smart homes, and others.
In fact, a classification algorithm can be firstly trained offline on the problem and then
used online in an IoT-based fully automated e-health monitoring system. In this latter,
sensors gather, in real time, ECG signals from a sleeping subject, send them to a device
where data is processed, HRV values are computed, sleep phase identification takes
place, and medical personnel, close or not to the subject, is immediately informed
of the subject’s sleeping phases.
To verify the applicability of this approach, in this chapter the publicly avail-
able Sleep Heart Health Study (SHHS) data set is taken into account, and several
276 G. Sannino and I. De Falco

machine learning classification algorithms are tested to classify the sleep stages.
Their performance is compared in terms of F1-score value because SHHS is highly
unbalanced.

2 Related Works

In the literature, various research groups have proposed different approaches for the
classification of sleep stages through the analysis of biomedical signals and by using
artificial intelligence techniques, such as, for example, machine learning or deep
neural networks. Many of them use the ECG signal and its characteristics, but, in
most cases, they also use further signals. In the following, a brief description of the
most recent related works is reported.
In [19], Ni, Mingzhe, and Wang present an approach to the recognition of pure
sleep and stage transition, the multi-classification of pure sleep, and the verification
analysis of the coarse–fine-grained classification of stage transition. They extract
ECG data from PSG data and then extract the HRV features (in the time domain,
frequency domain, and nonlinear domain) utilized for the classification of pure sleep.
They perform a feature selection based on the max-relevance min-redundancy
criteria and classification by only using three classifiers. Results show that the clas-
sification effect using hidden Markov model (HMM) classifier is the best, and the
resulting accuracy is equal to 81.6%.
In [20], Zhang et al. present a wearable multi-sensor system to capture subjects’
ECG and breathing signals. This system applies a bidirectional RNN architecture
with long short-term memory (BLSTM) [6] to classify four sleep stages (wake, light
sleep, deep sleep, and REM). The presented system extracts the RR intervals from
ECG signals and calculates the HRV features: ten features in the time domain and 21
features in the frequency domain. Unfortunately, in their paper, they do not specify
which those features are. Additionally, the system uses 25 statistical features from
the breathing signal, and other features extracted from cardiopulmonary coupling
(CPC), for a total of 152 features.
They leverage 417 subjects from the SHHS data set and 32 subjects from their
proprietary data set, and achieve a sleep stage prediction in terms of accuracy equal
to 80.25% and 80.75%, respectively.
In [21], Goldammer and his colleagues carry out the classification of sleep stages
by means of 237 records taken from SHHS. They use 49 features (four raw HRV
features per time window, three time windows, two -features for each raw feature;
two raw respiratory features per time window, two time windows, two -features for
each raw feature; one time index feature). As classifiers, they consider random forests
(RF) and k-nearest neighbors (kNN). Unfortunately, they do not provide results for
their performance by using the classic metrics like accuracy or F1-score, rather they
assess their numerical results through Cohen’s κ [22] for each patient.
Starting from a single-lead electrocardiogram (ECG), the algorithm discussed in
[23] by Li et al. classifies four different sleep stages, i.e., wake, rapid eye movement
Use of Machine Learning Algorithms to Identify … 277

(REM) sleep, non-REM (NREM) light sleep, and NREM deep sleep. This is accom-
plished through the use of a set of steps. Firstly, beat detection is carried out, which
is followed by cardiorespiratory coupling in the time–frequency domain. Finally, to
perform classification, a deep convolutional neural network (CNN) is utilized.
Aiming to train and validate their algorithm, the authors make use of a set of
databases, namely the MIT-BIH Polysomnographic Database (SLPDB), the Phys-
ioNet/Computing in Cardiology Challenge 2018 database (CinC2018), and the SHHS
database.
They perform tenfold cross-validation, the results of which evidence for their
algorithm an accuracy (Acc) of 75.4% and a Cohen’s kappa coefficient of κ = 0.54
on the out-of-sample validation data when classifying Wake, REM, NREM light and
deep sleep in SLPDB data set. These figures increase up to Acc = 81.6% and κ =
0.63 when categorizing wake, REM sleep, and NREM sleep, and to Acc = 85.1%
and κ = 0.68 when the goal is to discriminate NREM sleep from REM/wakefulness.
Referring to the CinC2018 data set, when classifying wake, REM, light sleep, and
deep sleep, the algorithm obtains on out-of-sample validation an accuracy value of
65.6% and a Cohen’s kappa coefficient of 0.31. When the problem is modified into
a three-class classification and Wake, REM sleep, and NREM sleep are taken into
account, the outcome shows an Acc = 76.5% and κ = 0.42 for the classification.
Moreover, the two-class discrimination of NREM sleep versus REM/wakefulness
obtains as results the values of 79.4% for Acc and 0.48 for κ.
Finally, referring to SHHSv1, for the classification based on all the four classes,
i.e., Wake, REM, light sleep, and deep sleep, the algorithm proposed allows obtaining
an accuracy of 65.9% and a Cohen’s kappa coefficient of 0.47 on out-of-sample vali-
dation data. When the problem is modified into a three-class one by only considering
wake, REM sleep, and NREM sleep, values of Acc = 75.3% and κ = 0.57 are
achieved, Also, when the discrimination of NREM sleep versus REM/wakefulness
is investigated, the numerical results consist in values of 80.8% for Acc and 0.61 for
κ.
In [24], Xiao and his colleagues present a methodology relying on HRV to classify
sleep stages. Their approach aims at classifying three stages, namely wake, rapid eye
movement (REM) sleep, and non-REM (NREM) sleep. To this aim, they consider a
set of 45 healthy individuals (28 male and 17 female), whose ages range between 16
and 61. From the RR sequences of these subjects, the authors extract a set of 41 HRV
features. Then, they apply a random forest (RF) algorithm to perform supervised
classification. The data for these individuals is extracted from the publicly available
Sleep Heart Rate and Stroke Volume Data Bank (SHRSV) database, whose records
were acquired in the Sleep Center Institute of Psychophysiology and Rehabilitation
of Kaunas University of Medicine, Lithuania.
The authors perform both a subject-specific classification and a subject-
independent one. In the former case, they obtain a mean accuracy of 88.67% and
Cohen’s kappa statistic κ of 0.7393. In the latter case, instead, the numerical results
drop down to 72.58% for the accuracy and 0.4627 for κ.
278 G. Sannino and I. De Falco

3 The Database

The Sleep Heart Health Study (SHHS) database [25] has been collected during a
cohort study performed in the mid-1990s to investigate the obstructive sleep apnea
(OSA) events and other sleep-disordered breathing (SDB) disorders, and the major
cardiovascular events, including myocardial infarction and stroke.
SHHS provides good statistical power for assessing OSA [26] and other SDB
as risk factors, also due to the numerosity of participants. In fact, more than 6,400
(between men and women, aged 40 years and over) were enrolled, as well detailed
in [27]. Each of the participants has undergone a home polysomnogram (PSG) and
has been part of a follow-up procedure with periodic meetings to take note of the
incidence and the recurrence of cardiovascular disease events.
In detail, the database is composed of two sets: SHHS1 and SHHS2.
The first one, i.e., the follow-up 1 data set, has been collected on cardiovascular
events collected during the first part of the study, conducted approximately 2–3 years
after the enrollment visit. The population of the SHHS 2 data set, the second follow-
up examination, includes surviving members of the previous follow-up examination
who indicated an interest in undergoing a second home visit and/or a PSG.
In the presented study, we make reference to the SHHS1_hrv_5min data set, which
contains the analysis of the heart rate variability (HRV) features related to a set of
500 SHHS subjects who have participated in both visits (follow-up 1 and follow-up
2) and that have high-quality data.
This set contains the analysis of consecutive 5-min segments with no overlap.
In detail, as part of the polysomnographic study, the continuous electrocardio-
graphic (ECG) signals have been recorded at 125 Hz, and a Notch filter of 60 Hz
has been applied. Each of these ECG signals has been divided into 5-min segments,
and, over each segment, the HRV analysis is performed by following the guide-
lines articulated by the Task Force of The European Society of Cardiology and The
North American Society of Pacing and Electrophysiology [28]. The processing steps
effected to obtain the HRV feature vectors are shown in Fig. 2.
A description of the HRV measures used in this study is reported in Table 1.
Traditional HRV measures are commonly divided into two broad categories: time
domain measures and frequency domain measures.
In our experiments, we only consider the 5-min segments in which no apnea events
are present, as already performed in other studies, like in [21] and in [24].
Additionally, for each 5-min segment, the SHHS data set reports ten sleep stages
detected during these 5 min of sleep monitoring. These ten stages could be different
because the subject could change her/his status frequently during the considered
5 min. In order to correctly study a correlation between the HRV features and a
single sleep stage, we have decided to consider and select only the items in which
no sleep stage changes are detected.
Based on the above considerations, each item of the database we have used in this
study is defined as follows:
Use of Machine Learning Algorithms to Identify … 279

Fig. 2 Overview of the processing steps to obtain the HRV features vectors

Table 1 HRV features included in the study

Feature Description
Time domain
IHR Mean value of instantaneous heart rate. For a given NN interval, the IHR is
calculated as 60/NN. IHR is usually expressed as beats per minute but
strictly speaking it is unitless
NN_RR The ratio of consecutive normal sinus beats (NN) over all the cardiac
inter-beat (RR) intervals
AVNN The average of all the NN intervals
SDNN The standard deviation of all NN intervals
rMSSD The square root of the mean of the squares of the difference between
adjacent NN intervals
pNNx The percentage of differences between adjacent NN intervals that are >
x ms. We used x = 10, 20, 30, 40 and 50 ms
Frequency domain
TOT_PWR Total NN interval spectral power up to 0.4 Hz
VLF Very-low-frequency power: the NN interval spectral power between 0.003
and 0.04 Hz
LF Low-frequency power: the NN interval spectral power between 0.04 and
0.15 Hz
LF_n Low-frequency power (normalized)
HF High-frequency power: the NN interval spectral power between 0.15 and
0.4 Hz
HF_n High-frequency power (normalized)
LF_HF The ratio of low- to high- frequency power
280 G. Sannino and I. De Falco

Table 2 Data description of

Class Number of Items
the database used in this study
1—wake 1549
2—non-REM 6048
3—REM 347
Total 7944


 15
item = idparticipant ; visitNumber; idsegment ; HRVFeatures  ; SleepStage
1

where
• idparticipant is the unique identifier of each participant;
• visitNumber is always equal to 1, because we consider in our experiments only
SHHS1 that refers to the first night of monitoring;
• idsegment is a number that identifies each 5-min segment of that participant;
• HRVFeatures is the vector that contains the 15 features extracted considering that
5-min segment;
• SleepStage is the class that identifies the stage in which the participant is.
In the SHHS data set, there are six values, five of which concern the sleep stages:
0 is the wake stage, 1–4 are related to the sleep stages 1–4, and 5 is the REM sleep.
In our experiments, class 1 corresponds to the wake stage, in class 2 we have joined
into a unique class the sleep stages 1–4, and class 3 corresponds to the REM sleep
stage.
A description of the database used is reported in Table 2.

4 Experiments

4.1 Experimental Setup

4.1.1 The Classifiers

We have based our experimental activity on the Waikato Environment for Knowledge
Analysis (WEKA) [29] tool, version 3.8.1. This choice is due to both the large number
and the variety of the classifiers made available by WEKA.
These classifiers are in WEKA divided into groups, where all the items belonging
to a group share some common features. For example, all the items in the rules group
perform classification by finding one or more rules that can be used to assign each
data set item into one of the available classes. Similarly, all the classifiers in the
function group rely on the idea of using one or more “functions,” although these can
Use of Machine Learning Algorithms to Identify … 281

be of different types, as for instance perceptrons or radial basis functions, depending

on the specific classifier.
From each such group, we have taken into account some classifiers, namely those
that are often acknowledged in the scientific literature as capable of obtaining good
classification results.
Table 3 reports the classifiers we have considered in this paper. In the table, the
first column shows the name of the group, the second contains the complete name of
the algorithm, and the third reports the acronym we have used in the paper, whereas
the fourth contains a literature reference for the algorithm.
Each of these algorithms contains a set of parameters. A preliminary phase to tune
the values for these parameters could help the algorithm to achieve better classifica-
tion results, yet such a tuning phase is typically very time-consuming, as it requires
a wide set of trials before getting a satisfactory set of parameter values. This should
be repeated for each of the considered algorithms. To avoid this, we have decided
not to perform any tuning phase. Rather, we use the default parameters as they are
contained in WEKA. This means that the results obtained in our experiments could
be further improved.
Another important issue is that many of these algorithms rely on an initial random
value, called seed, to be assigned before the execution starts. This means that different
values for the random seed might yield different executions for a given algorithm,

Table 3 Complete list of the classification algorithms used in this paper

Group Algorithm Acronym References
Bayesian Bayes net BN [30]
Naïve Bayes NB [31]
Functions Logistic Log [32]
Multilayer perceptron MLP [33]
Radial basis function RBF [34]
Support vector machine SVM [35]
Weightless RAM-based neural model (Wisard) Wis [36]
Lazy Instance-based learner (K–nearest neighbor) Ibk [37]
Meta AdaBoostM1 AB [38]
Bagging Bag [39]
Rules One rule OneR [40]
Repeated incremental pruning Jrip [41]
(JRip)
Partial decision tree (PART) PART [42]
Ripple-down rule RIDOR [43]
Trees C4.5 decision tree (J48) J48 [44]
Random forest RF [45]
REP Tree RT [46]
282 G. Sannino and I. De Falco

hence different classification results. From this, it follows that for these algorithms
we cannot perform one single run: It could be a lucky run or an unlucky one, so our
conclusions could be strongly biased by this seed value. To overcome this problem,
it is very common in the scientific literature to run a number of runs, say at least ten
or more, and then consider as the indicator of the classification ability the average
of the classification results obtained over these runs. This is what we do here, where
we consider a number of 25 runs. This number is widely used in the literature, as it
is considered sufficiently sound from a statistical viewpoint.
Some other algorithms, instead, do not depend on any random seed, so, for them,
we execute each of them just once.

