Lecture Notes in Bioengineering

Somsubhra Gupta
Sandip Bag
Karabi Ganguly
Indranath Sarkar
Papun Biswas Editors

Advancements
of Medical
Electronics
Proceedings of the First International
Conference, ICAME 2015
Lecture Notes in Bioengineering
More information about this series at http://www.springer.com/series/11564
Somsubhra Gupta Sandip Bag
•

Karabi Ganguly Indranath Sarkar

•

Papun Biswas
Editors

Advancements of Medical
Electronics
Proceedings of the First International
Conference, ICAME 2015

123
Editors
Somsubhra Gupta Indranath Sarkar
Information Technology Electronics and Communication
JIS College of Engineering Engineering
Kalyani, West Bengal JIS College of Engineering
India Kalyani, West Bengal
India
Sandip Bag
Biomedical Engineering Papun Biswas
JIS College of Engineering Electrical Engineering
Kalyani, West Bengal JIS College of Engineering
India Kalyani, West Bengal
India
Karabi Ganguly
Biomedical Engineering
JIS College of Engineering
Kalyani, West Bengal
India

ISSN 2195-271X ISSN 2195-2728 (electronic)

Lecture Notes in Bioengineering
ISBN 978-81-322-2255-2 ISBN 978-81-322-2256-9 (eBook)
DOI 10.1007/978-81-322-2256-9

Library of Congress Control Number: 2014958893

Springer New Delhi Heidelberg New York Dordrecht London

© Springer India 2015
This work is subject to copyright. All rights are reserved 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, express or implied, with respect to the material contained
herein or for any errors or omissions that may have been made.

Printed on acid-free paper

Springer (India) Pvt. Ltd. is part of Springer Science+Business Media (www.springer.com)

Invited Talks

E. Russell Ritenour, Ph.D.

Professor and Chief Medical Physicist
Department of Radiology and
Radiological Sciences
Medical University of South Carolina
96 Jonathan Lucas St., MSC 323
Charleston, SC 29425-3230
ph: 843-792-4884

Invited talk title Magnetic Resonance Imaging Safety Issues at High Field
Strength—E. Russell Ritenour, Ph.D.
Invited talk abstract The increasing trend toward higher magnetic fields in Magnetic
Resonance Imaging (MRI) brings with it some new challenges in
safety for patients and staff in the medical environment. In
particular, the stronger static magnetic field gradients have the
capability of producing measurable effects in diamagnetic
substances as well as paramagnetic and ferromagnetic
substances. The higher peak values of static magnetic field also
have the potential to produce unwanted inline forces and torques
upon paramagnetic and ferromagnetic substances.
In the work presented here, magnetic properties of materials
will be reviewed briefly. Artifacts and safety issues will be
discussed with emphasis on the increased severity of effects at
higher magnetic field levels. The current status of the United
States Food and Drug Administration regulation of equipment
will be presented with emphasis on equipment design and
labeling. The equipment discussed will include devices
implanted in the patient. Practical aspects of dealing with patients
and doing research with the high magnetic field large bore
magnets used in MRI will be emphasized
(continued)

v
vi Invited Talks

(continued)
Bio E. Russell (Russ) Ritenour received his Ph.D. in Physics from the
University of Virginia in 1980. He was selected for a National
Institutes of Health Postdoctoral Fellowship at the University of
Colorado and stayed on the faculty there as director of the
Graduate Program in Medical Physics until 1989 when he moved
to Minnesota to become Professor and Chief of Physics and
Director of the Graduate Program in Biophysical Sciences and
Medical Physics at the University of Minnesota. In 2014 he
became Professor and Chief Medical Physicist of the Department
of Radiology and Radiological Science of the Medical University
of South Carolina. He has served the American Board of
Radiology in various capacities since 1986 including
membership and then chair of the physics committee for the
radiology resident’s exam. He has served as a consultant to the
U.S. Army for resident physics training and is a founding
member of the Society of Directors of Academic Medical
Physics Programs. He is a fellow of the ACR and a fellow and
past president of the AAPM. He also served as imaging editor
of the journal Radiographics and as Board Member and Treasurer
of the RSNA Research and Education Foundation and as Board
Member of the International Organization of Medical Physics.
He is currently the Medical Physics and Informatics Editor of the
American Journal of Roentgenology. His research interests
include radiation safety, efficacy of diagnostic imaging, and the
use of high speed networks for medical education and clinical
communication
Xiaoping Hu, Ph.D.
The Wallace H. Coulter Department
of Biomedical Engineering
Georgia Institute of Technology and
Emory University
Atlanta, Georgia, USA
[email protected]

Invited talk title Recent Developments in Magnetic Resonance Neuroimaging and

Molecular Imaging—Xiaoping Hu, Ph.D.
(continued)
Invited Talks vii

(continued)
Invited talk abstract While MRI has been around for more than 40 years, significant
advances are still being made. In particular, there is remarkable
progress in the methodologies and applications of MRI in the
study of the brain, i.e., neuroimaging, and there are also
numerous innovations in molecular imaging with MRI. There are
two main directions in MRI of the brain: functional brain
imaging and structural connectivity. The former was ushered into
the field in the early 1990s and has generated an unprecedented
interest in a wide range of disciplines; it is now used not only for
mapping brain function but also for understanding brain activity
and ascertaining brain connectivity. The latter was introduced at
about the same time but has been exploding in the last decade
with the advances in gradient technology; it is now widely used
for assessing the structural connectivity of the brain. For
molecular imaging, MRI has been used for cell tracking, targeted
imaging of biomarkers, and reporting of gene express. In this
talk, I will provide an overview of these two broad directions and
highlight some of the recent in my lab. More highlights in
neuroimaging will include methodological developments and
applications of functional and structural connectivity of the brain
and MR detection of action current. As for molecular imaging, I
will highlight methods for better detection of magnetic
nanoparticles and development of MR reporter genes
Bio Dr. Hu obtained his Ph.D. in Medical Physics from the
University of Chicago in 1988 and his postdoctoral training there
from 1988 to 1990. From 1990 to 2002, he was on the faculty
of the University of Minnesota, where he became full professor
in 1998. Since 2002, he has been Professor and Georgia Research
Alliance Eminent Scholar in Imaging in the Wallace H. Coulter
joint department of biomedical engineering at Georgia Tech and
Emory University and the director of Biomedical Imaging
Technology Center in the Emory University School of Medicine.
Dr. Hu has worked on the development and biomedical
application of magnetic resonance imaging for three decades. Dr.
Hu has authored or co-authored 245 peer-reviewed journal
articles. As one of the early players, Dr. Hu has conducted
extensive and pioneering work in functional MRI (fMRI),
including methods for removing physiological noise,
development of ultrahigh field fMRI, systematic investigation
of the initial dip in the fMRI signal, Granger causality analysis of
fMRI data, and, more recently, characterization of the dynamic
nature of resting state fMRI data. In addition to neuroimaging,
his research interest also includes MR molecular imaging. Dr. Hu
was deputy editor of Magnetic Resonance in Medicine from 2005
to 2013, Associate Editor of IEEE Transactions on Medical
Imaging since 1994, editor of Brain Connectivity since its
inception, and editorial board member of IEEE Transactions on
Biomedical Engineering since 2012. He was named a fellow
of the International Society for Magnetic Resonance in Medicine
in 2004. He is also a fellow of IEEE and a fellow of the
American Institute of Medical and Biological Engineering
Keynote Talk

Hiro Yoshida, Ph.D.

Director, 3D Imaging Research, Department
of Radiology, Massachusetts General
Hospital
Associate Professor of Radiology, HMS
25 New Chardon St, Suite 400C, Boston,
MA 02114

Keynote talk title Advancement of diagnostic imaging decision

support systems
Keynote talk abstract In clinical practice, there is an increasing
demand for fast turnaround time in obtaining
high-quality multi-dimensional medical
images with quantitative analysis and
diagnostic results obtained from big data.
Access to such images and analysis results
from anywhere at any time facilitates
collaboration between clinicians and
specialists as well as patients and healthcare
providers, thus improves the quality and
timeliness of care for the patient. However,
such advanced imaging is computationally
intensive and requires high-end resources such
as computational servers and clusters. In
today’s economic climate, investment in such
resources is often cost-prohibitive, thus
limiting broader adoption of advanced
imaging techniques and quantitative analysis
(continued)

ix
x Keynote Talk

(continued)
tools. Cloud supercomputing, or high-
performance cloud computing, is an
integration of high-performance computing
with today’s ubiquitous cloud computing.
Recent advances of cloud technology
provides, for the first time in history, an
affordable infrastructure that delivers the
supercomputer power needed for real-time
processing of these state-of-the-art diagnostic
images on mobile and wearable devices with
an easy, gesture-based user interface.
This keynote shows advancements of
diagnostic imaging decision support system
that address the above clinical and technical
needs, by using computer-assisted virtual
colonoscopy for cancer screening as a
representative example. Virtual colonoscopy,
also known as computed tomographic (CT)
colonography, provides a patient-friendly
method for early detection of colorectal
lesions, and has the potential to solve the
problems of capacity and safety with
conventional colorectal screening methods.
Virtual colonoscopy has been endorsed as a
viable option for colorectal cancer screening
and shown to increase patient adherence in the
United States and in Europe. Anytime,
anywhere access to the VC images will
facilitate the high-throughput colon cancer
screening.
A could-supercomputer-assisted virtual
colonoscopy demonstration system will be
presented. In this system, virtual colonoscopy
images are processed by computationally
intensive algorithms such as virtual bowel
cleansing and real-time computer-aided
detection for improved detection performance
of colonic polyps—precursor of colon cancer.
A high-resolution mobile display system is
connected to the could-supercomputer-assisted
virtual colonoscopy to allow for visualization
of the entire colonic lumens and diagnosis of
colonic lesions at anytime, anywhere. The
navigation through the colonic lumen is driven
by a motion-based natural user interface for
easy navigation and localization of colonic
lesions. The current status, challenges, and
promises of the system for realizing efficient
diagnostic imaging decision support system
will be described.
(continued)
Keynote Talk xi

(continued)
Bio Hiro (Hiroyuki) Yoshida received his B.S. and
M.S. degrees in Physics and a Ph.D. degree in
Information Science from the University of
Tokyo, Japan. He previously held Assistant
Professorship in the Department of Radiology
at the University of Chicago. He was a tenured
Associate Professor when he left the university
and joined the Massachusetts General Hospital
(MGH) and Harvard Medical School (HMS),
where he is currently the Director of 3D
Imaging Research in the Department of
Radiology, MGH and Associate Professor of
Radiology at HMS. His research interests
include computer-aided diagnosis, quantitative
medical imaging, and imaging informatics
systems in general, and in particular, in the
area of the diagnosis of colorectal cancers with
CT colonography. In these research areas, he
has been the principal investigator on 11
national research projects funded by the
National Institutes of Health (NIH) and two
projects by the American Cancer Society
(ACS) with total approximate direct cost of
$6.5 million, and received 9 academic awards:
a Magna Cum Laude, two Cum Laude, four
Certificate of Merit, and two Excellences in
Design Awards from the Annual Meetings of
Radiological Society of North America
(RSNA), an Honorable Mention award from
the Society of Photo-Optical Instrumentation
Engineers (SPIE) Medical Imaging, and Gold
medal of CyPos Award from the Japan
Radiology Congress (JRC). He also received
two industrial awards on his work on system
developments: 2012 Innovation Award from
Microsoft Health Users Group and Partners in
Excellence award from Partners HealthCare.
He is author or co-author of more than 170
journal and proceedings papers and 16 book
chapters, author, co-author, or editor of 14
books, and inventor and co-inventor of seven
issued patents. He was guest editor of IEEE
Transaction on Medical Imaging in 2004, and
currently serves on the editorial boards of the
International Journal of Computer Assisted
Radiology, the International Journal of
Computers in Healthcare, Intelligent Decision
Technology: An International Journal, and ad
hoc Associate Editor of Medical Physics.
From the Convener’s Desk

The 1st International Conference on ‘Advancements

of Medical Electronics’ (ICAME 2015) scheduled from
29 to 30 January 2015 has been organized with the
objective to provide a platform for exchange of research
ideas, discussions on the arena of Medical Electronics,
development of new methodologies which reach out to
mankind and is beneficial for the society.
Understanding the need for interdisciplinary subjects
like Biomedical Engineering, Electronics and Commu-
nication Engineering, Nanotechnology, the idea cropped
up into the minds of a young group of Faculty members
at JIS College of Engineering to organize a Conference
which would reach out to the objective mentioned above and address the issues
related to Quality Improvement of Life, Rehabilitation after Surgery or Long-term
Illness that might have forced a person to compromise with a healthy lifestyle.
The amazing capabilities of Science and Technology in today’s world have
come up with numerous innovation and discoveries aimed towards improving the
quality of life even in critical cases like quadriplegia and terminal illnesses like
cancer. But often all such ‘Gifts of Science and Technology’ reach out to the
advanced countries while the Developing Countries like India remain still in a
dearth of knowledge and indigenous developments which could have catered to a
hundred million population like ours.
With the zeal to serve the nation and human life, ICAME 2015 was organized.
The international status of the conference will enable eminent speakers from fields
like Radio Imaging, Nanotechnology, Nanomedicine, Communication, Electronics
from countries outside India to amalgamate with national scientists and researchers.
This would, I firmly believe, generate resources for our country and serve our people
with affordable medical devices and thus increase the lifespan and quality life of our
fellow Indians which are perhaps the best rewards for technologists like us.

xiii
xiv From the Convener’s Desk

So, I invite all of you from the international and national arena to make the most
of the two days and enable us to come up with research works beneficial for
mankind.

Dr. Meghamala Dutta

Head of Department of Biomedical Engineering and
Convener ICAME 2015
Program Committee

International Conference on Advancements of Medical Electronics

29–30th January 2015
Organized by
Department of Biomedical Engineering and Department of Electronics and
Communication Engineering
JIS College of Engineering, Kalyani
Chief Patron
Sardar Jodh Singh, Chairman, JIS Group, India
Prof. S.M. Chatterjee, Former VC, BESU, Shibpur, Howrah
Patrons
Shri Taranjit Singh, Managing Director, JIS Group
Prof. Dr. Ajoy Roy, Vice Chancellor BESU, Shibpur, Howrah
Dr. Sajal Dasgupta, Director, Technical Education
Manpreet Kaur, CEO, JIS Group, India
Jaspreet Kaur, Trusty Member, JIS Group
Organizing Chair
Prof. Dr. Asit Guha, Advisor, JIS Group
Executive Committee Member
Mr. U.S. Mukherjee, Deputy Director, JIS Group
Prof. Sankar Ray, Principal, JISCE
Prof. Urmibrata Bandopadhyay, Former Principal, JISCE
Mrs. Shila Singh Ghosh, General Manager (Corporate Relation), JIS Group, India
Prof. Ardhendu Bhattacharya, Dean, JISCE
Prof. Debatosh Guha, Chair, IEEE Kolkata Section
Prof. S.K. Mitra, HOD-Department of Electrical Engineering, JISCE
Prof. P.K. Bardhan, HOD-Department of Mechanical Engineering, JISCE
Prof. Pranab K. Banerjee, Former Professor, Jadavpur University

xv
xvi Program Committee

Convener
Dr. Meghamala Dutta, HOD-Department of Biomedical Engineering, JISCE
Co-Convener
Mrs. Swastika Chakraborty, HOD-Department of Electronics and Communication
Engineering
Dr. Sandip Bag, Department of Biomedical Engineering
International Advisory Board
• Dr. Hiro Yoshida—Professor, Harvard Medical School
• Dr. E. Russel Ritenour—Professor, University of Minnesota
• Dr. Jim Holte—Emeritus Professor, University of Minnesota
• Dr. Milind Pimprikar—Caneus Canada
• Dr. Xiaoping Hu—Professor, Georgia Tech/Emory School
• Dr. Todd Parrish—Professor, North Western University
• Dr. Rinti Banerjee—Professor, IIT, Powai
• Dr. Amit K. Roy—Professor, IT-BHU
• Prof. T. Asokan—Professor, IIT Madras
• Prof. Amit Konar—Professor, Department of Electronics and Telecommunication,
JU
• Prof. Salil Sanyal—Professor, Department of Computer Science and Engineering,
JU
• Prof. D. Patronobis—Former Professor, Department of Instrumentation Engineering,
JU
• Prof. Amitava Gupta—Professor, Department of Power Plant Engineering, JU
• Prof. D.N. Tibarewala—Professor, School of Bioscience and Engineering, JU
• Prof. Chandan Sarkar—Department of Electronics Engineering, JU

Organizing Committee

Program Chair
Dr. Somsubhra Gupta, HOD-Department of Information Technology, JISCE
Publication Chair
Dr. Karabi Ganguly, Department of Biomedical Engineering, JISCE
Publication Committee
Dr. Sabyasachi Sen, HOD-Department of Physics
Dr. Indranath Sarkar, Department of Electronics and Communication Engineering,
JISCE
Partha Roy, Department of Electrical Engineering, JISCE
Program Committee xvii

Papun Biswas, Department of Electrical Engineering, JISCE

Ranjana Ray, Department of ECE, JISCE
Dr. Biswaroop Neogi, Department of Electronics and Communication Engineering,
JISCE
Finance Chair
Dr. Anindya Guha, HOD-Department of Humanities, JISCE
Finance Committee
Manas Paul, Department of Computer Application, JISCE
Anirban Patra, Department of Electronics and Communication Engineering, JISCE
Santanu Mondal, Department of Mathematics, JISCE
Rupankar Mukherjee, Department of Biomedical Engineering, JISCE
S.k. Malek, Accountant, JISCE
Reception and Registration Chair
Nilotpal Manna, HOD-Department of Electronics and Instrumentation Engineering,
JISCE
Reception and Registration Committee
Souvik Das, Department of Biomedical Engineering, JISCE
Bikash Dey, Department of Electronics and Communication Engineering, JISCE
Sourish Halder, Department of Electronics and Communication Engineering, JISCE
Suparna Dasgupta, Department of Information Technology, JISCE
Soumen Ghosh, Department of Electronics and Instrumentation Engineering,
JISCE
Sumanta Bhattacharya, Department of Computer Application, JISCE
Moumita Pal, Department of Electronics and Communication Engineering, JISCE
Sraboni Biswas, Administrative Staff, JISCE
Travel and Accomodation
Debashis Sanki, Department of Information Technology, JISCE
Debashis Majumder, Department of Mathematics, JISCE
Mainuck Das, Department of Electronics and Instrumentation Engineering, JISCE
Basudeb Dey, Department of Electrical Engineering, JISCE
Saptarshi Nandi, Department of Civil Engineering, JISCE
Kunal Banerjee, Department of Mechanical Engineering, JISCE
Hospitality
Debashis Sanki, Department of Information Technology, JISCE
Rupak Bhattacharjee, Department of Mathematics, JISCE
Aniruddha Ghosh, Department of Electronics and Communication Engineering,
JISCE
Dipankar Ghosh, Department of Mathematics, JISCE
Prolay Ghosh, Department of Information Technology, JISCE
xviii Program Committee

Saikat Chakraborty, Administrative Staff

Gourango Halder, Department of Biomedical Engineering, JISCE
G.C. Sarkar, HR Manager, JISCE
Sumindar Roy, Department of Electronics and Instrumentation Engineering, JISCE
Website and e-Communication
Soumyabrata Saha, Department of Information Technology, JISCE
Nilotpal Haldar, Department of Electronics and Communication Engineering,
JISCE
Sumanta Bhattacharya, Department of Computer Application, JISCE
Volunteer and Student Management
Avik Sanyal, CMS, JISCE
Subham Ghosh, Department of Electronics and Communication Engineering,
JISCE
Dr. Shubhomoy Singha Roy, Department of Physics, JISCE
Contents

Part I Medical Image Processing and Analysis

Proposed Intelligent System to Identify the Level of Risk

of Cardiovascular Diseases Under the Framework
of Bioinformatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Somsubhra Gupta and Annwesha Banerjee

Real Time Eye Detection and Tracking Method for Driver

Assistance System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Sayani Ghosh, Tanaya Nandy and Nilotpal Manna

Preprocessing in Early Stage Detection of Diabetic Retinopathy

Using Fundus Images. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Vijay M. Mane, D.V. Jadhav and Ramish B. Kawadiwale

Magnetic Resonance Image Quality Enhancement

Using Transform Based Hybrid Filtering . . . . . . . . . . . . . . . . . . . . . . 39
Manas K. Nag, Subhranil Koley, Chandan Chakraborty
and Anup Kumar Sadhu