4.1.2 The Choices for Classification

We wish to perform supervised learning, hence we have to divide the data set into a
training set and a testing one. On the first, the classifier builds some either internal
or external knowledge that helps explain the regularities underlying the data and
the class they belong to. On the second, instead, this acquired knowledge is used to
classify new items not shown to the classifier in the learning phase. We have decided
to divide the data set by assigning to the training set the first 66% of the items in
their order of appearance in the data set, and by leaving the last 34% of the items for
the testing set. Special care must be posed to the fact that all the items related to a
given subject are either in the training set or in the testing set; otherwise, classifiers
might learn about some subjects and test over the same subjects. This is the reason
why we have to preserve the order of the data set items, meaning with this that no
random shuffling can be allowed.
Moreover, many statistical measures exist through the values of which the quality
of a proposed classification can be evaluated; among them, we can recall here accu-
racy, sensitivity, specificity, area under the ROC curve, and so on. The by-far most
widely used measure is accuracy. The problem with it is that, whenever a data set
is highly unbalanced, as the one used here, accuracy tends to hide some problems
that may arise in the classification process. The most important of these problems is
represented by the fact that an algorithm might try to work well on the most populated
classes while taking no care at all for the least populated ones. In fact, if all the items
in a class with just few items are incorrectly assigned to another class, possibly the
majority one, accuracy would, in any case, return numerically good values, indepen-
dently of this problem. The ability of correct classification, on the contrary, heavily
relies on correctly dealing with the classes with few items.
Consequently, we have decided to make use here of the F-score. In fact, this can
help us get rid of the above problem, as its values immediately evidence the above-
described undesired behavior. For the definition of F-score, let us start from a two-
class problem, in which we have a positive class PC with the subjects suffering from
a given disease and a negative class NC with the healthy subjects. The classification
divides these items into four groups:
Use of Machine Learning Algorithms to Identify … 283

• true positives: the items of PC that are properly allocated to PC, let us denote
their quantity with tp;
• true negatives: the items of NC that are properly allocated to NC, let us denote
their quantity with tn;
• false positives: the items of NC that are wrongly allocated to PC, let us denote
their quantity with fp;
• false negatives: the items of PC that are wrongly allocated to NC, let us denote
their quantity with fn.
With these definitions, we can define the F-score as follows:
tp
F score =
(t p + 0.5 · ( f p + f n))

For a data set with more than two classes, the value of the F-score for a generic
ith class, let us denote it as F i , is computed by considering it as the positive class,
while taking into account all the other classes together as the negative class. Then,
the F-score value is calculated as the weighted average of all the F i values achieved
over each of the data set classes, where the weight assigned to the generic ith class
is equal to the percentage of its items in the data set.
F-score can take on values in the range [0.0–1.0], where a higher value represents
a higher-quality classification.

4.2 Numerical Results

Table 4 shows the results obtained by all the algorithms over the testing set. Namely,
for each algorithm, the table shows the average value of the F-score, the related
standard deviation, as well as the maximum and minimum F-score values obtained
in the 25 runs. Of course, for the algorithms that do not depend on a random seed,
only one run is performed, hence for them the average corresponds to the actual value
obtained, and the standard deviation is equal to 0.
Moreover, for some runs it may happen that an algorithm may run and perform
classification, thus obtaining results in terms of, e.g., accuracy, yet it could not provide
a valid value for F-score. This happens whenever an algorithm is not able to place at
least one item in each of the data set classes, which is a problem for the quality of the
obtained classification. An explanation for this is that in such a case the algorithm
behaves in a greedy way and tries to allocate all items to the most populated classes,
which has a low impact on classification accuracy value, given the relatively low
number of items in the least populated classes. We represent this situation as a NaN
in the table. The second column in Table 4 reports the number of runs for which this
problem does not take place, with reference to the total number of runs performed.
The best value for each of the five columns is reported in bold. The runner-up is
reported in italic for the columns reporting average, maximum and minimum values.
284 G. Sannino and I. De Falco

Table 4 Numerical results for all the algorithms investigated

Algorithm Number of successful runs Average Std. dev Maximum Minimum
BN 1/1 0.756 – 0.756 0.756
NB 1/1 0.776 – 0.776 0.776
Logistic 0/1 NaN – Nan NaN
MLP 1/25 0.825 NaN 0.825 0.825
RBF 0/25 NaN – NaN NaN
SVM 0/25 NaN – NaN NaN
Wisard 16/25 0.714 0.015 0.755 0.690
Ibk 1/1 0.733 – 0.733 0.733
AdaBoostM1 0/25 NaN – NaN NaN
Bagging 25/25 0.808 0.012 0.828 0.785
OneR 0/1 NaN – NaN NaN
Jrip 25/25 0.789 0.010 0.816 0.770
PART 25/25 0.784 0.000 0.784 0.784
RIDOR 25/25 0.792 0.009 0.810 0.774
J48 25/25 0.784 0.000 0.784 0.784
Random Forest 25/25 0.807 0.003 0.814 0.802
REPTree 25/25 0.791 0.011 0.815 0.771

As can be seen from the table, the MLP algorithm obtains the overall best average
performance. The problem with this algorithm is that this is the only run, out of the
25, in which MLP avoids yielding NaN. This is not a good performance for MLP, as
it seems that in the vast majority of the cases it fails computing F-score value, which
means it fails in assigning at least one item to each class.
Given this problem, we see that the runner-up in terms of average performance
is Bagging, immediately followed by Random Forest. Both algorithms succeed in
providing valid values for F-score in all the 25 runs, so we can see them as the
best-performing algorithms on average.
As concerns the best single run, the “Maximum” column of the table informs that
Bagging also yields the highest value, while the runner-up is, in this case, MLP with
its only successful run. In its turn, MLP is followed by JRip.
Instead, if the “Minimum” column is considered, we can see that MLP obtains in
its only run the highest such value, followed by Random Forest, Bagging being the
third-best technique here.
In conclusion, if we consider all the above results altogether, Bagging can
be considered as the best-performing and most-reliable algorithm, immediately
followed by Random Forest.
As a further remark to the table, a high number of unsuccessful runs can be seen:
It is equal to 110 out of 305, i.e., about 36.06% of the executed runs. This means
that the classification problem faced in this chapter is not trivial to face, even for
Use of Machine Learning Algorithms to Identify … 285

algorithms that are known for their goodness at dealing with this task. Going into
details, highly reputed classifiers as Radial Basis Function, Support Vector Machine,
AdaBoostM1, and MultiLayer perceptron, experience non-trivial problems with it.
This is a somehow surprising outcome of our experiments.

4.3 Statistical Analysis

We have also decided to perform statistical analysis on the achieved results to inquire
into whether or not the results obtained by the various algorithms are statistically
different. To this aim, we have chosen to only consider here all the algorithms that
are executed 25 times and obtain 25 runs in which they compute the F-score value.
With these assumptions, the set of algorithms involved in the statistical analysis is
composed of Bagging, JRip, PART, Ridor, J48, Random Forest, and REPTree.
As the experiments have only been executed on a data set, we firstly perform one-
way ANOVA analysis. We assume here as the H0 hypothesis the one of statistical
equivalence among the algorithms, i.e., we hypothesize that the difference between
the average values of all the algorithms is sufficiently low to be statistically not
significant. We run the test at a level of significance α = 0.05. This value designates
the highest admissible probability of wrongly rejecting a specified null hypothesis
that assesses the statistical equivalence in a case in which this hypothesis actually
holds true. Setting this value to 0.05 means that, in a case in which a given equivalence
hypothesis holds true, we have a probability equal to five percent of going wrong
and rejecting it.
Table 5 shows the results obtained with the ANOVA test.
Since the value of p-value is lower than α, the equivalence hypothesis H 0 is
rejected. This means that the difference between the averages of some algorithms is
sufficiently big to be significant from a statistical viewpoint, or, in other words, some
of the algorithms’ averages are considered to be not equal. Moreover, as p-value
equals 0.00000, the chance of type-1 error, i.e., rejecting a correct H0 , is small.
Once we have obtained this important information, we now wish to find out
which algorithms are statistically equivalent to which other algorithms. To this aim,
we consider each couple of the seven algorithms, and we run an unpaired t-test at a
level of significance α equal to 0.05. Here too, the hypothesis H0 is that of statistical
equivalence between the two algorithms being tested. Table 6 shows the complete set
of results. In it, the generic cell (i, j) reports two pieces of information. Firstly, in it,

Table 5 ANOVA table for the F-score values for the seven algorithms considered
Degrees of freedom Sum of square Mean square F statistic p-value
Groups 6 0.0150983 0.00251638 38.249764 0.00000
Error 168 0.0110524 0.0000657882
Total 174 0.0261507 0.000150292
286 G. Sannino and I. De Falco

Table 6 Complete t-test table for the F-score values

Bag JRip PART Ridor J48 RF REPT
Bag – < 0.0001 < 0.0001 < 0.0001 < 0.0001 0.46152 < 0.0001
(Y) (Y) (Y) (Y) (N) (Y)
JRip – – 0.01595 0.12191 0.01595 < 0.0001 (Y) 0.17851
(Y) (N) (Y) (N)
PART – – – < 0.0001 (Y) NaN < 0.0001 (Y) 0.00068 (Y)
(N)
Ridor – – – – < 0.0001 (Y) < 0.0001 (Y) 0.44462 (N)
J48 – – – – – < 0.0001 (Y) 0.00068 (Y)
RF – – – – – – < 0.0001 (Y)
REPT – – – – – – –

a number can be read that represents the p-value resulting from the t-test performed
over algorithms i and j. Secondly, a letter is reported that expresses whether the means
of these two algorithms are found by the t-test to be significantly different (Y ) or not
(N). In each cell, this letter is set based on the following: if in the cell the computed
p-value is lower than α = 0.05, then the statistical equivalence hypothesis H 0 can
be rejected; hence, the two algorithms are statistically different, and the letter is a Y;
otherwise, it is an N.
The results reveal that, for Bagging, the only algorithm for which the statistical
equivalence hypothesis cannot be rejected is Random Forest. The same conclusion
holds true for Random Forest as well. Hence, also the statistical analysis confirms
the statistical superiority of these two classifiers and their statistical equivalence as
well.
In the table, the cell for PART and J48 algorithms contains NaN because each of
them obtains the same F-score value over its 25 runs, and this value is the same for
the two algorithms, i.e., 0.784: In this case, NaN represents an extremely low value
meaning that the statistical equivalence hypothesis can be definitely accepted.

5 Summary/Conclusion

In this chapter, we have aimed at automatically identifying the different phases of

sleep for monitored subjects by taking advantage of machine learning capabilities.
To do this, we have shown here that it is sufficient to make reference to the signals
gathered from electrocardiogram (ECG) sensors placed on the subjects and suitably
process them so as to obtain heart rate variability (HRV) values. To prove this, we
have availed ourselves of the real-world Sleep Heart Health Study (SHHS) data set,
which is publicly available and contains information about both ECG signals and
sleep phases.
Use of Machine Learning Algorithms to Identify … 287

The main innovative feature of this chapter consists in a large set of different
machine learning algorithms being used to fulfill this task and in the comparison
of their effectiveness. They are contrasted in terms of classification ability with
reference to the F-score parameter because this latter is very significant when highly
unbalanced data sets are investigated, as is the case here.
The experiments performed in the WEKA tool and reported here have revealed that
Bagging turns out to be the best-performing and most-reliable algorithm, immedi-
ately followed by Random Forest. This has been confirmed by the statistical analysis
performed on the obtained results, which has informed us that the performance of
these two algorithms over this data set is statistically equivalent.
An important outcome of our investigation has been that many algorithms fail in
properly classifying the data set items: In fact, in many runs, at least one class is
assigned no item, which means that the lower-populated classes are not recognized.
This undesired behavior is shown by even some of the most widely used algorithms,
as for example Radial Basis Function, Support Vector Machine, AdaBoostM1, and
MultiLayer perceptron. Of course, this conclusion is true when the default parameter
settings reported in WEKA are used, as we have done here, and its generality should
be tested by varying them.
As concerns future work, we plan of adding Bagging and/or Random Forest
classifiers to a device, for example, a smartphone or a tablet. This can be firstly
trained offline on the problem, and then used online in an IoT-based fully automated
monitoring e-health system. In this latter, sensors gather, in real time, ECG signals
from a sleeping subject, send them to the device where data is processed, HRV
values are computed, sleep phase identification takes place through Bagging/Random
Forest, and medical personnel, close or not to the subject, are immediately and
continuously informed of the subject’s sleeping phases.