Histogram Based Thresholding for Automated Nucleus

Segmentation Using Breast Imprint Cytology . . . . . . . . . . . . . . . . . . . 49
Monjoy Saha, Sanjit Agarwal, Indu Arun, Rosina Ahmed,
Sanjoy Chatterjee, Pabitra Mitra and Chandan Chakraborty

Separation of Touching and Overlapped Human Chromosome

Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
V. Sri Balaji and S. Vidhya

xix
xx Contents

Combination of CT Scan and Radioimmunoscintigraphy

in Diagnosis and Prognosis of Colorectal Cancer. . . . . . . . . . . . . . . . . 67
Sutapa Biswas Majee, Narayan Chandra Majee
and Gopa Roy Biswas

Enhanced Color Image Segmentation by Graph Cut Method

in General and Medical Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
B. Basavaprasad and M. Ravi

A New Approach for Color Distorted Region Removal in Diabetic

Retinopathy Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Nilarun Mukherjee and Himadri Sekhar Dutta

Part II Biomedical Instrumentation and Measurements

A New Heat Treatment Topology for Reheating of Blood Tissues

After Open Heart Surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Palash Pal, Pradip Kumar Sadhu, Nitai Pal and Prabir Bhowmik

Real Time Monitoring of Arterial Pulse Waveform Parameters

Using Low Cost, Non-invasive Force Transducer . . . . . . . . . . . . . . . . 109
S. Aditya and V. Harish

Selection of Relevant Features from Cognitive EEG Signals

Using ReliefF and MRMR Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 125
Ankita Mazumder, Poulami Ghosh, Anwesha Khasnobish,
Saugat Bhattacharyya and D.N. Tibarewala

Generalised Orthogonal Partial Directed Coherence as a Measure

of Neural Information Flow During Meditation. . . . . . . . . . . . . . . . . . 137
Laxmi Shaw, Subodh Mishra and Aurobinda Routray

An Approach for Identification Using Knuckle and Fingerprint

Biometrics Employing Wavelet Based Image Fusion
and SIFT Feature Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Aritra Dey, Akash Pal, Aroma Mukherjee
and Karabi Ganguly Bhattacharjee

Development of a Multidrug Transporter Deleted Yeast-Based Highly

Sensitive Fluorescent Biosensor to Determine the (Anti)Androgenic
Endocrine Disruptors from Environment . . . . . . . . . . . . . . . . . . . . . . 161
Shamba Chatterjee and Sayanta Pal Chowdhury
Contents xxi

Simulation of ICA-PI Controller of DC Motor in Surgical Robots

for Biomedical Application. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
Milan Sasmal and Rajat Bhattacharjee

Development of a Wireless Attendant Calling System

for Improved Patient Care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Debeshi Dutta, Biswajeet Champaty, Indranil Banerjee,
Kunal Pal and D.N. Tibarewala

A Review on Visual Brain Computer Interface . . . . . . . . . . . . . . . . . . 193

Deepak Kapgate and Dhananjay Kalbande

Design of Lead-Lag Based Internal Model Controller for Binary

Distillation Column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Rakesh Kumar Mishra and Tarun Kumar Dan

Clinical Approach Towards Electromyography (EMG) Signal

Capturing Phenomenon Introducing Instrumental Activity . . . . . . . . . 215
Bipasha Chakrabarti, Shilpi Pal Bhowmik, Swarup Maity
and Biswarup Neogi

Brain Machine Interface Automation System: Simulation

Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Prachi Kewate and Pranali Suryawanshi

Part III DSP and Clinical Applications

Cognitive Activity Classification from EEG Signals

with an Interval Type-2 Fuzzy System . . . . . . . . . . . . . . . . . . . . . . . . 235
Shreyasi Datta, Anwesha Khasnobish, Amit Konar and D.N. Tibarewala

Performance Analysis of Feature Extractors for Object Recognition

from EEG Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Anwesha Khasnobish, Saugat Bhattacharyya, Amit Konar
and D.N. Tibarewala

Rectangular Patch Antenna Array Design at 13 GHz Frequency

Using HFSS 14.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
Vasujadevi Midasala, P. Siddaiah and S. Nagakishore Bhavanam
xxii Contents

Automated Neural Network Based Classification of HRV

and ECG Signals of Smokers: A Preliminary Study . . . . . . . . . . . . . . 271
Suraj Kumar Nayak, Ipsita Panda, Biswajeet Champaty, Niraj Bagh,
Kunal Pal and D.N. Tibarewala

Reliable, Real-Time, Low Cost Cardiac Health Monitoring System

for Affordable Patient Care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Meghamala Dutta, Sourav Dutta, Swati Sikdar, Deepneha Dutta,
Gayatri Sharma and Ashika Sharma

Part IV Embedded Systems and Its Applications in Healthcare

An Ultra-Wideband Microstrip Antenna with Dual Band-Filtering

for Biomedical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Subhashis Bhattacharyya, Amrita Bhattacharya and Indranath Sarkar

Design of Cryoprobe Tip for Pulmonary Vein Isolation. . . . . . . . . . . . 307

B. Sailalitha, M. Venkateswara Rao and M. Malini

Designing of a Multichannel Biosignals Acquisition System

Using NI USB-6009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
Gaurav Kulkarni, Biswajeet Champaty, Indranil Banerjee,
Kunal Pal and Biswajeet Mohapatra

Arsenic Removal Through Combined Method Using Synthetic

Versus Natural Coagulant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
Trina Dutta and Sangita Bhattacherjee

Development of Novel Architectures for Patient Care Monitoring

System and Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
M.N. Mamatha

Review on Biocompatibility of ZnO Nano Particles . . . . . . . . . . . . . . . 343

Ananya Barman

Tailoring Characteristic Wavelength Range of Circular Quantum

Dots for Detecting Signature of Virus in IR Region. . . . . . . . . . . . . . . 353
Swapan Bhattacharyya and Arpan Deyasi

Methodology for a Low-Cost Vision-Based Rehabilitation System

for Stroke Patients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Arpita Ray Sarkar, Goutam Sanyal and Somajyoti Majumder
Contents xxiii

Coacervation—A Method for Drug Delivery . . . . . . . . . . . . . . . . . . . . 379

Lakshmi Priya Dutta and Mahuya Das

A Simulation Study of Nanoscale Ultrathin-Body

InAsSb-on-Insulator MOSFETs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
Swagata Bhattacherjee and Subhasri Dutta

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393

About the Editors

Dr. Somsubhra Gupta is presently the Head of the

Department of Information Technology, JIS College of
Engineering (An Autonomous Institution). He graduated
from the University of Calcutta and completed his
Master’s from Indian Institute of Technology, Kharag-
pur. He received Ph.D. from University of Kalyani. His
area of teaching is Algorithm and allied domains and
research area is Machine Intelligence. In research, he has
around 56 papers including Book Chapters so far in
national/international journals/proceedings and over 40
citations. He is Principal Investigator/Project Coordinator
of some Research projects (viz. RPS scheme AICTE).

Dr. Sandip Bag is Assistant Professor in the Department

of Biomedical Engineering, JIS College of Engineering,
Kalyani since 2005. He completed his Ph.D., Postgrad-
uation and Graduation from Jadavpur University, Kolk-
ata. Dr. Bag has published over 22 papers in national and
international journals and proceedings and also presented
papers in India and abroad. He received financial and
research grant from DST and UGC, respectively. He
developed various laboratories such as Biomaterials and
Biomedical Instrumentation laboratory in JISCE.

xxv
xxvi About the Editors

Dr. Karabi Ganguly is Assistant Professor in the

Department of Biomedical Engineering, JIS College of
Engineering, Kalyani. She received her Ph.D. degree
from Jadavpur University, India and completed her
Postrgaduation and Graduation from the University of
Calcutta. She published many papers in national/inter-
national journals and proceedings. She has been invited
as the reviewer in many international and national
Conferences. Her research interest includes Cellular
Biochemistry, Physiology, and Clinical Oncology.

Dr. Indranath Sarkar obtained his Ph.D. from Uni-

versity of Kalyani in the year 2013. He earned M.E.
from Jadavpur University in the year 2002 and B.E. in
Electronics and Communication Engineering from
Regional Engineering College (Presently known as
National Institute of Technology), Durgapur in the year
1999. He is presently working as Assistant Professor in
the Department of Electronics and Communication
Engineering, JIS College of Engineering Kalyani, India.
He has published many papers in national/international
journals and proceedings.

Papun Biwas is Assistant Professor in the Department

of Electrical Engineering, JIS College of Engineering,
Kalyani, West Bengal, India. He received his M.Tech.
degree in Electrical Engineering from the University of
Calcutta in 2007. Currently, he is a Research Fellow in
the Department of Computer Science and Engineering,
University of Kalyani. His research interests pertain to
different areas of Soft and Evolutionary Computing in
the area of Fuzzy Multi-objective Decision Making.
 Part I
Medical Image Processing and Analysis
Proposed Intelligent System to Identify
the Level of Risk of Cardiovascular
Diseases Under the Framework
of Bioinformatics

Somsubhra Gupta and Annwesha Banerjee

Abstract This paper proposed a method to implement an intelligent system to find out
the risk of cardiovascular diseases in human being. Genetics play a direct and indirect
role in increasing the risks of cardiovascular diseases. Habits and individual symptom
viz. suffering from diabetes, obesity and hypertension also can influence the risk of the
said diseases. Excessive energy accumulation in ones body can create fatal problem in
health. In this paper, method has been proposed to the proposed to investigate three
major factors i.e. family history of CVD, Other diseases and Average Energy
Expenditure and find out the of level of risks of cardiovascular diseases.

Keywords Bioinformatics

Cardiovascular disease
Energy expenditure




Genetics Intelligent systems Production system

1 Introduction

In recent times, the cardiovascular diseases are one of the major causes of mortality
in human beings. Numerous factors are there that increases the risks of CVD like
obesity, diabetes, hypertensions which in turn are being caused by less energy
expenditure in human being and heredity also plays a major role in the causing on
CVD. However, it is also likely that people with a family history of heart disease
share common environments and risk factors that increase their risk.
Work is one form of energy, often called mechanical energy. When to throw a
ball or run a mile, work has been done; mechanical energy has been produced.
The sun is the ultimate source of energy. Solar energy is harnessed by plants,
through photosynthesis, to produce plant carbohydrates, fats, or proteins, all forms

S. Gupta (&)
A. Banerjee

Department of Information Technology, JIS College of Engineering,
Kalyani Block A Phase III, Nadia, West Bengal, India
e-mail: [email protected]
A. Banerjee
e-mail: [email protected]

© Springer India 2015 3

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_1
4 S. Gupta and A. Banerjee

of stored chemical energy. When humans consume plant and animal products, then
carbohydrates, fats, and proteins undergo a series of metabolic changes and are
utilized to develop body structure, to regulate body processes, or to provide a
storage form of chemical energy. Less energy expenditure causes obesity that in
turn increases the risk on CVD. By 2015, nearly one in every three people
worldwide is projected to be overweight, and one in ten is expected to be obese [1].
Cardiovascular disease is one of the alarming threats. World Health Report
showed that cardiovascular disease will be the major cause of death [2]. Sitting time
and no exercise activity have been linked in epidemiological studies to rates of
metabolic syndrome, type 2 diabetes, obesity, and CVD Regional Body fat distri-
bution is one of the major factors that increase the CVD [3]. As a definite cause of
cardiovascular morbidity and mortality [4] it is important to consider the potential
impact of dietary sugar on weight gain.
Sugar intake can increase carbohydrate fuel reserves and physical performance
[5]. There have been a number of studies that link sugar consumption to hyper-
tension in animals [6]. In humans, there is one report that high dietary sugar intake
enhances the risk of CHD in diabetic individuals who use diuretics [7].
Hypertension increases the risk of stroke in individuals [8]. It was been reported
by Kornegay et al. [9] that there is a reasonable agreement between proband-
reported family history of stroke and self-reported personal history of stroke in
members of the proband’s family. It is been examine the role of sedentary
behaviors, especially sitting, on mortality, cardiovascular disease, type 2 diabetes,
metabolic syndrome risk factors, and obesity [10].
Significant evidence for linkage heterogeneity among hypertensive sib pairs
stratified by family history of stroke suggests the presence of genes influencing
susceptibility to both hypertension and stroke [11].
Physical inactivity may induce negative effects on relatively fast-acting cellular
processes in skeletal muscles or other tissues regulating risk factors like plasma
triglycerides and HDL cholesterol [12–14].
More than 90 % of the calories expended in all forms of physical activity were
due to this pattern of standing and no exercise ambulatory movements because
individuals did not exercise and because the energy expenditure associated with no
exercise activity thermogenesis (NEAT) while sitting was small [15]. Obviously, 6–
12 h/day of no exercise activity is beyond what anyone would exercise regularly.
Laboratory rats housed in standard cages without running wheels also recruit
postural leg muscles for 8 h/day [16].
In this paper, how to find the influence of heredity and energy expenditure in the
cardiovascular disease has been presented.

2 Genetic Pattern and Inheritance

DNA is treated as “blue print of life”. It contains all the information to create life.
DNA contains the information needed to create the amino acids sequences of
proteins. The unit of building block of DNA Adenine (A), Cytosine (C), Guanine
Proposed Intelligent System to … 5

Fig. 1 mRNA to protein

(G), and Thymine (T) are the four bases in DNA. A pairs with T that is 2H bond
and C pairs with G that is 3H bond [17].
Protein is a linear sequence of amino acids, shown in the Fig. 1 form a very long
chain via peptide linkage. Gene is a segment of DNA. Inheritance pattern are the
predictable pattern seen in the transmission of genes from one generation to the next
and their expression in the organism that possesses them.
Offspring inherits genotypes from their parents. Diseases runs in family hier-
archy. A team of Portugal Hospital de Criancas studied the issue of genetic sus-
ceptibility in stokes [18]. It was concluded that two genes did contribute to the
development of disease.

2.1 Genetics and Cardiovascular Diseases

Life style and environment play a role in causing cardiovascular disease in indi-
vidual. But it is also proved by research that heredity also has a major role in
cardiovascular disease. Gene mutation and polymorphism are sometimes directly be
related to cardiovascular diseases. A study of University of Texas showed that
chromosome carries polymorphism for hypertension and stroke for Caucasian
patient and chromosome 19 for African and American [19].
It was been reported by Kornegay et al. [9] that there is a reasonable agreement
between proband-reported family history of stroke and self-reported personal his-
tory of stroke in members of the proband’s family. The accuracy of reporting is
high for other common diseases, such as myocardial infarction [20] coronary heart
disease, diabetes, hypertension, and asthma [21]. Positive family history was
defined by proband-reported history of stroke or cerebral hemorrhage diagnosed by
a physician for either biologic parent or at least 1 full biologic sibling.

3 Other Diseases and CVD

Other diseases also influences the heart disease that genetic also responsible indi-
rectly for the heart diseases. Hypertension, obesity, diabetes play a major role in
increasing risk of heart diseases and at the same time behaviours like smoking
habits also influences the probability of heart diseases. In humans, there is one
6 S. Gupta and A. Banerjee

report that high dietary sugar intake enhances the risk of CHD in diabetic indi-
viduals who use diuretics [7].
A number of studies have shown that specific topographic features of adipose
tissue are associated with metabolic complications that are considered as risk fac-
tors for CVD such as insulin resistance, hyperinsulinemia, glucose intolerance and
type II diabetes mellitus, hypertension, and changes in the concentration of plasma
lipids and lipoproteins. Metabolic correlates the body fat distribution [3].

4 Energy Expenditure and Disease

Recent studies have suggested that respiratory diseases, such as chronic obstructive
pulmonary disease (COPD) and obstructive sleep apnea syndrome (OSAS), influ-
ence energy expenditure (EE) [22]. Energy can be measured in either joules or
calories. A joule (J) can be defined as the energy used when 1 kilogram (kg) is
moved 1 metre (m) by the force of 1 newton (N). A calorie (cal) can be defined as
the energy needed to raise the temperature of 1 g of water from 14.5 to 15.50 °C. In
practice, both units are used just as different units are used to measure liquids, e.g.
pints, liters. One calorie is equivalent to 4.184 J.
There are three main components of Total Energy Expenditure (TEE) in humans:
1. Basal Metabolic Rate (BMR)—Energy expended at complete rest in a post-
absorptive state; accounts for approximately 60 % of TEE in sedentary
individuals.
2. Thermal Effect of Food (TEF)—Increase in energy expenditure associated
with digestion, absorption, and storage of food and nutrients; accounts for
approximately 10 % of TEE.
3. Energy Expenditure of Activity—Further classified as Exercise-related
Activity Thermo genesis
4. Growth—The energy cost of growth has two components: (1) the energy
needed to synthesize growing tissues; and (2) the energy deposited in those
tissues. The energy cost of growth is about 35 % of total energy requirement
during the first 3 months of age, falls rapidly to about 5 % at 12 months and
about 3 % in the second year, remains at 1–2 % until mid-adolescence, and is
negligible in the late teens.
5. Pregnancy—During pregnancy, extra energy is needed for the growth of the
foetus, placenta and various maternal tissues, such as in the uterus, breasts and
fat stores, as well as for changes in maternal metabolism and the increase in
maternal effort at rest and during physical activity.
6. Lactation—The energy cost of lactation has two components: (1) the energy
content of the milk secreted; and (2) the energy required to produce that milk.
Well-nourished lactating women can derive part of this additional requirement
from body fat stores accumulated during pregnancy.
Proposed Intelligent System to … 7

Associated with active sports or exercise and Non-Exercise Activity Thermo

genesis (NEAT) associated with activities of daily living, fidgeting, spontaneous
muscle contraction etc. Exercise-related Activity.

4.1 Energy Expenditure Measurement

There are few methods are available to measure the energy of human body. Following
tables summaries the commonly used methods to measure energy (Table 1):
Over the last few decades, research in public health has associated inactivity
with a number of ailments and chronic diseases, such as colon cancer, type II
diabetes, osteoporosis and coronary heart disease.
Humans have been increasingly spending more time in sedentary behaviours
involving prolonged sitting. It is found by research that most of the time people
sitting physically idle i.e. with out any exercise. The amount of exercise is very
limited, generally to <150 min/week [23].

Table 1 Human body energy measuring methods

Sl. No. Methods Description
1. Accelerometer Techniques for measuring energy expenditure involve
either measuring heat loss directly (direct calorimetry)
or measuring a proxy of heat loss (indirect calorimetry)
such as oxygen (O2) consumption or carbon dioxide
(CO2) production [27]
2. Direct calorimeter Done under control-led laboratory conditions in
insulated chambers that measure changes in air
temperature associated with the heat being released by
a subject
3. Indirect calorimeter Measures inspired and expired oxygen and carbon
dioxide in order to calculate resting energy expenditure
(REE) and respiratory quotient (RQ) using the
abbreviated Weir equation, where: REE (Kcal/day)
= [(VO2 × 3.94) + (VCO2 × 1.11)] × 1,440 min/day
VCO2 (carbon dioxide production) VO2 (oxygen
consumption)
4. DLW Technique in which stable isotopes of hydrogen and
oxygen in water (2H2 18O) are ingested
5. First beat beat-by-beat Probably the most frequently used indirect parameter
based heart rate method in EE-assessment, mainly due to following reasons:
• HR is easy to measure
• EE is easily accessible from HR data
• HR-based EE-estimates are relatively accurate in
steady exercise conditions [28]
8 S. Gupta and A. Banerjee

Fig. 2 Physical activity rate versus mortality

Figure 2 shows that the less activity tends to the risk of premature myocardial
infarction (A) and mortality from coronary artery disease (C) [24]. These general
findings were subsequently confirmed in studies in middle-aged women (B) [25]
and an elderly group (D) [26].

5 Proposed Method

Through our paper we proposed a intelligent system that working on the prediction
of probability Cardiovascular diseases. Some Genetics test to identify the gene
which influences the CVD but theses types of tests are costly from most the
countries. So we proposed the methods depending upon the following parameters.
Figure 3 shows the block diagram of the proposed method.

Family History of Cardiovascular disease: We proposed to investigate up to two

level of hierarchy in parents.
Here H is the set of data values assigned for different level of CVD associated in
family hierarchy.
H = {0, 0.25, 0.50, 0.75, 1}
0: no family history of CVD
0.25: CVD history in one level of ancestor in parents
Proposed Intelligent System to … 9

Family History of CVD Energy Expense

Other
Diseases Decision Making

Prediction

Fig. 3 Proposed intelligent systems to predict: block diagram

0.50: CVD history in one level of ancestors in either of the parents

0.75: CVD history in two levels of ancestors in either of the parents
1: CVD history in two level ancestors in parents
Energy Expenditure (EE)

Average weekly energy expenditure will be measured for individuals.