Acknowledgements The Sleep Heart Health Study (SHHS) was supported by National Heart,
Lung, and Blood Institute cooperative agreements U01HL53916 (University of California, Davis),
U01HL53931 (New York University), U01HL53934 (University of Minnesota), U01HL53937 and
U01HL64360 (Johns Hopkins University), U01HL53938 (University of Arizona), U01HL53940
(University of Washington), U01HL53941 (Boston University), and U01HL63463 (Case Western
Reserve University). The National Sleep Research Resource was supported by the National Heart,
Lung, and Blood Institute (R24 HL114473, RFP 75N92019R002).

References

1. Gubbi, J., Buyya, R., Marusic, S., & Palaniswami, M. (2013). Internet of Things (IoT): A
vision, architectural elements, and future directions. Future Generation Computer Systems,
29, 1645–1660.
2. Qadri, Y. A., Nauman, A., Zikria, Y. B., Vasilakos, A. V., & Kim, S. W. (2020). The future
of healthcare internet of things: A survey of emerging technologies. IEEE Communications
Surveys & Tutorials, 22, 1121–1167.
288 G. Sannino and I. De Falco

3. De Falco, I., De Pietro, G., & Sannino, G. (2020). Evaluation of artificial intelligence techniques
for the classification of different activities of daily living and falls. Neural Computing and
Applications, 32, 747–758.
4. Porumb, M., Stranges, S., Pescapè, A., & Pecchia, L. (2020). Precision medicine and artificial
intelligence: A pilot study on deep learning for hypoglycemic events detection based on ECG.
Scientific Reports, 10, 1–16.
5. Sannino, G., De Falco, I., & De Pietro, G. (2020). Non-invasive risk stratification of hyperten-
sion: A systematic comparison of machine learning algorithms. Journal of Sensor and Actuator
Networks, 9.
6. Montesinos, L., Castaldo, R., & Pecchia, L. (2018). Wearable inertial sensors for fall risk assess-
ment and prediction in older adults: A systematic review and meta-analysis. IEEE Transactions
on Neural Systems and Rehabilitation Engineering, 26, 573–582.
7. Castaldo, R., Melillo, P., Izzo, R., De Luca, N., & Pecchia, L. (2016). Fall prediction in hyperten-
sive patients via short-term HRV analysis. IEEE Journal of Biomedical and Health Informatics,
21, 399–406.
8. Pecchia, L., Bath, P. A., Pendleton, N., & Bracale, M. (2010). Web-based system for assessing
risk factors for falls in community-dwelling elderly people using the analytic hierarchy process.
International Journal of the Analytic Hierarchy Process, 2.
9. Porumb, M., Iadanza, E., Massaro, S., & Pecchia, L. (2020). A convolutional neural network
approach to detect congestive heart failure. Biomedical Signal Processing and Control, 55.
10. Längkvist, M., Karlsson, L., & Loutfi, A. (2012). Sleep stage classification using unsupervised
feature learning. Advances in Artificial Neural Systems.
11. Carskadon, M. A., & Dement, W. C. (2005). Normal human sleep: an overview. Principles and
Practice of Sleep Medicine, 4, 13–23.
12. Sharma, S., & Kavuru, M. (2010). Sleep and metabolism: an overview. International Journal
of Endocrinology.
13. Kwon, H., Yoon, H., Jung, D., Choi, S., Choi, J., Lee, Y., Jeong, D., & Park, K. (2017). Heart
rate variability in patients with major depressive disorder and healthy controls during non-REM
sleep and REM sleep. In 39th Annual International Conference of the IEEE Engineering in
Medicine and Biology Society (EMBC), pp. 2312–2315.
14. Tan, H. L., Kheirandish-Gozal, L., & Gozal, D. (2019). Sleep, sleep disorders, and immune
function. In Allergy and Sleep, pp. 3–15.
15. Roux, F., D’Ambrosio, C., & Mohsenin, V. (2000). Sleep-related breathing disorders and
cardiovascular disease. The American Journal of Medicine, 108, 396–402.
16. Iranzo, A., Santamaria, J., & Tolosa, E. (2009). The clinical and pathophysiological relevance
of REM sleep behavior disorder in neurodegenerative diseases. Sleep Medicine Reviews, 13,
385–401.
17. Hirshkowitz, M., Whiton, K., Albert, S. M., Alessi, C., Bruni, O., DonCarlos, L., Hazen, N.,
Herman, J., Katz, E. S., Kheirandish-Gozal, L., Neubauer, D. N., Ohayon, M., Peever, J.,
Rawding, R., Sachdeva, R. C., Setters, B., Vitiello, M. V., Catesby, W. J., & Adams, J. (2015).
National Sleep Foundation’s sleep time duration recommendations: methodology and results
summary. Sleep Health Journal, 1, 40–43.
18. Friedman, G. D., Cutter, G. R., Donahue, R. P., Hughes, G. H., Hulley, S. B., Jacobs, D. R., Jr.,
Liu, K., & Savage, P. J. (1988). CARDIA: Study design, recruitment, and some characteristics
of the examined subjects. Journal of Clinical Epidemiology, 41, 1105–1116.
19. Ni, H., Mingzhe, G. U. O., & Wang Y. (2020). Sleep Stage Classification based on two-cycle
sleep model recognition using continuous heart rate variability. Preprints.
20. Zhang, Y., Yang, Z., Lan, K., Liu, X., Zhang, Z., Li, P., Cao, D., Zheng, J., & Pan, J. (2019).
Sleep stage classification using bidirectional LSTM in wearable multi-sensor systems. In IEEE
Conference on Computer Communications Workshops (INFOCOM), pp. 443–448.
21. Goldammer, M., Weber, L., Malberg, H., Zaunseder, S. (2019). Individualized sleep stage clas-
sification from cardiorespiratory features. In 13th IEEE International Symposium on Medical
Information and Communication Technology (ISMICT).
Use of Machine Learning Algorithms to Identify … 289

22. Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological
Measurement, 20, 37–46.
23. Li, Q., Li, Q., Liu, C., Shashikumar, S. P., Nemati, S., & Clifford G. D. (2018). Deep learning
in the cross-time frequency domain for sleep staging from a single-lead electrocardiogram.
Physiological Measurement, 39.
24. Xiao, M., Yan, H., Song, J., Yang, Y., & Yang, X. (2013). Sleep stages classification based
on heart rate variability and random forest. Biomedical Signal Processing and Control, 8(6),
624–633.
25. Zhang, G. Q., Cui, L., Mueller, R., Tao, S., Kim, M., Rueschman, M., Mariani, S., Mobley, D.,
& Redline, S. (2018). The national sleep research resource: Towards a sleep data commons.
Journal of the American Medical Informatics Association, 25(10), 1351–1358.
26. De Falco, I., De Pietro, G., Della Cioppa, A., Sannino, G., Scafuri, U., & Tarantino, E. (2019).
Evolution-based configuration optimization of a deep neural network for the classification of
obstructive sleep Apnea episodes. Future Generation Computer Systems, 98, 377–439.
27. Quan, S. F., Howard, B. V., Iber, C., Kiley, J. P., Nieto, F. J., O’Connor, G. T., Rapoport, D. M.,
Redline, S., Robbins, J., Samet, J. M., & Wahl, P. W. (1997). The Sleep Heart Health Study:
design, rationale, and methods. Sleep, 20(12), 1077–1085.
28. Malik, M., Bigger, J. T., Camm, A. J., Kleiger, R. E., Malliani, A., Moss, A. J., & Schwartz,
P. J. (1996). Heart rate variability: standards of measurement, physiological interpretation and
clinical use. European Heart Journal, 17(3), 354–381.
29. Garner S. R. (1995). Weka: The waikato environment for knowledge analysis. In Proceedings
of the New Zealand Computer Science Research Students Conference, pp. 57–64.
30. Russell S., & Norvig P. (2002). Artificial intelligence: a modern approach. The Knowledge
Engineering Review.
31. John G. H., & Langley, P. (2013). Estimating continuous distributions in bayesian classifiers.
arXiv preprint arXiv:1302.4964.
32. Le Cessie, S., & Van Houwelingen, J. C. (1992). Ridge estimators in logistic regression. Journal
of the Royal Statistical Society: Series C (Applied Statistics), 41(1), 191–201.
33. Rumelhart D. E., Hinton G. E., & Williams R. J. (1988). Learning representations by back-
propagating errors. Cognitive Modeling, 5(3).
34. Broomhead D. S., & Lowe D. (1988). Radial basis functions, multi-variable functional inter-
polation and adaptive networks. Royal Signals and Radar Establishment Malvern (United
Kingdom), Tech. Rep.
35. Zeng Z.-Q., Yu H.-B., Xu H.-R., Xie Y.-Q., & Gao J. (2008). Fast training support vector
machines using parallel sequential minimal optimization. In 3rd International Conference on
Intelligent System and Knowledge Engineering, vol. 1. IEEE, pp. 997–1001.
36. De Gregorio, M., & Giordano, M. (2018). An experimental evaluation of weightless neural
networks for multi-class classification. Applied Soft Computing, 72, 338–354.
37. Aha, D. W., Kibler, D., & Albert, M. K. (1991). Instance-based learning algorithms. Machine
Learning, 6(1), 37–66.
38. Freund, Y., & Schapire, R. E. (1996). Experiments with a new boosting algorithm. In
Proceedings of the 13th International Conference on Machine Learning (ICML), vol. 96,
pp. 148–156.
39. Breiman, L. (1996). Bagging predictors. Machine Learning, 24(2), 123–140.
40. Holte, R. C. (1993). Very simple classification rules perform well on most commonly used
datasets. Machine Learning, 11(1), 63–90.
41. Cohen W. W. (1995) Fast effective rule induction. In Machine learning proceedings (pp. 115–
123). Elsevier.
42. Frank, E., & Witten I. H. (1998). Generating accurate rule sets without global optimization.
Computer Science Working Papers.
43. Compton, P., & Jansen, R. (1990). A philosophical basis for knowledge acquisition. Knowledge
Acquisition, 2(3), 241–258.
44. Quinlan J. R. (2014). C4. 5: Programs for machine learning. Elsevier
290 G. Sannino and I. De Falco

45. Breiman, L. (2001). Random forests. Machine learning, 45(1), 5–32.

46. Breslow, L. A., & Aha, D. W. (1997). Simplifying decision trees: A survey. The Knowledge
Engineering Review, 12(1), 1–40.
Emerging Technologies for Pandemic
and Its Impact

Dinesh Kumar Saini, Sandhiya Bala, Akhilesh Kumar Sharma,

and Kashif Zia

Abstract This is the era of modern and highly sophisticated research which is on-
going at various geographical locations and there are so many consequences that
are also associated with the security and ethical responsibility. Hence knowingly or
unknowingly the human civilization would face the kind of traumas and pandemics
which has already been happened in the recent past. This chapter discusses these
kind of consequences and their remedies with the latest technological developments
whether it is the security, tracking, IOT based healthcare systems, healthcare semi-
nars, conferences, etc. The healthcare systems touches every acumen of the human
life and impact greatly the day-to-day human life. The tracking, surveillance and
monitoring, etc. are not that difficult due to the recent advancements and IOT based
researches. The robotic diagnostic arm system, robotic patrolling and surveillance,
and cobotics are the ideal inventions of the modern era by which the medical prac-
titioners now easily can monitor, diagnose as well as do the surgery to the remotely
placed patient or the medication seekers sitting at geographically distant places.
Technological interventions helped human life easier, but at the same time when the
minute level of mistakes will create the havocs and its capable of and can destroy the
entire human race in very less time just like Covid’19 is the latest example. These
pandemics are very difficult to control, hence, there is an utmost care needed, while
developing and performing the research with these viruses whether to control, test,
vaccine researches, trials on these vaccines, etc. The pandemics and their threat to
the humans and any living being are dangerous in any form. The entire world shown
the united support and courage to control the Covid’19 as the latest pandemic, but it
is very necessary that people understand the necessity to act wisely, this chapter also
discusses if the situations wrong and how to control the lockdown situations using

D. K. Saini · A. K. Sharma (B)

Manipal University Jaipur, Jaipur, India
S. Bala
Nielsen Global Connect Chennai, Chennai, India
K. Zia
Sohar University Oman, Sohar, Oman
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 291
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_14
292 D. K. Saini et al.

sophisticated surveillance using drone cameras, IOT devices, robotics, unmanned

vehicles, etc. used to control stampede and festive gathering or any other kind of
people gatherings. Advanced people monitoring and machine learning based predic-
tions systems utilized in order to reduce such situations where the gathering control
is difficult. These devices helping the government and the police cops to perform
the duties and to avoid any such situations where this is uncontrollable. The world
economy and mental health aspects, lockdown control and employee sanitization
automation systems are also discussed in this chapter.