EE is the set of data values for energy expenditure amount.
EE = {0,0.20, 0.40, 0.60, 0.80, 1}
0: 0 % of EE
0.20: (>0 % to ≤25 % of EE)
0.60: (>25 % to ≤50 % of EE)
0.80: (>50 % to ≤75 % of EE)
1: (>75 % to ≤100 % of EE)
Other diseases: Hypertension, Obesity, diabetes:

OD is the set of data values for other disease that influence the risk of CVD
OD = {0, 0.25, 0.50, 1}
0: Having none of the diseases
0.25: Having one diseases
0.50: Having two diseases
1: Having all three diseases
The production system of the proposed Intelligent System has been presented
below diagrammatically in the Fig. 3.
Probability of CVD:

P ðCVDÞ ¼ w1 ðV1Þ þ w2 ðV2Þ þ w3 ðV3Þ

where w1, w2 and w3 are the weight values for the considered three factor that
influence the risk of CVD i.e. w1 is the weight for family history, w2 is weight for
energy expenditure and w3 is the weight for having other diseases.
10 S. Gupta and A. Banerjee

Fig. 4 Probability trends of

CVD considering three factor
of our proposed methods

We considered the value of weight as follows:

w: {0.5, 0.3, 0.2}

v1, v2 and v3 are the data value for the three factors being considered for our
proposed method i.e. v1 is for family history, v2 is for EE and v3 is for having other
diseases.
Figure 4 shows the Probability trend of CVD among individuals considering the
Family History of CVD, Total Energy expenditure and Effects of other diseases.

6 Implication and Future Scope

From the investigation of our proposed method it is being proved that low energy
expenditure, previous family history of CVD and having hypertension. Obesity and
diabetes increases the risk of CVD. Previous family history of CVD and having
other diseases are directly proportional in causing of CVD where Energy Expen-
diture is inversely proportional to the causing of CVD.
1. Family History of CVD ∞ CVD
2. Other Diseases ∞ CVD
3. Energy expenditure ∞ 1/CVD
Our future scope of study is to gene wide scan to find out the probability of CVD
and other diseases among family history and to find out the relationship among
them.
Proposed Intelligent System to … 11

7 Conclusions

In this paper, the heredity of CVD, Other diseases and Energy expenditure of
human body has been investigated and then implement an intelligent machine to
find out the probability of CVD in individuals has been proposed. The proposed
method is deterministic and generalized one but in practical the establishment of the
factor is trivial.

References

1. World Health Organization (2011) Obesity and Overweight. Fact sheet No. 311. http://www.
who.int/mediacentre/factsheets/fs311/en/index.html. Accessed 20 Jan 2011
2. World Health Organization (2002) World health report 2002: reducing risks, Promoting
healthy life. WHO, Geneva, 2002
3. Despres JP, Moorjani S, Lupien PJ, Tremblay A, Adeau A, Bouchard C (2014) Obesity favors
apolipoprotein E- and C-III-containing high density lipoprotein subfractions associated with
risk of heart disease. J Lipid Res 55(10):2167–2177
4. Eckel RH, Krauss RM (1998) American heart association call to action obesity as a major risk
factor for coronary heart disease. AHA Nutr Committee Circ 97:2099–2100
5. Hill JO, Prentice AM (1995) Sugar and body weight regulation. Am J Clin Nutr 62(suppl
1):264S–273S
6. Preuss HG, Zein M, MacArthy P et al (1998) Sugar-induced blood pressure elevations over the
lifespan of three substrains of Wistar rats. J Am Coll Nutr 17:36–47
7. Sherman WM (1995) Metabolism of sugars and physical performance. Am J Clin Nutr 62(
suppl):228S–241S
8. Klungel O, Stricker B, Paes A, Seidell J, Bakker A, Voko Z, Breteler M, Boer A (1999)
Excess stroke among hypertensive men and women attributable to undertreatment of
hypertension. Stroke 30:1312–1318
9. Kornegay C, Liao D, Bensen J, Province M, Folsom A, Ellison C (1997) The accuracy of
proband-reported family history of stroke: the FHS Study. Am J Epidemiol 145:S82(Abstract)
10. Hamilton T, Deborah G. Hamilton T, Theodore W. Zderic1 (2007) Role of low energy
expenditure and sitting in obesity, metabolic syndrome, type 2 diabetes, and cardiovascular
disease. Marc Diabetes 56(11):2655−2667
11. Morrison AC, Brown A, Kardia SLR, Turner ST, Boerwinkle E (2003) Evaluating the context-
dependent effect of family history of stroke in a genome scan for hypertension. Stroke
34:1170–1175. doi:10.1161/01.STR.0000068780.47411.16
12. Bey L, Hamilton MT (2003) Suppression of skeletal muscle lipoprotein lipase activity during
physical inactivity: a molecular reason to maintain daily low-intensity activity. J Physiol
551:673–682
13. Hamilton MT, Hamilton DG, Zderic TW (2004) Exercise physiology versus inactivity
physiology: an essential concept for understanding lipoprotein lipase regulation. Exerc Sport
Sci Rev 32:161–166
14. Zderic TW, Hamilton MT (2006) Physical inactivity amplifies the sensitivity of skeletal
muscle to the lipid-induced downregulation of lipoprotein lipase activity. J Appl Physiol
100:249–257
15. Levine JA, Lanningham-Foster LM, McCrady SK, Krizan AC, Olson LR, Kane PH, Jensen
MD, Clark MM (2005) Interindividual variation in posture allocation: possible role in human
obesity. Sci 307:584–586
16. Hennig R, Lømo T (1985) Firing patterns of motor units in normal rats. Nat 314:164–166
12 S. Gupta and A. Banerjee

17. Wong L (2011) Some new results and tools for protein function prediction, RNA target site
prediction, genotype calling, environmental genomics, and more. J Bioinform Comput Biol 9
(6):5−7
18. Wang JG, Staessen Ja (2000) Genetic polymorphism in the rennin-angiotenisn system:
relevance for susceptibility to cardio vascular diseases. Eur J Phamacol 410(2–3):289–302
19. Morrison AC, Brown A, Kardia SL, Turner ST Boerwinkle E Genetic Epidemiology Network
of Arteriopathy (GENOA) Study
20. Kee F, Tiret L, Robo J, Nicaud V, McCrum E, Evans A, Cambien F (1993) Reliability of
reported family history of myocardial infarction. BMJ 307:1528–1530
21. Bensen J, Liese A, Rushing J, Province M, Folsom A, Rich S, Higgins M (1999) Accuracy of
proband reported family history: the NHLBI Family Heart Study (FHS). Genet Epidemiol
17:141–150
22. Gregersen NT, Chaput JP, Astrup A, Tremblay A (2008) Human Energy Energy expenditure
and respiratory diseases: is there a link. Expert Rev Respir Med 2(4):495–503. doi:10.1586/
17476348.2.4.495
23. Morrison AC, Brown A, Kardia SL (2003) Evaluating the context-dependent effect of family
history of stroke in a genome scan for hypertension. Stroke 34(5):1170–1175 (Epub 2003 Apr
24)
24. Morris JN, Heady JA, Raffle PA, Roberts CG, Parks JW (1953) Coronary heart-disease and
physical activity of work. Lancet 265:1053–1057
25. Weller I, Corey P (1998) The impact of excluding non-leisure energy expenditure on the
relation between physical activity and mortality in women. Epidemiol 9:632–635
26. Manini TM, Everhart JE, Patel KV, Schoeller DA, Colbert LH, Visser M, Tylavsky F, Bauer
DC, Goodpaster BH, Harris TB (2006) Daily activity energy expenditure and mortality among
older adults. JAMA 296:171–179
27. An Energy Expenditure Estimation Method Based on Heart Rate, Measurement Firstbeat
Technologies Ltd
28. Kruger J, Yore MM, Kohl HW 3rd (2007) 3rd Leisure-time physical activity patterns by
weight control status: 1999–2002 NHANES. Med Sci Sports Exerc 39:788–795
Real Time Eye Detection and Tracking
Method for Driver Assistance System

Sayani Ghosh, Tanaya Nandy and Nilotpal Manna

Abstract Drowsiness and fatigue of automobile drivers reduce the drivers’ abilities
of vehicle control, natural reflex, recognition and perception. Such diminished
vigilance level of drivers is observed at night driving or overdriving, causing
accident and pose severe threat to mankind and society. Therefore it is very much
necessary in this recent trend in automobile industry to incorporate driver assistance
system that can detect drowsiness and fatigue of the drivers. This paper presents a
nonintrusive prototype computer vision system for monitoring a driver’s vigilance
in realtime. Eye tracking is one of the key technologies for future driver assistance
systems since human eyes contain much information about the driver’s condition
such as gaze, attention level, and fatigue level. One problem common to
many eye tracking methods proposed so far is their sensitivity to lighting condition
change. This tends to significantly limit their scope for automotive applications.
This paper describes real time eye detection and tracking method that works under
variable and realistic lighting conditions. It is based on a hardware system for the
real-time acquisition of a driver’s images using IR illuminator and the software
implementation for monitoring eye that can avoid the accidents.

Keywords Vigilance level Eye tracking Deformable template Edge detection



Template-based correlation IR illuminator

S. Ghosh (&)
T. Nandy
N. Manna

Department of Electronics and Instrumentation Engineering,
JIS College of Engineering, Kalyani, Nadia 741235, India
e-mail: [email protected]
T. Nandy
e-mail: [email protected]
N. Manna
e-mail: [email protected]

© Springer India 2015 13

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_2
14 S. Ghosh et al.