Keywords Pandemics · Covid’19 · Health care devices · IoT · Surveilance ·

Cobotics · Healthcare predictions · Diagnostics

1 Introduction

Up until now, the world has seen a pandemic so rarely that enough actions have not
been taken by governments for being adequately equipped for such crises. COVID’19
could be the big wake-up call that the world needs to prioritize the health of earthlings.
If you have lived in the year 2020, you would not need an introduction. For the
others, an epidemic that is spread across multiple countries and continents is what
we call a pandemic.
But defining it as ‘dealing with a global-spreading disease’ does not precisely
describe this scenario. Different governments use all their intelligence and expertise
to formulate their battle against a pandemic. Ample health care, lockdowns, quaran-
tines, fighting the falling economy, public surveillance, are a few from the long list
of things-to-deal-with for governing bodies during a pandemic [1].
On 26 March 2020, The UN secretary-general, António Guterres said in his
response to COVID’19—“Different countries have followed different, sometimes
contradictory strategies and we are all paying a heavy price”. It is indeed necessary
for different countries to come together to formulate a plan-of-action that is generic
enough and can be adopted without many conflicts.
The world is quite familiar with pandemics [2]. There is a whole decade between
the 2009 H1N1 pandemic and 2020. We also saw the West-African outbreak of Ebola
Virus from 2014 to 2016. We clearly need to streamline and pace up the development
of firm strategies for mitigation of a global medical emergency.
The existence of a standard, generalized pandemic management framework with
directives for different scenarios of severity would greatly shorten the pandemic
life-cycle [3]. Such a framework would involve efficiently tackling every trivial to
complex aspect of this global crisis. To name a few of these aspects—distribution
and accessibility of healthcare and supplies, safety of citizens, remote surveillance,
economic pitfalls, student education, travelers’ safeguarding and quarantining, etc.
Undoubtedly, much of the emerging technologies could be brought into play to build
a robust system that will save the nations’ time of discussion of “how to proceed
now?” and help to keep life going as normal as possible [4].
Emerging Technologies for Pandemic and Its Impact 293

The rest of the chapter discusses different emerging technologies that can be
utilized during a pandemic. With the service of such engineering marvels, any global-
spreading disease would no longer be an alarming threat to the basic functioning of
the world.
• Technology for Surveillance
– Location Data
– Health tracking mobile applications
– Robotic diagnostic system
– Robotic patrolling system.
• Technology for Healthcare
– 3D printing supplies
– Advanced isolation cubicles with automation (to avoid space contingency)
– Autonomous vehicles
– Cobotics for Treatment.
• Technology for Economy
– 3D remote work with XR (AR or VR)
– Sanitization Systems for Essential Workers.
• Technology for Lockdown
– Drones for goods delivery
– AI based entertainment streaming methods
– Automation for Cleaning.
• Technology for Education
– Interactive Mixed Reality (MR) classrooms
– AI for analyzing student mental health

2 Surveillance

2.1 Location Data

Being able to detect where people are could prove to be of great help during a
pandemic. It can help track gatherings, track the geographical virus prone areas,
identify traffic, unnecessary public movement, understanding the effectiveness of
social distancing, etc.
Location data can be harnessed in different ways:
• Using GPS (Global Positioning System), a feature enabled in almost every smart
device.
• Mobile network towers that can approximately infer the location of a device.
294 D. K. Saini et al.

• Using Wi-Fi Network’s databases that contain the unique identifiers of wireless
routers and their known locations (of the connected devices).
• Use of IoT (Internet of Things) like Bluetooth Beacons attached to devices
at different locations and mobile apps could send proximity based location
information.
However any such system must be guaranteed by robust security and protection of
human rights. The question arises about who the authorities for safeguarding civilian
sensitive information? The chain of responsibilities to safeguard sensitive civilian
information begins with the institutions that have the access to location data, like:
• Mobile network operators or the “SIM-card providers”.
• Mobile operating system providers like IoS, Android, etc., mobile applications
use multiple sensors within the handset to provide precise location data [5].
• Location analytics providers who store data of connected devices (Internet of
things) like mobile phones, smart watches, vehicles, etc. This data is used by
airports, retailers, etc., for tracking their busy hours in a non-pandemic time.
Quick amassing of structured location information would greatly uncomplicate
the planning, research and mitigation of a pandemic with other technology [6].

2.2 Health Tracking Mobile Applications

There are plenty of fitness apps that are widely used for maintaining a health regime,
making and tracking diet plans, keeping a tab on antibiotics and drug intakes, etc.
However, what we need for a pandemic is a full-fledged body and mental health
tracking system that is an all round personalized doctor.
Such an application would help in being precautious and hence avoiding
unanticipated health risks.
The relevant technologies that such an application could include:
• Advanced Body Mass Index calculator,
• Blood pressure metrics calculator with oscillometric finger pressing method, i.e.,
to sense the blood flow by pressing the finger on a mobile camera and measuring
factors like blood pressure. In [7], the authors have developed a mobile application
with the oscillometric finger pressing method that requires no extra hardware
(Fig. 1).
• Blockchain to prevent all types of vulnerabilities and threats that concern the
security of user medical data.
• Deep learning methodologies for best treatment decisions of a patient computed
using electronic health records [8].
• Artificial intelligence for virtual medical assistance personalised for individuals.
Emerging Technologies for Pandemic and Its Impact 295

Fig. 1 Mobile tracking of health [7]

2.3 Robotic Diagnostic System

The ultimate concern during a pandemic is the prevention of spreading of the

disease. Technology could be used at every point of human contact. As doctors and
other medical employees would be the most vulnerable, they can be protected from
unknowingly catching the disease causing microbes during the diagnosis of a patient.
The small proportion of doctor to patient during a pandemic raises concerns over
a medically qualified human to be available to attend patients at all times. Robotic
systems are a viable solution to this problem.
How exactly could we expect this to work?
Diagnostic arm’s are not new to the research world. However, the concept seems
to be neglected for implementation because of the lack of need.
In [9], the researchers present a tele-diagnostic system with a robotic arm for
remote monitoring of patients with endemic deadly diseases. The highlight here is
that the proposed system requires low cost of acquisition and almost no interaction
between patient and medical attendant.
This system proposes a 3 degree of freedom (DOF) robotic arm that allows moni-
toring body vitals with no or minimal contact between patients and health workers
using sensors and motors.
If we look at the device in two perspectives—mechanical and electrical, we under-
stand that the arm uses servo motors to achieve the mechanical movements and
sensors programmed for interaction and monitoring of patients health.
296 D. K. Saini et al.

Fig. 2 a Robot arm free body diagram, b basic robot arm design

The project also flaunts a graphic user interface (GUI) developed for effective
visualization of the patient’s body temperature (Fig. 2).
Such research and development can be utilized to maintain safety of the physician
by keeping the doctor distanced from the affected yet providing maximum control
to the physicians themselves during the diagnosis.

2.4 Robotic Patrolling System

Another set of vulnerable essential workers are the Policemen who keep the cities
under control and surveillance during a lockdown. Robotics may come in handy not
only to reduce human to human proximity but also to make the surveillance and
patrolling system all the more efficient.
A connected robotic patrolling system could also help in quick amassing of accu-
rate data of large gatherings, lockdown violators, etc. We have already discussed how
constructive such information can be in tackling a pandemic.
There are different forms of patrolling that a pandemic demands.
• Patrolling on roads for restriction of unnecessary movement.
• Surveillance in and around hospitals, clinics and Pharmacies.
• Patrolling at city borders with special restrictions.
• Patrolling at public areas like parks, theaters, malls, etc.,
• Monitoring and dispersing crowds.
Emerging Technologies for Pandemic and Its Impact 297

A governing body can include many more layers of patrolling other than the
above. However, the key here is that we utilize Robotics to minimize or completely
eradicate the requirement of human patrol officers.
Video surveillance robots are already in the market. Here’s a couple of examples
that might interest you:
• In [10] SMP Robotics have developed the Security Robots of S5.2 with exceptional
video surveillance. They come with numerous features that make them a great
candidate for a situation like the one under discussion here.
To mention a few:
– Low-light mobility
– Obstacle avoidance
– 360° panoramic video surveillance
– Knowledge and rectification of path deviation
– Intelligent collaborative patrolling
– Audible warning system (Fig. 3).
• In [11] The Knightscope K-5 has been commodiously used at the Pechanga Resort
and Casino in the US for enhanced security in-addition to it’s human security staff
and installed security cameras.
Few of the many impressive capabilities are:
– Inflicting deterrence to defectors in the patrolling area.
– Vehicle license plate recognition
– Human-to-human two-way intercom

Fig. 3 Automated support vehicle for radio surveillance [10]

298 D. K. Saini et al.

Fig. 4 La–fi travel briefcase [12]

– Remote monitoring using the user-interface\

– Broadcast of custom messages/announcements (Fig. 4).
Patrolling can thus be completely free of human physical interaction, hence
bringing us one step closer to the end of a pandemic.

3 Healthcare

3.1 3D Printing Supplies

The medical and pharmaceutical industry cannot possibly have in-stock or quickly
manufacture supplies during the pandemic. Countries would be frantically looking
for solutions to suffice the surge in demand for drugs, protection kits, medical tools,
etc. This would not be a problem with the introduction of 3D printing machines that
could be programmed to run 24 × 7 and produce for the bulk requirement of medical
equipment and antibiotics [13].
3-D printing is rapidly gaining popularity in the medical field.
• A team at Duke University, USA during the 2020 COVID Pandemic, designed
a 3D-printable modification to unused surgical helmets which could be used as
an alternative to PPEs (Personal Protective Equipment) for the pandemic doctors
[14]. This proves to be a potential solution to the shortage of PAPRs (powered
air-purifying respirator) and hence keep physicians out of danger.
Emerging Technologies for Pandemic and Its Impact 299

• The University of Nebraska Medical Center (UNMC), USA designed a protective

face shield for protecting physicians especially anesthesia providers who are at
greater risk during shortage of PPEs in a pandemic. This shield is constructed by
a combination of 3D-printing and assembling commonly available products. It
has also been designed for decontamination, hence reusability [15].
The key advantage of 3D printing that makes it a promising technological solution
for tackling a pandemic is the reduction in production cost and time that enables
engineers to quickly develop and test prototypes with multiple configurations and
pick the apt one [16].

3.2 Advanced Isolation Cubicles with Automation

The infrastructure of a city cannot accommodate hospital facilities that are robust
enough to attend for a pandemic. The influx of patients and the clinical space required
by them can be unpredictable. A smart and ad-hoc solution would be to use isola-
tion cubicles that can be temporarily developed anywhere, hence providing medical
infrastructure to people during a pandemic without disturbing the city-plan.
‘MediCab’ developed by a start-up in India is a portable microstructure that was
launched in Kerala’s Wayanad district in 2020 for treating COVID’19 patients living
in remote areas. The structure is foldable and has 4 zones that facilitates isolated
treatments with professional medical attenders [17] (Fig. 5).

Fig. 5 Medicab facility att Kerala Waynads location [17]

300 D. K. Saini et al.

Such an innovation can be enhanced for a pandemic with automated medical

assistance in the form of.
• smart monitoring systems
• cobotics for doctors to treat patients without physical contact
• autonomous portability of the isolation cubicles, etc.
Such technology would not only largely help curb a pandemic but also bring a
revolution in contemporary health infrastructure.

3.3 Autonomous Vehicles

Despite lockdowns, transportation of people and goods cannot be completely elim-

inated due to exceptions like patients, doctors, medical bulk equipment delivery,
government officials, essential workers. Vehicles today require a human driver for
most of this transportation. In order to minimize the number of human movement,
autonomous vehicles could put the entire driver population out of vulnerability.
Although technology of driverless transportation is gaining popularity, it is still
uncertain if autonomous ambulances would be welcomed. In [18], the authors have
recorded the lack of willingness by patients to travel in a driverless ambulance using
multiple factors. Besides, ambulances could be enhanced by connecting them to a
5G network for easy, collaborative tracking and management.
Self-driving vehicles can therefore be developed for the purpose of goods trans-
portation and in a pandemic perspective for sterile medical supplies transportation.
This largely reduces the number of essential workers required in the healthcare sector,
thereby reducing the number of people at risk. Autonomous vehicles designed to
carry goods rather than human passengers is potentially a near-future possibility and
a great technology to fight pandemics.

3.4 Cobotics for Treatment

A robot assisted by humans is essentially the best usage of AI for treating patients
without endangering the doctors and specialists. This is where the technology of
Cobotics fits perfect which is radically gaining popularity.
There could never come a time when a machine can be let to independently treat
patients. Cobotics could be used during a pandemic or could also become a new
normal in the medical industry for keeping doctors from contracting the disease,
utilising expertise from anywhere across the world, and many more applications.
Cobotics because of the concept of partial automation empowers the human who
is to perform a task rather than replacing them. Their skill and instincts are preserved.
The usage of cobots has already been identified in several aspects of healthcare.
Emerging Technologies for Pandemic and Its Impact 301

Fig. 6 Robot guided COBOT for laser bone ablation [20]

In [19] AOT’s CARLO® (Cold Ablation Robot-guided Laser Osteotome) system

uses a cobot which is tactile robotic to perform a medical procedure called laser
bone ablation. The system uses cold laser technology to cut bone without contact
[20] (Fig. 6).
• In [19] A neurosurgical cobot has been developed by Synaptive Medical, a
medical technology company. This cobot enables surgeons to microscopically
position their surgical view and thus increase precision. It comes with multiple
features including advanced visualization with 3D, voice-activated control,
enabling heads-up operations, etc. [https://www.synaptivemedical.com/products/
modus-v/].
Other healthcare areas where cobotics can help could be sterilized medical
supplies and tools packaging, automation in labs for medical tests, automated
massaging tools [19], the post-mortem and forensics department, etc.

4 Economy

4.1 3D Remote Work with XR (AR or VR)

The impact on the employee class of the society during a pandemic must not be
neglected at any cost. This is because that very impact rolls over to other sections of
the economy in unprecedented ways. To give an example, if an employee is unable
302 D. K. Saini et al.

to work, his income reduces. As a result his role as a consumer of goods and services
becomes smaller and this in-turn affects the supply chain. It is obvious how crucial
it is to keep a community’s economic cycle going during the pandemic while the
population stays safe and quarantined.
Most work these days already requires some form of digital support. Remote work
is when such digital support for productivity and communication is available at the
employee’s personal location. “Work from home” is common among the working
population. However, remote work is not generally recommended as it lacks the
collaborative experience which is a major factor necessary for productivity and tech-
nology today is not utilized to an extent to provide the infrastructure that office work
provides.
Although workplace employment cannot be completely banished for cultural and
emotional reasons of the society, a global pandemic may put forth a need for extended
remote work. To make long work-from-home periods encouraging and effective, we
look at what makes it difficult.
• The lack of physical collaboration.
• Uncomfortable virtual meetings
• The feeling of immobility for the employees
• Monotonous routine
• Cybersecurity and maintaining intellectual property of organizations Ahmad,
Tabrez [21].
Now we look at few ways we can overcome those issues and keep the productivity
growing despite the pandemic:
• Enhancing virtual conferences and meetings using augmented reality for better
communication with a sense of physical collaboration.
• Blockchain for security of information.
In [22], the authors introduce us to “Hawk” which is a framework built with a
blockchain based-model of cryptography. Hawk enables parties to record transac-
tional information and keep it out of the public’s view. These “Smart Contracts”
provide two forms of security, the on-chain privacy and the contractual security, i.e.,
neither the public nor the parties involved in making the contract can autonomously
manipulate or control the information stored in the contract. The Hawk program
encrypts the private data to the blockchain hence sealing it from any forgery.

4.2 Sanitization Systems for Essential Workers

The demand for certain human-to-human interactions cannot be denied despite the
automation and advancement of technology. An important instance of this would be
the meetings of governing authorities, health professionals and medical researchers
who would need to work collaboratively without any form of hindrance to find a
Emerging Technologies for Pandemic and Its Impact 303

Fig. 7 Disinfectant walkthrough tunnel installed in Islamabad, Pakistan [24]

solution to the global emergency. Sanitization of these professionals can be tech-

nologically enhanced to decrease human as well as medical resource consumption
during this process.
A few effective methods could be:
• Automated fumigation can help keep inanimate things risk-free.
• A more economical solution would be a Walkthrough sanitization system with
which we could keep essential workers out of danger [23]. These gates could
be easily installed in public places and industrial areas. These walk-in chambers
spray disinfectant solution on the individual who passes (Fig. 7).
• Cobots could be used to work along with workers to keep them protected yet effec-
tively utilising their human intelligence. In [24], we get a glimpse of how robotics,
Internet of Things (IoT), locationing and augmented reality (AR) technologies aid
to preserve productivity in a warehouse setting. It is crucial to understand this can
and needs to be done without eliminating the business processes that require
human intelligence and intuition.

No solution can be proclaimed to be generic for diseases across the pathology

spectrum. Each pandemic may demand for a different sanitization technique [23]. The
decision-making bodies must be informed well early of the resources and innovative
solutions already available to us to act quickly and curate the most suitable way out
of an emergency [24].
304 D. K. Saini et al.

5 Lockdown

Lockdown is a restrictive order placed generally by a governing body to confine

the inhabitants of a society within their households for a specific period of time.
A lockdown could be practiced for various political, social or economic reasons.
During a pandemic it would be to curb the spread of a communicable disease.

5.1 Drones

As we discussed in the previous section about keeping the economy cycle moving,
for anything to ‘move’ in the economy, logistics need to function seamlessly. Only
a global emergency that would put everything to a halt would make one realize how
strong of a pillar is logistics to a smooth functioning economy. To save the day when
goods transportation becomes difficult due to lack of workforce, we have Drones.
Drones are automated aerial vehicles that come with features that are almost perfect
for our crisis here. A drone which is unmanned, is aerially mobile, can carry at least
2 kg of goods, illumination capable, has surveillance features, and driven by solar
power can aid battling a pandemic in numerous ways [25, 26, 27, 39]:

Drones for delivery of medical supplies, essentials like groceries, etc.

Surveillance drones for crowd monitoring with a loud speaker for making
announcements and first aid kits for emergencies.
Drones for fumigation and disinfection of inanimate things and geographical
areas.
Mass temperature screening of individuals with thermographic imaging drones
for periodic check-ups of people thus efficient identification of affected people
(Fig. 8).

5.2 AI Based Entertainment Streaming

Lockdowns and extended quarantines can take a toll on people’s entertainment

routines. Considering the seriousness of a global pandemic, this might seem like
a petty topic to discuss, but it is indeed necessary to understand that paying attention
to such things may help a great deal in retaining citizen cooperation. If enhancing
the streaming technology could help in controlling collective human agitation when
most necessary, then why not?
In [28] Research shows that during the COVID-19 viral pandemic, there has been
a great reach of YouTube with partners in the entertainment industry for creating
awareness among the public about the severity of the pandemic and their role in it’s
mitigation.
Emerging Technologies for Pandemic and Its Impact 305

Fig. 8 Remote location surveillance drones [25]

There could be many opposition and challenges to using entertainment media for
the mitigation plan of a pandemic due to various reasons:
• The addiction caused by media and the subsequent health issues
• The threat of audience being misled by half-baked/wrong/ambiguous information
• The requirement of private and/or sensitive information about the viewers and the
possible risk of its misuse, etc.
By consciously attending to the challenges and developing trust among the
users with proofs, guarantees, efficient customer support and clear user-rights, such
advancement in the entertainment industry particularly to battle a pandemic would
help cater to the mental state of the audience. This can then be utilized to widen the
viewers’ wisdom about the situation, help prevent collective agitation, etc.

5.3 Automation and Innovation in Cleaning

Nothing, not even a global lockdown, can stop the inhabitants of this planet from
producing waste. Waste treatment and management could see major setbacks during
a pandemic due to various factors like unavailability of labor, logistics, cost, etc.
In [29], the authors talk in detail about the severity of waste management chal-
lenges during a pandemic as there can be different waste material contaminated with
306 D. K. Saini et al.

Fig. 9 Waist management systems in various studies [29]

the endemic disease. It is evident that there is a large scope for automation and AI
in this sector. The potential areas may be but not limited to:
• Smart waste collection and material-wise separation
• Automated biomedical and/or hazardous waste treatment
• Intelligent domestic waste management systems
• Robotic cleaners for public areas.
The COVID’19 has already seen a rapid technological advancement in the waste
management industry. The following image shows the various practices adopted to
battle the global pandemic caused by the Coronavirus (Fig. 9).
Automation in waste management could create a form of safety in the society, not
just during a pandemic but always. By curbing the interaction of humans with waste,
one is able to stop the spread of trash-borne diseases [30].

6 Education

The economy is not the only cycle that needs to be kept going during a pandemic.
Such emergencies can be long in an unprecedented manner. The nurturing of the
budding generation cannot be compromised. Education systems’ authorities must
take the planning and execution of solutions into their own hands rather than waiting
for a government’s order to follow by. In this era, technology can be kept ready to be
utilized when educational services cannot reach the students by conventional ways
[31].
Emerging Technologies for Pandemic and Its Impact 307

6.1 Interactive Mixed Reality (MR) Classrooms

Online classes are not effective because of various reasons, mainly the inability of
students to maintain self-discipline. Toddlers and younger kids are the toughest to
reach with online classes, yet they have the most brain growth rate than any other
age-group. Hence they need to be given equal importance if not more in terms of the
quality of education they are provided with.
One cannot deny, education is affected across the age spectrum of students.
Customized innovative solutions that pertain to each field of study and age-group
is possibly the best way to attend to the issue. The surgical residency programs is
one such example. Solutions like social-media based platforms for daily practice
questions and discussions, simulation softwares for Operating Room courses, etc.,
are a few examples that are specific to the field and hence could prove extremely
efficient [32].
Moreover, virtualization of schooling and quarantining restrictions would lead to
students losing their focus and the little zeal for going to school brought by the P
E classes. Courses like P.E, yoga etc., that need interaction that plans further than
audio and video are generally compromised during virtual schooling [33].
Mixed reality is bringing together the physical and the digital world by producing
virtual, simulated and interactive environments and visualizations to provide a real-
time experience to the user.
MR technology would enable students and teachers to interact and learn from
simulated objects from their curriculum thus gaining practical knowledge. For
example, students could learn about the revolution of planets by physically moving a
virtual planet around a virtual sun. It is also identified that physical body movement
aids the learning capacity of students [34] (Fig. 10).

Fig. 10 AR-VR and mixed reality technology in schools for teaching during pandemic and normal
time [35]
308 D. K. Saini et al.

In a pandemic, a mixed reality classroom may be enhanced with teacher and

student virtualization with avatars along with other elements of a classroom to provide
the students a sense of physical presence and interaction. Such an immersive learning
platform could easily reduce the effect that a pandemic would cause on students’
schooling.

6.2 AI for Analyzing Student Mental Health

Children may not take extended home quarantine in the most calm way. A pandemic
situation could be extremely strange and could be difficult to process for a school-
going child. They may not be completely expressive about their concerns or the
adults might be unable to comprehend their ways of putting forth their concerns.
Since schooling would inevitably be virtualized during those times, the computer is
basically what the kids would be interacting with the most [36]. Keeping aside the
fact that this exposure needs to be reduced as much as possible, we could utilize it
to determine if the youngster is in any way uncomfortable with the sudden set-up
alterations [37, 27, 39].
The scope of understanding a student’s mental state still seems vague, but it could
be game changing. From a surge in learning and upskilling capability to immediate
recognition of students in need of treatment, there are multiple positive outcomes of
feeding the right kind of knowledge according to the mind’s state [38].