1 Introduction

The increasing number of traffic accidents due to a drivers’ diminished vigilance

level is a serious problem for the society. Drivers’ abilities of vehicle control,
natural reflex, recognition and perception decline due to drowsiness and fatigue,
reducing the drivers’ vigilance level. These pose serious danger to their own lives as
well as lives of other people. According to the U.S. National Highway Traffic
Safety Administration (NHTSA), drowsiness and falling asleep while driving are
responsible for at least 100,000 automobile crashes annually [1, 2]. An annual
average of roughly 40,000 nonfatal injuries and 1,550 fatalities result from these
crashes. These figures only present the casualties happening during midnight to
early morning, and underestimate the true level of the involvement of drowsiness
because they do not include crashes during daytime hours. Vehicles with driver
intelligence system that can detect drowsiness of the driver and send alarm may
avert fatal accidents.
Several efforts to develop active safety systems have been reported in the lit-
erature [3–11] for reducing the number of automobile accidents due to reduced
vigilance. Among deferent techniques, measurement of physiological conditions
like brain waves, heart rate, and pulse rate [10, 12] yields maximum detection
accuracy. However these techniques are intrusive as the sensing elements (elec-
trodes) for measurement require physical contact with drivers causing annoyance.
Less intrusive techniques like eyelid movement or gaze or head movement moni-
toring techniques [13] with head-mounted devices as eye tracker or special contact
lens also deliver good results. These techniques, though less intrusive, are still not
practically acceptable. A driver’s state of vigilance can also be characterized by the
behaviors of the vehicle he/she operates. Vehicle behaviors including speed, lateral
position, turning angle, and moving course are good indicators of a driver’s
alertness level. While these techniques may be implemented non-intrusively, they
are, nevertheless, subject to several limitations including the vehicle type, driver
experiences, and driving conditions [4].
Fatigue in people can be easily observed by certain visual behaviors and changes
in their facial features like the eyes, head, and face. The image of a person with
reduced alertness level exhibits some typical visual characteristics that include slow
eyelid movement [14, 15], smaller degree of eye openness (or even closed), fre-
quent nodding [16], yawning, gaze (narrowness in the line of sight), sluggish in
facial expression, and sagging posture. To make use of these visual cues,
increasingly popular and non-invasive approach for monitoring fatigue is to assess a
driver’s vigilance level through visual observation of his/her physical conditions
using a camera and state-of-the-art technologies in computer vision. Techniques
using computer vision are aimed at extracting visual characteristics that typically
characterize a driver’s vigilance level from his/her video images. In a recent
workshop [17] sponsored by the Department of Transportation (DOT) on driver’s
vigilance, it is concluded that computer vision represents the most promising non-
invasive technology to monitor driver’s vigilance.
Real Time Eye Detection and Tracking … 15

There are many literatures reporting the fatigue monitoring systems based on
active real-time image processing technique [3–7, 9, 10, 15, 18–21]. Detection of
driver fatigue is primarily focused in efforts. Characterization of a driver’s mental
state from his facial expression is discussed by Ishii et al. [9]. A vision system from
line of sight (gaze) to detect a driver’s physical and mental conditions is proposed
by Saito et al. [3]. A system for monitoring driving vigilance by studying the eyelid
movement is described by Boverie et al. [5] and results are revealed to be very
promising. A system for detection of drowsiness is explained at Ueno et al. [4] by
recognizing the openness or closeness of driver’s eyes and computing the degree of
openness. Qiang et al. [18] describes a real-time prototype computer vision system
for monitoring driver vigilance, consisting of a remotely located video CCD
camera, a specially designed hardware system for real-time image acquisition and
various computer vision algorithms for simultaneously, real-time and non-intru-
sively monitoring various visual bio-behaviors typically characterizing a driver’s
level of vigilance. The performance of these systems is reported to be promising
and comparable to the techniques using physiological signals. This paper focuses an
effort to develop a low cost hardware system that may be incorporated at dashboard
of vehicles to monitor eye images movements pertaining to drowsiness of driver.
The paper is organized with background theory describing various processes of eye
detection, followed by proposed scheme and implementation. Finally experimental
observations and results are tabulated and discussed.

2 Background Theory

To analyze the eye movement and tracking, image processing technique is

employed which treats images as two dimensional signals while applying already
set signal processing methods. It is a type of signal dispensation in which input is
image, like video frame or photograph and output may be image or characteristics
associated with that image. The images from camera are converted into digital form
that are thereby enhanced and performed some filtering and logic operations on
them, to extract some useful and desired information.
Methods used for eye detection and tracking rely on various prior assumptions of
the image data and in general two classes of approaches exist. One common and
effective approach is to exploit active illumination from infrared (IR) light emitters.
Through appropriate synchronization schemes and by using the reflective properties
of the pupil when exposed to near infrared light (dark or bright pupil effects) the eye
can be detected and tracked effectively. In addition to controlling the light condi-
tions, IR also plays an important role for some gaze estimation methods. Other
approaches avoid the use of active illumination, and rely solely on natural light. But
eye detection much difficult in absence of active illumination as some assumptions
on the image data is to be made. A method based on active light is the most
predominate in both research and in commercial systems. Ebisawa and Satoh [22]
16 S. Ghosh et al.

use a novel synchronization scheme in which the difference between images

obtained from on axis and off axis light emitters are used for tracking. Kalman
filtering, the Mean shift algorithm, and combinations of Kalman and mean shift
filtering are applied for eye tracking. The success of these approaches is highly
dependent on external light sources and the apparent size of the pupil. Efforts are
made to focus on improving eye tracking under various light conditions. Sun light
and glasses can seriously disturb the reflective properties of IR light.
Eye tracking and detection methods fall broadly within three categories, namely
deformable templates, appearance-based, and feature-based methods. Deformable
template and appearance-based methods are based on building models directly on
the appearance of the eye region while the feature-based method is based on
extraction of local features of the region. In general appearance models detect and
track eyes based on the photometry of the eye region. A simple way of tracking
eyes is through template-based correlation. Tracking is performed by correlation
maximization of the target model in a search region. Grauman et al. [23] uses
background subtraction and anthropomorphic constraints to initialize a correlation-
based tracker. Matsumoto and Zelinsky [24] present trackers based on template
matching and stereo cameras. Excellent tracking performance is reported, but the
method requires a fully calibrated stereo setup and a full facial model for each user.
The appearance of eye regions share commonalities across race, illumination and
viewing angle. Rather than relying on a single instance of the eye region, the eye
model can be constructed from a large set of training examples with varying pose
and light conditions. Based on the statistics of the training set a classifier can be
constructed for detection purposes over a larger set of subjects. Eye region local-
ization by Eigen images uses a subset of the principal components of the training
data to construct a low-dimensional object subspace to represent the image data.
Recognition is performed by measuring distances to the object subspace. The
limitations of the methods, which are purely based on detection of eyes in indi-
vidual frames, are that they do not make use of prior information from previous
frames which can be avoided by temporal filtering.
Deformable template-based method relies on a generic template which is mat-
ched to the image. In particular deformable templates, an eye model is constructed
in which the eye is located through energy minimization. The system is to be robust
to the variations of the template and the actual image for analytical approximations.
Yuille et al. [25] uses statistical measures in a deformable template approach to
account for statistical variations. The method uses an idealized eye consisting of
two regions with uniform intensity. One region corresponds to the iris region and
the other the area of the sclera. Ivins and Porrill [26] describe a method of tracking
the three-dimensional motion of the iris in a video sequence. A five-parameter
scalable and deformable model is developed to relate translations, rotation, scaling
due to changes in eye-camera distance, and partial scaling due to expansion and
contraction. The approximate positions of the eyes are then found by anthropo-
morphic averages. Detected eye corners are used to reduce the number of iterations
of the optimization of a deformable template. This model consists of parabolas for
Real Time Eye Detection and Tracking … 17

the eyelids and a subset of a circle for iris outline. A speedup is obtained compared
to Yuille et al. [25] by exploiting the positions of the corners of the eye. This
method requires the presence of four corners on the eye, which, in turn, only occur
if the iris is partially occluded by the upper eyelid. When the eyes are wide open,
the method fails as these corners do not exist. Combination of deformable template
and edge detection is used for an extended iris mask to select edges of iris through
an edge image. The template is initialized by manually locating the eye region
along with its parameters. Once this is done the template is allowed to deform in an
energy minimization manner. The position of the template in an initial frame is used
as a starting point for deformations that are carried out in successive frames. The
faces must be nearly frontal-view and the image of the eyes should be large enough
to be described by the template. The deformable template-based methods seem
logical and are generally accurate. They are also computationally demanding,
require high contrast images and usually needs to be initialized close to the eye.
While the shape and boundaries of the eye are important to model so is the texture
within the regions. For example the sclera is usually white while the region of the
iris is darker. Larger movements can be handled using Active Appearance Models
for local optimization and a mean shift color tracker. These effectively combine
pure template-based methods with appearance methods. This model shares some of
the problems with template-based methods; theoretically it should be able to handle
changes in light due to its statistical nature. In practice they are quite sensitive to
these changes and especially light coming from the side can have a significant
influence on their convergence.
Feature-based methods extract particular features such as skin-color, color dis-
tribution of the eye region. Kawato et al. [27] use a circle frequency filter and
background subtraction to track the in-between eyes area and then recursively
binaries a search area to locate the eyes. Sommer et al. [28] utilize Gabor filters to
locate and track the features of eyes. They construct a model-based approach which
controls steerable Gabor filters: The method initially locates particular edge (i.e. left
corner of the iris) then use steerable Gabor filters to track the edge of the iris or the
corners of the eyes. Nixon demonstrates the effectiveness of the Hough transform
modeled for circles for extracting iris measurements, while the eye boundaries are
modeled using an exponential function. Young et al. [29] show that using a head
mounted camera and after some calibration, an ellipse model of the iris has only
two degrees of freedom (corresponding to pan and tilt). They use this to build a
Hough transform and active contour method for iris tracking using head mounted
cameras. propose the Fast Radial Symmetry Transform for detecting eyes in which
they exploit the symmetrical properties of the face. Explicit feature detection (such
as edges) in eye tracking methods relies on thresholds. In general defining
thresholds can be difficult since light conditions and image focus change. Therefore,
methods on explicit feature detection may be vulnerable to these changes.
In this paper real time eye detection and tracking method is presented that works
under variable and realistic lighting conditions which is applicable to driver assis-
tance systems. Eye tracking is one of the key technologies for future driver assistance
18 S. Ghosh et al.

systems since human eyes contain much information about the driver’s condition
such as gaze, attention level, and fatigue level. Thus, non-intrusive methods
for eye detection and tracking are important for many applications of vision-based
driver-automotive interaction. One common problem to many eye tracking methods
is their sensitivity due to lighting condition change. This tends to significantly limit
the scope for automotive applications. By combining image processing and IR light
the proposed method can robustly track eyes.

3 Proposed Scheme

To detect and track eye images with complex background, distinctive features of
user eye are used. Generally, an eye-tracking and detection system can be divided
into four steps: (i) Face detection, (ii) Eye region detection, (iii) Pupil detection and
(iv) Eye tracking.
Image processing technique is incorporated for detection of these. Figure 1
illustrates the scheme. Camera is incorporated in the dashboard of vehicle which
takes the images of the driver regularly at certain interval. From the images first the
face portion is recognized from the complex background. It is followed by eye
region detection and thereafter the pupil or eyelid detection. The detection algo-
rithm finally detects the eyelid movement or closeness and openness of eyes. In the
proposed method, eye detection and tracking are applied on testing sets, gathered
from different images of face data with complex backgrounds. This method com-
bines the location and detection algorithm with the grey prediction for eye tracking.
The accuracy and robustness of the system depends on consistency of image
acquisition of the driver face in real time under variable and complex background.
For this purpose the driver’s face is illuminated using a near-infrared (NIR)
illuminator. It serves three purposes:

Eye Images Face Eye Region

Camera Detection Detection

Pupil Eye
Detection Tracking Alarm

Fig. 1 Image acquisition scheme

Real Time Eye Detection and Tracking … 19

• It minimizes the impact of different ambient light conditions, and hence the
image quality is ensured under varying real-world conditions including poor
illumination, day, and night;
• It allows producing the bright pupil effect, which constitutes the foundation for
detection and tracking the visual cues.
• As the near-infrared illuminator is barely visible to the driver, any interference
with the driver’s driving will be minimized.
If the eyes are illuminated with a NIR illuminator at certain wavelength beaming
light along the camera optical axis, a bright pupil can be obtained. At the NIR
wavelength, almost all IR light is reflected from the pupils along the path back to
the camera. Thus bright pupil effect is produced which is very much similar to the
red eye effect in photography. The pupils appear dark if illuminated off the camera
optical axis, since the reflected light will not enter the camera lens which is called
dark pupil effect. It is physically difficult to place IR light-emitting diodes (LEDs)
as illuminators along the optical axis since it may block the view of the camera,
limiting the camera’s operational field of view. Therefore quite a few numbers of IR
illuminator LEDs are placed evenly and symmetrically along the circumference of
two coplanar concentric rings, the center of both rings coincides with the camera
optical axis as shown at Fig. 2. In the proposed scheme, the camera acquires the
images of face of the driver at certain interval. Every time the image is analyzed and
bright pupil effect is detected. Whenever dark pupil effect is detected i.e., eyelid is
at closed condition at prolonged time, it may be assumed that driver’s vigilance
level has been diminished. Subsequently alarm is activated to draw the attention of
the driver.

Fig. 2 IR illuminator with

camera
20 S. Ghosh et al.

4 Implementation

A laboratory model has been developed to implement above scheme. A web camera
with IR illuminators has been employed focusing the face region of a person
(driver) to acquire the images of face. The acquired image signal is fed to Data
Acquisition Card and subsequently to a microcontroller. The microcontroller
analyses the images and detects the pupil characteristics. If the eyelid is closed for
several seconds, it may be assumed that the drowsiness occurred to the person and
alarm is activated by the microcontroller. The circuit scheme is shown at Fig. 3.
Microcontroller ATMEGA 8 is employed here in association with voltage regulator
IC 7805 and driver IC L2930 for buzzer.
To find the position of pupil, first, face region must be separated from the rest of
the image using boundaries function, which is a process of segmentation. This will
cause the images background to be non-effective. Region prop technique is used to
separate a region from total face and the region containing eyes and eyebrow. This
will result in decreasing the computational complexity. Finally, in proposed method
points with the highest values are selected as the eye candidate’s using centroid
function. The eye region is well detected among these points. If eye is detected then
the next frame is simulated, and if the eye is not detected then a signal is passed to
the micro controller for raising the alarm. The indicator also turns red. And when
the eye is detected the indicator turns green with raising no alarm. This is how the
system can awake the driver in long drive or in fatigue condition. The imple-
mentation flow chart is given at Fig. 4.

Fig. 3 Circuit scheme

Real Time Eye Detection and Tracking … 21

Fig. 4 Flow chart

START

INPUT COM PORT NAME FROM

OPEN COM PORT WITH SPECIFIC

BAUD RATE

INITIALIZE CAMERA

FORM STRUCTURING ELEMENT

FOR NOISE REMOVAL

GET A SNAPSHOT FROM VIDEO

CONVERT PICTURE TO RGB

CONVERT PICTURE TO BINARY

STRUCTURING ELEMENT TO
REMOVE NOISE

EXECUTE FOR BOUNDARY ELEMENTS

FIND AREA OF ALL BOUNDARY

ELEMENTS

YES NO
IF AREA IS WITHIN
PRE-SPECIFIED LIMITS

EYE DETECTED EYE NOT DETECTED

NO
IF N NUMBER OF
SNAPSHOT ANALYZED

YES

GO FOR NEXT RAISE ALARM

22 S. Ghosh et al.

Fig. 5 Experimental results

OBSERVATION 1

OBSERVATION 2

OBSERVATION 3

OBSERVATION 4

OBSERVATION 5

OBSERVATION 6

OBSERVATION 7
Real Time Eye Detection and Tracking … 23

Table 1 Observations on alarm conditions with respect to the eye condition

Eye condition Alarm condition
Observation 1 Open Green LED is ON. No buzzer
Observation 2 Open Green LED is ON. No buzzer
Observation 3 Open (at night) Green LED is ON. No buzzer
Observation 4 Closed Red LED is ON. Buzzer raise the alarm
Observation 5 Open Green LED is ON. No buzzer
Observation 6 Closed Red LED is ON. Buzzer raise the alarm
Observation 7 Open Green LED is ON. No buzzer

5 Observations

Experiments have been carried out on different person and different time. These
indicate the high correct detection rate which is indicative of the method’s supe-
riority and high robustness. In the experimental set up two different colors of LEDs
—Red and Green are used to indicate fatigue condition (closed eyes) and normal
condition (open eyes) respectively. A buzzer is also incorporated whenever fatigue
condition is detected. The experimental results for image sequence of eye tracking
are given at Fig. 5, and observations on alarm conditions with respect to the eye
condition are also tabulated at Table 1. It may be noticed that at closed eye con-
dition the Red LED glows as well as buzzer is activated. These observations show
that the model can track eye region robustly and correctly and can avoid the
accident as well.

6 Discussions

The experimental model as developed in the laboratory is of minimum complexi-

ties. Experiments and observations have been carried out at different time, person
and environmental condition that prove its good robust performance. It is helpful in
driver vigilance and accident avoidance system. However the performance and
effectiveness of the project depends on the quality of the camera and finding out the
threshold while removing noise from the acquired picture.
In this model the alarming system is only to alert the driver. The system can be
integrated with brake and accelerator system of the vehicle. Also this alarming
system may be attached with the front end and the back end light indicator with
good audible sound to alert other drivers and passer-by on that road to minimize the
fatal rate.
24 S. Ghosh et al.

References

1. Elzohairy Y (2008) Fatal and injury fatigue-related crashes on ontario’s roads: a 5-year review.
In: Working together to understand driver fatigue: report on symposium proceedings, february
2008
2. Dingus TA, Jahns SK, Horowitz AD, Knipling R (1998) Human factors design issues for crash
avoidance systems. In: Barfield W, Dingus TA (eds) Human factors in intelligent
transportation systems. Lawrence Associates, Mahwah, pp 55–93
3. Saito H, Ishiwaka T, Sakata M, Okabayashi S (1994) Applications of driver’s line of sight to
automobiles—what can driver’s eye tell. In: Proceedings of vehicle navigation and
information systems conference, Yokohama, Japan, pp 21–26
4. Ueno H, Kaneda M, Tsukino M (1994) Development of drowsiness detection system. In:
Proceedings of vehicle navigation and information systems conference, Yokohama, Japan,
pp 15–20
5. Boverie S, Leqellec JM, Hirl A (1998) Intelligent systems for video monitoring of vehicle
cockpit. In: International Congress and exposition ITS: advanced controls and vehicle
navigation systems, pp 1–5
6. Kaneda M et al (1994) Development of a drowsiness warning system. In: The 11th
international conference on enhanced safety of vehicle, Munich
7. Onken R (1994) Daisy, an adaptive knowledge-based driver monitoring and warning system.
In: Proceedings of vehicle navigation and information systems conference, Yokohama, Japan,
pp 3–10
8. Feraric J, Kopf M, Onken R (1992) Statistical versus neural bet approach for driver behaviour
description and adaptive warning. The 11th European annual manual, pp 429–436
9. Ishii T, Hirose M, Iwata H (1987) Automatic recognition of driver’s facial expression by
image analysis. J Soc Automot Eng Jap 41:1398–1403
10. Yammamoto K, Higuchi S (1992) Development of a drowsiness warning system. J Soc
Automot Eng Jap 46:127–133
11. Smith P, Shah M, da Vitoria Lobo N (2000) Monitoring head/eye motion for driver alertness with
one camera. In: The 15th international conference on pattern recognition, vol 4, pp 636–642
12. Saito S (1992) Does fatigue exist in a quantitative of eye movement? Ergonomics 35:607–615
13. Anon (1999) Perclos and eye tracking: challenge and opportunity. Technical Report Applied
Science Laboratories, Bedford
14. Wierville WW (1994) Overview of research on driver drowsiness definition and driver
drowsiness detection. ESV, Munich
15. Dinges DF, Mallis M, Maislin G, Powell JW (1998) Evaluation of techniques for ocular
measurement as an index of fatigue and the basis for alertness management. Dept Transp
Highw Saf Publ 808:762
16. Anon (1998) Proximity array sensing system: head position monitor/metric. Advanced safety
concepts, Inc., Sante Fe, NM87504
17. Anon (1999) Conference on ocular measures of driver alertness, Washington DC, April 1999
18. Qiang J, Xiaojie Y (2002) Real-Time Eye, Gaze, and face pose tracking for monitoring driver
vigilance. Real-Time Imag 8:357–377
19. D’Orazio T, Leo M, Guaragnella C, Distante A (2007) A visual approach for driver inattention
detection. Pattern Recogn 40(8):2341–2355
20. Boyraz P, Acar M, Kerr D (2008) Multi-sensor driver drowsiness monitoring. Proceedings of
the institution of mechanical engineers, Part D: J Automobile Eng 222(11):2041–2062
21. Ebisawa Y (1989) Unconstrained pupil detection technique using two light sources and the
image difference method. Vis Intell Des Eng, pp 79–89
22. Grauman K, Betke M, Gips J, Bradski GR (2001) Communication via eye blinks: detection
and duration analysis in real time. In: Proceedings of IEEE conference on computer vision and
pattern recognition, WIT Press, pp 1010–1017
Real Time Eye Detection and Tracking … 25