7 Conclusion

Despite any advancement in the tech industry, usability and security will be a constant
vulnerability. It is important that new tech systems must be developed with a “people
first” mentality. Designers of such digital transformation must thoroughly analyze
the capability of users to effectively as well as safely utilize it. The accessibility of
technology must encompass enabling users to completely understand and control
their data and how it is used.
The crux of technology to fight a pandemic would be automation. It is crucial to
identify the right scenarios that truly require automation. It so happens many-a-times
that automation devices increase the need for human labor in forms of supervision,
operation, troubleshooting, etc. This is the paradox of automation and is not useful
in any sense.
It should be effectively noted that the scope of above technologies are not limited
to the mentioned applications. Every pandemic is generic in a few ways and unique
in many. Technology could be cleverly planned and applied with the sole goal
of reducing human effort and interaction, while maintaining integrity, safety and
security.
Emerging Technologies for Pandemic and Its Impact 309

References

1. Whitelaw, S., et al. (2020). Applications of digital technology in COVID-19 pandemic planning
and response. The Lancet Digital Health.
2. Quinn, T. C. (1996). Global burden of the HIV pandemic. The Lancet, 348(9020), 99–106.
3. Nacoti, M., et al. (2020). At the epicenter of the COVID-19 pandemic and humanitarian crises
in Italy: Changing perspectives on preparation and mitigation. NEJM Catalyst Innovations in
Care Delivery, 1(2).
4. Shaw, R., Kim, Y., & Hua, J. (2020). Governance, technology and citizen behavior in pandemic:
Lessons from COVID-19 in East Asia. Progress in Disaster Science, 100090.
5. Sonn, J. W., Kang, M., & Choi, Y. (2020). Smart city technologies for pandemic control without
lockdown. 1–3.
6. Dong, E., Hongru, Du., & Gardner, L. (2020). An interactive web-based dashboard to track
COVID-19 in real time. The Lancet Infectious Diseases, 20(5), 533–534.
7. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6120863/.
8. Norgeot, B., Glicksberg, B. S., & Butte, A. J. (2019). A call for deep-learning healthcare.
Nature Medicine, 25(1), 14–15.
9. Folorunso, T. A., Olaniyi, O. M., Nuhu, B. K., Bima, M. E., & Ajani, O. S. Development of
robotic arm based diagnostic system for clinical tele-monitoring.
10. https://smpsecurityrobot.com/products/.
11. https://www.knightscope.com/knightscope-k5.
12. https://www.latimes.com/business/la-fi-travel-briefcase-pechanga-robots-20180706-story.
html.
13. Belhouideg, S. (2020). Impact of 3D printed medical equipment on the management of the
COVID-19 pandemic. The International Journal of Health Planning and Management.
14. Erickson, M. M., et al. (2020). Helmet modification to PPE with 3D printing during the
COVID-19 Pandemic at Duke University Medical Center: A novel technique. The Journal
of Arthroplasty.
15. Armijo, P. R., et al. (2020). 3D printing of face shields to meet the immediate need for PPE in
an anesthesiology department during the COVID-19 pandemic. American Journal of Infection
Control.
16. Attaran, M. (2020). 3D printing role in filling the critical gap in the medical supply chain during
COVID-19 pandemic. American Journal of Industrial and Business Management, 10(05), 988.
17. https://indianexpress.com/article/education/iitians-develop-portable-hospital-units-to-treat-
covid-19-patients-6511644/.
18. Winter, S. R., et al. (2018). Driverless ambulances: A possibility, but will patients ride? In
Proceedings of the Human Factors and Ergonomics Society Annual Meeting (Vol. 62(1)).
SAGE Publications.
19. https://blog.robotiq.com/5-ways-cobots-are-used-in-medicine-and-healthcare.
20. https://aot.swiss/en/carlo/.
21. Corona Virus (COVID-19) Pandemic and work from home: Challenges of cybercrimes and
cybersecurity. Available at SSRN 3568830.
22. Kosba, A., et al. (2016). Hawk: The blockchain model of cryptography and privacy-preserving
smart contracts. In 2016 IEEE Symposium on Security and Privacy (SP). IEEE.
23. Mallhi, T. H., et al. (2020). Walkthrough sanitization gates for COVID-19: A preventive
measure or public health concern? The American Journal of Tropical Medicine and Hygiene,
tpmd200533.
24. https://www.zebra.com/us/en/blog/posts/2019/co-bots-working-alongside-people-in-wareho
use.html.
25. Chamola, V., et al. (2020). A comprehensive review of the COVID-19 pandemic and the role
of IoT, drones, AI, blockchain, and 5G in managing its impact. IEEE Access, 8, 90225–90265.
26. Bhoi, A. K., Mallick, P. K., Liu, C. M., & Balas, V. E. (Eds.) (2021). Bio-inspired
neurocomputing. Springer.
310 D. K. Saini et al.

27. Oniani, S., Marques, G., Barnovi, S., Pires, I. M., & Bhoi, A. K. (2020). Artificial intelligence for
Internet of Things and enhanced medical systems. In Bio-inspired neurocomputing (pp. 43–59).
Springer.
28. Basch, C. E., et al. (2020). The role of YouTube and the entertainment industry in saving
lives by educating and mobilizing the public to adopt behaviors for community mitigation of
COVID-19: Successive sampling design study. JMIR Public Health and Surveillance, 6(2),
e19145.
29. Sharma, H. B., et al. (2020). Challenges, opportunities, and innovations for effective solid waste
management during and post COVID-19 pandemic. Resources, Conservation and Recycling,
162, 105052.
30. Wang, J., et al. (2020). Disinfection technology of hospital wastes and wastewater: Suggestions
for disinfection strategy during coronavirus Disease 2019 (COVID-19) pandemic in China.
Environmental Pollution, 114665.
31. Conroy, M. A., Durrheim, D. N., & Dalton, C. B. (2008). Likely impact of school and childcare
closures on public health workforce during an influenza pandemic: A survey. Communicable
Diseases Intelligence Quarterly Report, 32(2), 261.
32. Chick, R. C., et al. (2020). Using technology to maintain the education of residents during the
COVID-19 pandemic. Journal of Surgical Education.
33. Laddu, D. R., et al. (2020). Physical activity for immunity protection: Inoculating populations
with healthy living medicine in preparation for the next pandemic. Progress in Cardiovascular
Diseases.
34. Lindgren, R., & Johnson-Glenberg, M. (2013). Emboldened by embodiment: Six precepts for
research on embodied learning and mixed reality. Educational Researcher, 42(8), 445–452.
35. https://www.indiatoday.in/education-today/featurephilia/story/role-of-augmented-virtual-rea
lity-in-education-1417739-2018-12-26.
36. Wiederhold, B. K. (2020). Connecting through technology during the coronavirus disease 2019
pandemic: Avoiding “Zoom Fatigue”. 437–438.
37. Ferdig, R. E., et al. (2020). Teaching, technology, and teacher education during the COVID-19
pandemic: Stories from the field. Association for the Advancement of Computing in Education
(AACE).
38. Mohr, D. C., Zhang, Mi., & Schueller, S. M. (2017). Personal sensing: Understanding mental
health using ubiquitous sensors and machine learning. Annual Review of Clinical Psychology,
13, 23–47.
39. Marques, G., Bhoi, A. K., de Albuquerque, V. H. C., K.S., H. (Eds.) (2021). IoT in healthcare
and ambient assisted living. Springer.
Impact of Artificial Intelligence in Health
care: A Study

Devyani Bairagya, Hrudaya Kumar Tripathy, Akash Kumar Bhoi,

and Paolo Barsocchi

Abstract The acceptance of AI in health care relates to the analysis of the huge
amount of information that is generated each day and the limitation of physicians
to address these needs. The growth of data complexity in the medical domain refers
to the increasing usage of artificial intelligence in that sector. Some vital func-
tionaries include care providers, diagnostic recommendation systems, and adherence
of patients among others. Likewise, there exist several applications where AI can be
successfully deployed in clinical applications. In this study, the role of AI in this
critical healthcare sector is highlighted. Some popular existing research works in the
healthcare domain are discussed. Software projects involving AI in this sector are
summarized. Finally, a real-time implementation of medical imaging using different
computational methods is demonstrated. Maximum accuracy of 94.2% was noted
for prostate cancer. It is also found that it takes maximum time for the analysis of
lung cancer (2.43 s) and minimum for brain cancer (1.12 s).

Keywords Artificial intelligence (AI) · Health care · Machine learning · Medical

imaging · Classification accuracy

1 Introduction

Artificial intelligence is the study and design of intelligent models and programs that
mimic the human mind, that is, to think, reason, and perform tasks that are similar
to the human brain. From Cortana and Siri to drones and driverless cars, artificial

D. Bairagya · H. K. Tripathy (B)

School of Computer Engineering, Kalinga Institute of Industrial Technology (KIIT), Deemed To
Be University, Bhubaneswar, Odisha, India
A. K. Bhoi
Department of Computer Science and Engineering , Sikkim Manipal Institute of Technology
(SMIT), Sikkim Manipal University (SMU), Majitar, Sikkim 737136, India
A. K. Bhoi · P. Barsocchi
Institute of Information Science and Technologies, National Research Council, 56124 Pisa, Italy
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 311
A. Kumar Bhoi et al. (eds.), Hybrid Artificial Intelligence and IoT in Healthcare,
Intelligent Systems Reference Library 209,
https://doi.org/10.1007/978-981-16-2972-3_15
312 D. Bairagya et al.

Fig. 1 Few applications of

AI

intelligence (AI) has been advancing rapidly. AI includes everything from Bing and
Google to the autonomous military. AI is an extremely broad field that consists of
several sub-fields [1]. In today’s fast-paced world, much of the world’s information
is produced by individuals and machines. This transcends our ability to understand,
interpret, and make sound decisions. AI is the future of all decision making processes
as AI-powered algorithms are highly efficient at calculating combinations and going
through multiple decisions to arrive at the best conclusion [2, 3]. AI utilizes repetitive
learning and discovery through data and also incorporates intelligence into existing
machines. It performs frequent computerized tasks reliably and without weariness.
For example, Siri was an added feature to a new set of Apple products. AI adapts
by finding structure and regularities in data through learning algorithms. So, just
as it can teach itself how to play games like chess or tic-tac-toe, it can teach itself
what product should be recommended, just like in Amazon or Netflix. The algorithm
adapts when new data is provided. It analyzes data using neural networks that have
many hidden layers. It also achieves tasks with commendable accuracy. Figure 1
shows a few domains where artificial intelligence is widely used. Apart from these,
there are various other sectors where AI is used with outstanding brilliancy.

1.1 Autonomous Vehicles

Information collected by the sensors in self-driving cars is useless unless it is

processed. This is where AI comes up and can be matched up to the human brain.
Companies like Waymo carried out numerous test drives before applying their
Impact of Artificial Intelligence in Health care: A Study 313

first AI-based public ride-hailing service [4]. This technique gathers data from the
cameras, radar, and GPS of the vehicle to produce instructions that can operate
it. Another great example of an autonomous vehicle is the Tesla self-driving car.
AI executes computer vision and image detection with DL to build cars that can
recognize objects and drive without human instructions.

1.2 Cybersecurity

AI is being used in this sector as it faces significant challenges like large-scale

hacking that harm organizations and create huge business damage. AI combines
understanding and uses its techniques to create relationships between threats, such
as destructive files, suspicious addresses and so on [5]. This investigation eliminates
time-consuming research activities and provides systematic risk analysis, allowing
security analysts to respond to threats almost 50 times faster and initiate a response
to mitigate the threat. It helps to focus on the attacks that can be a real hazard and
keeps data safe and secure, thus enabling organizations to not become victims of
cyber-attacks.

1.3 Agriculture

AI provides innovative approaches to improve crop yield and also provides efficient
ways to protect crops from weeds. “PEAT,” a Berlin-based agricultural start-up, has
developed “Plantix” that distinguishes prospective defects and nutrient deficiencies
in the soil by a series of images with a calculated accuracy of up to 93%. Precision
cultivation uses AI technology to identify diseases, pests, and inadequate nutrition in
crops. Its sensors can detect and identify weeds and then determine which herbicides
will be used [6]. This process helps to combat the overuse of herbicides and toxins
that damage plants and ultimately end up in our food supply.

1.4 Social Media and Gaming

Nowadays, a tremendous amount of data posts, chats, etc., lead to a huge amount of
data. Here, AI and machine learning (ML) are involved. For example, on Facebook,
AI is used for facial recognition that receives facial features and is also used for
tagging people. ML concepts are used to design our feeds based on our specific
interests and to filter offensive content. AI algorithms can also play systematic and
interactive games where the machine needs to think and analyze a large number
of moves and possible scenarios. For a better game experience, AI is increasingly
transferring control to the player.
314 D. Bairagya et al.

1.5 Military

Nations are developing AI technologies for better analysis, asset performance, cyber
performance, command, and control, as well as a variety of private and indepen-
dent weapons and vehicles. This is also used to enable sensor coordination, threat
detection and detection, detection of signals, enemy positions, and drones capable of
independent operation. The most prominent example of AI on the battlefield is the
Project Maven, which used ML algorithms to filter through hours of motion videos,
tracking suspected terrorists, and insurgents. However, it was publicly said that this
project only looks for tanks and other targets of interest in a war.

1.6 Finance and Business

AI systems can respond when a business is slowing down and reduce financial crime
by monitoring unusual or negative changes. AI-based e-commerce platforms have
made it easy to balance demand and turnover. Customer relationship management
(CRM) platforms are used to get more information on how to better serve customers.
Also, interview chats have already been installed for immediate action. AI algorithms
are great for trading because they can analyze large amounts of data in a short period
of time and can be trained to store patterns and predict how this can happen again.
AI is a vast domain with a huge number of applications, with even more projects
under development. Few other areas of it include its application in Law, Hospitality,
Advertising, Education, Art, Travel, etc. The list is never-ending!

2 AI in Health care

AI in health care is used to automate several tasks like the evaluation of diseases,
performing surgeries, administrative work, and to identify high-risk patients [7]. It is
also applied to dosage issues which is a high-cost problem where it could save more
than $15 billion. It receives patient data to form a hypothesis, which it then displays
with a confidence score. Artificial intelligence can also assist doctors. Microsoft has a
“Hanover” project that will memorize all the papers, articles, and information needed
for cancer and predict which drug combinations will work best for all patients. There
is also a project which is working on fighting myeloid leukaemia. An algorithm also
conducted a diagnosis with more than 94% accuracy and produced a result within
40 s which was a decision as to whether or not whether not, the patient should be
undergoing treatment.
It is used to automatically perform initial CT scans or ECG scans or to identify
high-risk patients. It is already being used for the most expensive volume problem
where it can save about $16 billion. Cambio health care has established a stroke
Impact of Artificial Intelligence in Health care: A Study 315

support system that provides a warning when a patient is at risk of having a stroke.
Coala life has a device that can monitor heart disease. Artificial intelligence can also
assist doctors.
Artificial intelligence has been playing a significant role in society and has only
been expanding its domain for the past few decades [8]. What we do not realize is that
it presents itself in almost everything we use which directly or indirectly impacts our
daily life. Using e-mail, social media, e-commerce platforms, and app-cab services,
all involve AI algorithms to improve our experience and to make life easier. A leading
domain for AI in the healthcare sector and it is already serving major roles here. AI
in health care includes treatment methods, their outcomes, automating routine tasks
in medical practice, survival rates, increasing recovery rate, managing patients and
medical resources, and interconnected health conditions. Here, the data mining and
pattern recognition ability of AI-powered systems come in handy. It uses statistical
techniques to give machines the ability to learn and get trained with new data. It
then identifies patterns to finally make decisions without human instructions. It can
detect and examine trends in data and even make predictions that could identify
potential health hazards. It can help doctors personalize care for patients. Diagnostic
AI applications gather and process data to compare it with predefined categories to
help with treatment. This helps the work of doctors as AI carries out tasks in lesser
time and reduces the cost. This even allows patients to track their glucose monitors,
ultrasounds, electrocardiograms, thermometers, and so on.
Machine learning techniques can extract specific patient traits via their algorithms,
which includes the data collected when a patient visits a doctor [9]. These traits,
for example, symptoms, physical exam reports, medications, disease-related data,
images of diagnoses, and laboratory tests all contribute to structured data. ML uses the
physical features of the patient, along with its huge database of gathered information
to provide a targeted and accurate diagnosis. In deep learning, the same inputs are fed
into a programmed neural network which has hidden layers that merge the provided
information and close into a curated output. This is beneficial when multiple possible
outcomes can then be narrowed down to one or two results, thus, allowing the doctor
to come to a concrete decision [10].
In natural language processing, the information is documented from notes and
speech to text processing when a doctor sees a patient. This includes reports from
physical examinations, previous exposure to diseases, laboratory tests, and so on.
NLP does this by using databases that have disease-relevant keywords which aid
the final decision procedure for a correct diagnosis. This improves diagnoses as it
provides an efficient output which saves time and can also speed up the treatment
process so the patient can recover faster. AI-based platforms allow physicians to gain
a better grasp on significant patterns from data which saves time, effort, and expen-
diture. This incorporation of AI tools and techniques results in efficiently improving
patient care.
AI can also pinpoint any risk a patient might have, thereby helping in the early
detection of a health hazard. One study involving high-risk stroke patients was
conducted which used AI programs including their present and past symptoms, and
genetic information. Here, if any unusual physical movement in the patient was
316 D. Bairagya et al.

recorded, an alert would be triggered. This enabled doctors to send patients to an

MRI or CT scan earlier due to which they were able to arrange for treatment sooner
and predict whether the patient had a possibility of future stroke. This early detection
had an accuracy of 85% in a diagnostic evaluation.
Quite often, a patient could have several symptoms that point to different condi-
tions which delay the diagnosis. AI improves accuracy in early detection, diagnosis,
and treatment. It can also be used to help newer physicians, who do not have that
much experience, to be able to make the right decisions. Ophthalmology and radi-
ology are popular fields because of the AI process of image analysis. It uses images
of the human eye to provide a diagnosis that would otherwise involve an ophthalmol-
ogist. Using these methods, professionals, therapists, and even patients can come to
conclusions. Physicians can share HD 3D images and consult with other physicians
locally and internationally to improve the efficiency of the procedure.
AI can perform tasks that are tedious in their operation. Too much time is spent
working on records, bills, study screens, laboratory testing rooms, and more. AI
systems can handle these tasks efficiently and filter the results which could save a
significant amount of time, manage patients, and maximize the quality for diagnosis.
AI also manages medical resources. For example, it can predict which patients could
benefit the most from rare medical resources, which departments are likely to need
additional staff, which practitioner is most required in which department, and so on.
AI-powered robots are also quite well known by now as there are over 200,000
robots that are installed every year, all over the world. These robots can perform
several tasks from lifting or assembling objects to the delivery of supplies in health-
care units. These days, robots are more cooperative with humans and can be trained
easily to carry out specific tasks. With other capabilities being embedded in their oper-
ating systems, they are also becoming more intelligent. There are surgical robots as
well which help surgeons by enhancing their ability to explore, create precise inci-
sions, perform tests, stitch wounds, and so on. Gynaecologic surgery, prostate surgery,
and head and neck surgery are a few common surgeries that employ robots.
Wearable technologies like FitBits and smartwatches use AI to track data and
alert users about potential health issues and hazards. Being able to analyze one’s
health eases professionals and prevents unnecessary hospital visits. It is important to
guide new trainees on the capabilities of AI and also train them to function efficiently
alongside new technology. The ultimate aim should be to achieve a proper balance
between the productive use of automated procedures and the judgment of trained
doctors. This is a necessity as the thought of AI completely replacing humans in the
realm of medicine is a concern as this could end up hindering the privileges which
are obtained from it.
Impact of Artificial Intelligence in Health care: A Study 317

3 Existing Applications Integrating AI in the Healthcare

Sector

Artificial intelligence in health care has a wide scope and is regularly developing.
Many existing AI-powered services are already implemented in the healthcare
domain. Figure 2 shows a few application of AI in health care. AI applications
are not only limited to healthcare but a far more vast and constantly increasing the
usability in different sectors.

3.1 Virtual Nurses and Digital Consultation

AI-powered virtual assistants are good for nursing. They can keep doctors and
patients in touch with each other all the time thus lowering sudden hospital visits
and load on medical professionals. The start-up Sense.ly had developed “Molly,” a
digital nurse to observe the patient’s health and follow up for treatments. ML is used
to support patients with chronic illnesses.

Fig. 2 AI in health care

318 D. Bairagya et al.

3.2 Robots

Robots use actual surgical encounter information to upgrade existing surgical

approaches and techniques and can also assist surgeons since their accuracy, flexi-
bility, and control go well beyond human potential. This reduces treatment time and
error. Analyzing tests, X-Rays, lab reports, CT and MRI scans, billing, data entry,
and so on can all be done faster and precisely by robots.

3.3 Cybersecurity

It is the responsibility of a hospital to keep the data of their patients safe and secure.
Instinctive and prudent, AI cybersecurity applications provide that extra layer
of safety. Advanced protection protects the LAN and responds in nano-seconds to
anomalous behavior patterns. Biomedical security tools use artificial intelligence to
handle and secure the increasing number of devices as they connect and disconnect
from hospital networks.

3.4 Administration and Workflow

This technique helps to maximize the time given by nurses, doctors to their jobs and
make their everyday tasks easier. It merges AI algorithms with existing technological
infrastructure. AI can keep a record of all treatments and patients to recommend
predictive diagnosis and also assemble, store, re-format, and detect data to provide
swift and accurate results.

3.5 Dosage and Treatment Design

The dosage of a patient is important as even a little drop of medicine can make a
huge difference. The applications are designed to lessen the margin of errors that may
occur when giving medicines to patients. There are elaborate artificial intelligence
systems that are designed to load and analyze data. From records and reports to
expertise or external research, these applications will help converge to the correct
and individualized treatment for every patient.
Impact of Artificial Intelligence in Health care: A Study 319

3.6 Fraud Detection

Both patients and healthcare providers can face damage due to frauds. Fraud detection
with AI can analyze data within seconds and detect complex patterns in the best way
possible. AI also automates time-consuming tasks which enable the fraud analysts
to focus on high-risk, critical cases. The work quality and efficiency of the analysts
also increase due to these automated AI algorithms. This also helps to steer through
elaborate payment and service fraud endeavors.

3.7 Health Monitoring

Wearable well-being trackers like FitBit, Garmin, and others monitor the heart rate
and activity levels of the body to make sure that the person is healthy and fit. It
sends alerts and notifications about improvements and can share this information
with doctors as well.

3.8 Drug Creation and Clinical Trial Participation

AI-powered applications can make the procedure of drug creation faster and inex-
pensive. By examining existing medicines, they can remodel them to counter disease.
AI-based trials collect data and assemble them to get the right theory and facilitate
its outcome.

3.9 Treatment Design and Precision Medicine

Correct diagnosis is the pillar of the healthcare domain. The sooner it is, the better the
treatment will be. AI uses data architecture to refine the precision of the initial diag-
nosis. It analyzes data and reports of the patient, conducts external research, involves
clinical expertise, and helps to choose the accurate customized treatment. Also, with
the help of AI, full-body scans can spot cancerous cells and vascular diseases early
and predict health issues that patients might face based on their genetics. Table 1
highlights the working principle, merits, and applications of some popular products
implementing AI in health care.
320 D. Bairagya et al.

Table 1 Comparative analysis of projects implementing AI

Product Working principle Applications
PathAI • Minimizing error in cancer diagnosis • Cancer diagnosis
• Designing methods for precision • Precision medicine
treatment
Enlitic • DL to facilitate radiology diagnostics • Radiology
Freenome • Merges deep expertise in molecular • Radiology
biology with advanced computational
techniques
• Routine blood draw
XtalPi • Predicts the chemical composition • Drug design and development
and pharmaceutical features of micro
molecules by integrating AI, the
cloud, and quantum physics
Olive • AI-powered algorithms to automate • Administration and workflow for fast
and optimize workflow and stable operations
Qventus • AI-based algorithm to solve • Simplifying healthcare operations
operational issues
IBM Watson • Infusing AI into applications to make • Individualized health plans
more precise predictions, automate • Early detection of diseases
decisions and procedures, and
optimize staff time
ICarbonX • Uses AI and big data to better analyze • All aspects of health care
aspects of human life to digitize and
understand life

4 Popular AI Products in Health care

There are hundreds of other healthcare companies and hospitals that benefit from the
services AI provides. Here are a few of them.

4.1 PathAI

PathAI has developed a technology to aid pathologists to enhance accurate diagnoses.

Its current goal is to reduce the risk of cancer diagnosis and innovative treatment
options. PathAI has partnered with drug developers such as Bristol-Myers Squibb,
clinical networks such as LabCorp, and organizations such as the Bill & Melinda
Gates Foundation to expand its technology.
Impact of Artificial Intelligence in Health care: A Study 321

4.2 Enlitic

Enlitic uses DL to forge ahead in radiology. It examines medical data (radiology

imaging, blood tests, medical history of the patient, and more) to allow doctors to
better assess the patient’s real-time needs and allow them to get more precision,
efficiency, and quality.

4.3 Freenome

It focuses on diagnosing cancer in its early stages and over time, developing new
therapies using AI for tests, diagnostic tests, and blood pressure. By combining
cellular biology with advanced computer techniques to detect and interpret disease-
related patterns among billions of circulating biomarkers, it improves simple and
accurate blood tests for the early detection of cancer.

4.4 Bioxcel Therapeutics

It is a clinical biopharmaceutical company that uses AI to identify and develop

new drugs in the field of immuno-oncology and neuroscience and approaches to
drug development. Additionally, the company’s drug rehabilitation program uses AI
to find new uses for existing drugs. It was called “Most Innovative Healthcare AI
Development of 2019.”

4.5 XtalPi

Combining AI, cloud, and quantum physics, XtalPi’s ID4 system predicts chemical
and pharmaceutical properties of micro-chemical in the development and devel-
opment of drugs. Besides, it claims that its crystal structure forecasting technology
envisions intricate cell formation in a matter of days. XtalPi investors include Google,
Tencent, and Sequoia Capital.

4.6 BenevolentAI

BenevolentAI’s main goal is to find the right treatment for patients promptly by using
artificial intelligence to produce better choices and to provide previously unpublished
322 D. Bairagya et al.

information about DL. BenevolentAI has partnered with major pharmaceutical orga-
nizations to issue licenses for drugs and organizations that help the community to
develop drugs that are more susceptible to rare diseases.

4.7 Olive

The Olive AI platform is designed to perform very repetitive tasks thus freeing
executives to operate at a higher level. It makes everything automated from qualifi-
cation testing to data migration so doctors can focus on providing the best patient
service. Olive AI-as-a-Service integrates easily within existing hospital software thus
eliminating the need for costly integration.

4.8 Qventus

Qventus supports former teams and develops genes using AI and behavioral science
that solves operational issues including those linked to emergency rooms and patient
safety. It prioritizes the patient’s illness or injury, follows waiting times and can even
map the immediate ambulance routes. CB insights named Qventus as one of the 100
most starting AI startups for 2019.

4.9 IBM Watson

Watson’s IBM experts help use their data to increase hospital productivity, better
engagement with patients, and improve treatment by combining human experts and
artificial intelligence. Watson currently uses his skills in everything from designing
personal health programs to interpreting genetic testing results and capturing early
signs of disease.

4.10 ICarbonX

ICarbonX uses AI and big data to carefully study aspects of human life in a way
they describe as “digital life.” By examining people’s health and actions in the
“carbon cloud,” they hope it will be strong enough to handle all aspects of life.
ICarbonX expects that when it will eventually collect enough data, it will better
diagnose symptoms, change treatments, and find healthy people.
Impact of Artificial Intelligence in Health care: A Study 323

4.11 Vicarious Surgical

Vicarious surgical integrates VR with AI-enabled robots so that surgeons can perform
less invasive tasks. Surgeons almost shrink and examine the inside of the patient’s
body in detail. With these new technologies, they seek to improve the lives of patients,
improve the efficiency of surgeons, and increase global access to quality care.
In addition to the existing establishment, there are some advances in various stages
of development, which help physicians to become better physicians.
Apart from the inventions which already exist, there are certain advances in various
phases of development, which help physicians be better doctors.

5 Background Study

In [11], Mohammad Shafenoor Amin et al. targets critical features and data mining
skills for the accurate prediction of heart disorders. Here Naïve Bayes and logistic
regression are used. The number of attributes for this proposed method is nine, and
the accuracy for the vote is 86%. Sarath Babu et al. in [12] focused on the prediction
of heart disorders by the effective use of data mining and sequencing of character-
istics. This approach uses genetic algorithm, k-means algorithm, maximal frequent
itemset algorithm (MAFIA), and decision trees. Meenal saini et al. in [13] used nine
classifiers and some modern procedures to predict heart diseases. This hybrid classi-
fier gave an accuracy of 82%. Purushottam et al. [14] has helped untrained physicians
to make an informed choice in matters involving the heart. KEEL tool was used for
its implementation, and the generated classification decision rule gave an accuracy of
86% [15]. Bandarage shehsni et al. presented a comparative analysis of the strategies
for distinguishing the prediction of cardiovascular disease. This method used Naïve
Bayes, neural networks, and decision trees. Naïve Bayes had an accuracy of 86%,
neural networks of 89%, and decision trees of 85% [16]. The analysis by Dandıl,
(2018) set out a unique automated channel of computed tomography (CT) diagnostic
test for the diagnosis of lung cancer in early stages by the division of benign and
malignant nodes [17]. The research work by Naresh and Shettar (2014) proposed
a method for the identification of the presence of lung cancer during the prelimi-
nary stages itself by utilizing CT screening images taken from patients [18]. The
research by Wu et al. (2017) proposed the use of computational histomorphometry
image classifier that works on nuclear sequencing, texture, composition, and tumor
structure to anticipate the possibility of recurrence of lung cancer [19]. The study by
Win et al. (2014) unfolds one of the most advanced ML strategies, termed averaged
one-dependence estimators with subsumption resolution that forecasts the reappear-
ance of the tumor by using DNA microarray gene expression data [20]. The study by
Teramoto et al. (2017) introduced a computer-aided program to identify lung cancer,
found in microimages using deep convolutional neural network.
324 D. Bairagya et al.

6 Deep Learning in Radiographic Imaging

The fundamental advancements and enhancements of AI in health care are mostly

in medical imaging. Medical imaging involves the identification of the image of
interest and the important areas of the image [21]. The image is then segregated and
interpreted as required. Recent advancements in ML and deep learning have further
heightened computational strategies that can refine image quality and boost accuracy
to optimize clinical decisions [22–25]. These methods begin with the gathering of
data in huge amounts. This data is curated and integrated to facilitate automated
analysis which provides a better view of the collected data, which then proceeds
for screening. The application of information collected from datasets, continuous
monitoring along with screening reports result in automated detection of potential
tumors. The application then proceeds with specific treatments optimized for each
patient by the categorization of different types of tumors. Table 2 highlights the
number of articles and datasets which were taken to demonstrate the feasibility of
radiographic imaging.
Figure 3 illustrates the result of the model discussed above. It is seen that it
classifies brain cancer with an accuracy of 93.8%, prostrate cancer with an accuracy

Table 2 Datasets and articles

Type of cancer Number of datasets Number of articles
taken for reference
Brain cancer 03 10
Prostrate cancer 02 12
Skin cancer 03 14
Lung cancer 04 15
Breast cancer 04 12

Brain Cancer 93.8

Prostrate Cancer 94.2

Skin Cancer 90.4

Lung Cancer 89.6

Breast Cancer 92.5

87 88 89 90 91 92 93 94 95

Classification Accuracy (%)

Fig. 3 Classification accuracy for various types of cancer

Impact of Artificial Intelligence in Health care: A Study 325

3
2.67
2.43
2.5

2 1.78
1.55
1.5
1.12
1

0.5

0
Breast Cancer Lung Cancer Skin Cancer Prostrate Cancer Brain Cancer

Execution Time Delay (sec)

Fig. 4 Execution time delay in seconds for various types of cancer

of 94.2%, and so on. This process provides maximum accuracy for prostrate cancer
and minimum for lung cancer.
Figure 4 illustrates the execution time day for the five types of cancer that we have
referred to. It is found that it takes maximum time for the analysis of lung cancer
(2.43 s) and minimum for brain cancer (1.12 s).
Recently, an automated detection system for radiography utilizing deep learning
algorithms was developed to distinguish between common diseases for diagnosis.
This particular algorithm integrated data from various streams into an integrated
diagnosis to come up with a specified treatment plan by mining data to find links. This
system also learnt and improved from previous patterns and images. The outcome
of this method proved that this AI-powered process has superior image recognition
abilities than when done manually. This brings forward the vast potential of AI in
this field for finer quality, boosted accuracy, and greater efficiency. Medical imaging
is a helpful and essential model for the detection of cancer, progression monitoring,
and prediction prognosis of the disease. Early detection of disease with this is vital
for lowering mortality rates.
Figure 5 shows how AI algorithms in health care begin with the gathering of
large amounts of data. The curation and integration of this data are done to facilitate
automated analysis which provides a better view of the collected data, which then
proceeds for screening. Medical imaging involves the identification of the image
of interest and the important areas of the image. The image is then segregated and
interpreted as required. The application of information from datasets, continuous
monitoring along with the results of patient screening results in automated detection
of potential tumors. Through the categorization of different types of tumors, the
application then proceeds with the specific treatments optimized for each patient.
326 D. Bairagya et al.

Fig. 5 Artificial intelligence in cancer medical imaging

7 Conclusion

The acceptance of AI in health care relates to the analysis of the huge amount of
information that is generated each day and the limitation of physicians to address
these needs. With advanced computing ability, medical AI programs are overcoming
these limitations to a large extent. Hardly a week goes by without research claiming
that it has come up with a new approach to using AI for diagnosis or treatment
with greater accuracy than humans. However, the most prominent challenge to AI in
this domain is ensuring its application in daily practice. For extensive adoption, AI
programs must be approved by regulators, standardized to a degree such that similar
products work similarly, taught to physicians, paid for by organizations, and updated
regularly. These challenges will be overcome, but that will take a substantial amount
of time. As a result, we can expect to see wider and more extensive use of AI in the
health care sector within 10–15 years.

References

1. Patel, V., Shortliffe, E., Stefanelli, M., Szolovits, P., Berthold, M., Bellazzi, R., et al. (2009).
The coming of age of artificial intelligence in medicine. Artificial Intelligence in Medicine,
46(1), 5–17.
Impact of Artificial Intelligence in Health care: A Study 327

2. Buch, V. H., Ahmed, I., & Maruthappu, M. (2018). Artificial intelligence in medicine: Current
trends and future possibilities. British Journal of General Practice, 68, 143–144.
3. Mishra, S., Tripathy, H. K., & Mishra, B. K. (2018). Implementation of biologically motivated
optimisation approach for tumour categorisation. International Journal of Computer Aided
Engineering and Technology, 10(3), 244–256.
4. Mishra, S., Tripathy, H. K., Mallick, P. K., Bhoi, A. K., & Barsocchi, P. (2020). EAGA-MLP—
An enhanced and adaptive hybrid classification model for diabetes diagnosis. Sensors, 20(14),
4036.
5. Rath, M., & Mishra, S. (2020). Security approaches in machine learning for satellite commu-
nication. In Machine learning and data mining in aerospace technology (pp. 189–204).
Springer.
6. Sahoo, S., Mishra, S., Panda, B., & Jena, N. (2016). Building a new model for feature optimiza-
tion in agricultural sectors. In 2016 3rd International Conference on Computing for Sustainable
Global Development (INDIACom) (pp. 2337–2341), New Delhi.
7. Jena, L., Patra, B., Nayak, S., Mishra, S., & Tripathy, S. (2019). Risk prediction of kidney
disease using machine learning strategies. In Intelligent and cloud computing (pp. 485–494).
Springer.
8. Schlessinger, D. I., Chhor, G., Gevaert, O., Swetter, S. M., Ko, J., & Novoa, R. A. (2019). Arti-
ficial intelligence and dermatology: Opportunities, challenges, and future directions. Seminars
in Cutaneous Medicine and Surgery, 38, E31–E37.
9. Mishra, S., Mallick, P. K., Jena, L., & Chae, G. S. (2020). Optimization of skewed data using
sampling-based preprocessing approach. Frontiers in Public Health, 8, 274. https://doi.org/10.
3389/fpubh.2020.00274
10. Mishra, S., Mallick, P. K., Tripathy, H. K., Bhoi, A. K., & González-Briones, A. (2020). Perfor-
mance evaluation of a proposed machine learning model for chronic disease datasets using an
integrated attribute evaluator and an improved decision tree classifier. Applied Sciences, 10(22),
8137.
11. Amin, M. S., Chiam, Y. K., & Varathan, K. D. (2018). Identification of significant Features
and data mining techniques in predicting heart disease. Telematics and Informatics.
12. Babu, S., Vivek, E. M., Famina, K. P., Fida, K., Aswathi, P., Shanid, M., & Hena, M. (2017).
Heart disease diagnosis using data mining technique. In 2017 IEEE International Conference
on Electronics, Communication and Aerospace Technology (ICECA).
13. Saini, M., Baliyan, N., & Bassi, V. (2017). Prediction of heart disease severity with hybrid
data mining. In International Conference on Energy, Communication, Data Analytics and Soft
Computing (ICECDS)—IEEE 2017.
14. Purushottam, Saxena, K. Dr., & Sharma, R. (2016). Efficient heart disease prediction system.
Procedia Computer Science.
15. Rathnayake, B. S. S., & Ganegoda, G. U. (2017). Heart disease prediction with data mining
and neural network techniques. In 2017-IEEE 4th International Conference for Convergence
in Technology.
16. Dandıl, E. (2018). A computer-aided pipeline for automatic lung cancer classification on
computed tomography scans. Journal of Healthcare Engineering.
17. Naresh, P., & Shettar, D. R. (2014). Early detection of lung cancer using neural network
techniques. International Journal of Engineering Research and Applications, 4.
18. Wu, Y. C., Wei, N. C., Hung, J. J., Yeh, Y. C., Su, L. J., Hsu, W. H., & Chou, T. Y. (2017).
Generating a robust prediction model for stage I lung adenocarcinoma recurrence after surgical
resection. Oncotarget, 8(45), 79712.
19. Win, S. L., Htike, Z. Z., Yusof, F., & Noorbatcha, I. A. (2014). Cancer recurrence predic-
tion using machine learning. International Journal of Computational Science and Information
Technology (IJCSIT), 6(1).
20. Teramoto, A., Tsukamoto, T., Kiriyama, Y., & Fujita, H. (2017). Automated classification of
lung cancer types from cytological images using deep convolutional neural networks. BioMed
Research International.
328 D. Bairagya et al.

21. Mishra, S., Dash, A., & Jena, L. Use of deep learning for disease detection and diagnosis.
In Bio-inspired neurocomputing (pp. 181–201). Springer.
22. Mishra, S., Tripathy, H. K., & Acharya, B. A precise analysis of deep learning for medical
image processing. In Bio-inspired neurocomputing (pp. 25–41). Springer.
23. Sahoo, S., Das, M., Mishra, S., & Suman, S. (2021). A hybrid DTNB model for heart disorders
prediction. In Advances in electronics, communication and computing (pp. 155–163). Springer.
24. Mallick, P. K., Mishra, S., & Chae, G. S. (2020). Digital media news categorization using
Bernoulli document model for web content convergence. Personal and Ubiquitous Computing.
https://doi.org/10.1007/s00779-020-01461-9
25. Mishra, S., Mallick, P. K., Tripathy, H. K., Jena, L., & Chae, G.-S. (2021). Stacked KNN with
hard voting predictive approach to assist hiring process in IT organizations. The International
Journal of Electrical Engineering and Education. https://doi.org/10.1177/0020720921989015