23. Matsumoto Y, Zelinsky A (2000) An algorithm for real-time stereo vision Implementation of
Head pose and gaze direction measurements. In: Proceedings of IEEE 4th international
conference on face and gesture recognition, pp 499–505
24. Yuille AL, Hallinan PW, Cohen DS (1992) Feature extraction from faces using deformable
templates. Int J Comput Vis 8(2):99–111
25. Ivins JP, Porrill J (1998) A deformable model of the human iris for measuring small
3-dimensional eye movements. Mach Vis Appl 11(1):42–51
26. Kawato S, Tetsutani N (2002) Real-time detection of between-the-eyes with a circle frequency
filter. In: Asian conference on computer vision
27. Sommer G, Michaelis M, Herpers R (1998) The SVD approach for steerable filter design. In:
Proceedings of international symposium on circuits and systems 1998, Monterey, California,
vol 5, pp 349–353
28. Yang G, Waibel A (1996) A real-time face tracker. In: Workshop on applications of computer
vision, pp 142–147
29. Loy G, Zelinsky A (2003) Fast radial symmetry transform for detecting points of interest.
IEEE Trans Pattern Anal Mach Intell 25(8):959–973
Preprocessing in Early Stage Detection
of Diabetic Retinopathy Using Fundus
Images

Vijay M. Mane, D.V. Jadhav and Ramish B. Kawadiwale

Abstract Automated retinal image processing is becoming a primary important

screening tool for early detection of diabetic retinopathy (DR). An automated
system reduces human errors and also reduces the burden on the ophthalmologists.
The accurate detection of microaneurysms (MAs) is an important step for early
detection of DR. This paper present some methods to improve the quality of input
retinal image and extraction of blood vessels, as a preprocessing step in automatic
early stage detection of DR. Experimental results are performed for preprocessing
and blood vessel extraction techniques using standard fundus image database.

Keywords Preprocessing
Contrast enhancement
Segmentation
Fundus

images Blood vessels

1 Introduction

Diabetes mellitus (DM) is the name of a systemic and serious disease [1]. It occurs
when the pancreas does not produce an adequate amount of insulin or the body is
unable to process it properly. This results in an abnormal increase of the glucose
level in the blood. Eventually this high level of glucose causes damage to blood
vessels. This damage affects almost all organs like eyes, nervous system, heart,
kidneys etc. Diabetes mellitus commonly results in diabetic retinopathy (DR),

V.M. Mane (&)

JSPM’S RSCOE Tathwade, Savitribai Phule Pune University, Pune, India
e-mail: [email protected]
D.V. Jadhav
TSSM BSCOE&R, Narhe, Pune, India
e-mail: [email protected]
R.B. Kawadiwale
Department of Electronics, Vishwakarma Institute of Technology, Pune, India
e-mail: [email protected]

© Springer India 2015 27

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_3
28 V.M. Mane et al.

Fig. 1 Fundus image

showing features of diabetic
retinopathy

which is caused by pathological alteration of the blood vessels which nourish the
retina. As a result of this damage, the capillaries leak blood and fluid on the retina
[2]. DR is the main cause of new cases of blindness among adults aged 20–74 years
[3–7]. Microaneurysms, hemorrhages, exudates, cotton wool spots or venous loops
etc. can be seen as visual features on retinal images as shown in Fig. 1. Micro-
aneurysms (MAs) are a common and often early appearance of DR. So the MA
detector is an attractive candidate for an automatic screening system able to detect
early findings of DR. The main processing components for detection of MAs using
retinal fundus images include: preprocessing, selection of a candidate MA, feature
extraction and classification as shown in Fig. 2. The performance of lesion detection
algorithms solely depends on quality of retinal images that are captured by fundus
camera.
Preprocessing step is used to minimize image variations and improve image
quality. MAs do not appear in a vessel but many MA candidates can be detected in
retinal vessels. This false detection may be due to the dark red dots in the blood
vessels. To reduce false MA detection, blood vessels have to be removed to prevent

Fig. 2 Detection of MAs

Fundus Image acquisition

Image Pre-processing

Candidate MA detection

Feature Extraction

Classification/Final MA
Preprocessing in Early Stage Detection of Diabetic … 29

a misclassification. In this paper, we present comparison of retinal image prepro-

cessing techniques that enhance the fundus images by extracting Green plane of
color fundus image. Contrast enhancement using Histogram stretching, histogram
equalization and adaptive histogram equalization are implemented to enhance the
input fundus image. Blood vessels are extracted using simple thresholding, top hat
transform, K means clustering and fuzzy C means clustering methods. Paper is
organized in five sections. Sections 2 and 3 present the techniques for colored
retinal image enhancement and vessels segmentation respectively. Experimental
results of the tests on the images of the DRIVE database and their analysis are given
in Sect. 4 followed by conclusion in Sect. 5.

2 Contrast Enhancement

The main objective of pre-processing technique is to attenuate image variation by

normalizing the original retinal image against a reference model or data set for
subsequent viewing, processing or analysis. Variations typically arise within the
same image (intra-image variability) as well as between images (inter-image vari-
ability). Intra-image variation occurs due to differences in light diffusion, presence
of abnormalities, variations in fundus reflectivity and fundus thickness. Inter-image
variations arise due to differences in cameras, illumination, acquisition angle and
retinal pigmentation.

2.1 Green Plane Selection

It is observed that the green channel of color fundus images is commonly used
by unsupervised methods to detect MAs from fundus images, as it has the best
vessel—background contrast as observed in middle column of Fig. 3.

2.2 Histogram Stretching [8]

The contrast of an image is the distribution of intensity of its pixels. A low contrast
image show small differences between its light and dark pixel values. Human eyes
are sensitive to contrast rather than absolute pixel intensities. a perceptually better
image could be obtained by stretching the histogram of an image, in which dynamic
range of the image is filled. Figure 4a shows the original and Fig. 4b, c are log and
power law contrast enhanced images.
30 V.M. Mane et al.

Fig. 3 RGB colour band with respective histograms

Fig. 4 Contrast stretched images. a Input image. b LOG transformed image. c Power law
corrected
Preprocessing in Early Stage Detection of Diabetic … 31

• Take input fundus image, RGB image

• Extract Green channel component from this image
• Calculate:
1. L = C * log(1 + intensity of pixel)
2. P = C * (intensity of pixel)ɤ

2.3 Histogram Equalization [8, 9]

A histogram equalized image is obtained by mapping each pixel in the input image
to a corresponding pixel in the output image using an equation based on the
cumulative distribution function. Figure 5a, b are input image and its histogram
whereas Fig. 5c, d are histogram equalized image and its histogram.
• Take input fundus image.
• Extract Green channel component from this image.
• Get the histogram of the image.
• Find probability density function and cumulative distribution function for each
intensity level.

Fig. 5 Histogram equalized images and respective histograms. a Input image. b Histogram of
input image. c Histogram equalized image. d Histogram of equalized image
32 V.M. Mane et al.

• Calculate new intensity values using equalization equations.

• Build new image by replacing original gray values with the new gray values.

2.4 Adaptive Histogram Equalization (AHE) [9]

Adaptive method computes several histograms, each corresponding to a distinct

section of the image, and uses them to redistribute the lightness values of the image.
Figure 6a, b shows the original image and its histogram and Fig. 6c, d are image
after adaptive histogram equalization and its histogram respectively.
• Take input fundus image.
• Extract Green channel component from this image.
• Select appropriate window size, centered at a pixel called grid point.
• For each grid point calculate the histogram of the region around it, having area
equal to window size and centered at the grid point.
• Enhance the contrast of window selected and map the grid point to new image.

Fig. 6 Adaptive histogram equalized images and respective histograms. a Green plane.
b Histogram of green plane. c AHE. d Histogram of AHE.
Preprocessing in Early Stage Detection of Diabetic … 33

3 Blood Vessels Extraction

3.1 Shade Correction Followed by Thresholding [10]

A simple manually selected thresholding is applied to segment blood vessels from

background as shown in Fig. 7.
• Apply adaptive histogram equalization algorithm to improve the contrast.
• Apply median filter to contrast enhanced image with filter size greater than
blood vessel width.
• Subtract adaptive histogram equalized image from median filtered image.
• Threshold this shade corrected image to get blood vessels.

3.2 Top Hat Transform [9, 11, 12]

Top-hat transform is an operation that extracts small elements and details from
given images. The top-hat transformation applied on the filtered image with a disk
structure element with a size large enough to fill all the holes in blood vessels. The
top hat transformation is then performed by a closing operation, as shown in Fig. 8a
input image and Fig. 8b the output.
• Take shade corrected image as input.
• Define suitable structuring element.
• Apply morphological opening with structuring element on shade corrected
image.
• Manually threshold the opened image.
• Top hat transform—Subtract the threshold opened image from shade corrected
image.

Fig. 7 Output of shade correction algorithm. a Input image. b Shade corrected. c Thresholding
output
34 V.M. Mane et al.

Fig. 8 Top hat transform.

a Input image. b Output of
top hat transform

• T ð f Þ ¼ f  ðf o bÞ Where T ð f Þ is the top-hat transformation, (o) is the opening

operation, b is the structuring element and f is the shade corrected image.
• Subtract T ð f Þ pixels from shade corrected image.
• Threshold this image to get blood vessels.

3.3 K-Means Clustering [13]

K-means clustering is one of the simplest unsupervised learning algorithms. The

procedure follows a simple and easy way to classify a given data set through a
certain number of clusters fixed a priori. This algorithm aims at minimizing an
objective function:

X n 
k X 2
 j 
J¼ Xi  Cj 
j¼1 i¼0

where jjXji  Cj jj2 is a chosen distance measure between a data point Xji and the
cluster centre Cj is an indicator of the distance of the n data points from their
respective cluster centres.
• Take pre-processed image as input.
• Empirically set threshold to divide the image into blood vessels and background
clusters.
• Find cluster centers.
• Find distance of each data point from cluster center.
• Create an array whose columns are two distance vectors.
• According to the minimum distance, assign original image pixel to corre-
sponding cluster.
Preprocessing in Early Stage Detection of Diabetic … 35

3.4 Fuzzy C-Means Clustering [14]

Fuzzy clustering technique allows each of data object to belong to more than one
cluster. Membership value, which specifies degree of belongingness to each cluster,
is assigned to each data object. The membership value of each data object is
updated in each iteration. Data point may partially or fully belongs to a cluster.
• Take pre-processed image as input of size [m n] and convert into single vector
of m*n data objects.
• Set threshold empirically, so as to divide image into 2 clusters, blood vessels
and background.
• Find cluster centers “Ci” by using equation:
Pn
um
ij xj
Ci ¼ Pn
j¼1
u m
j¼1 ij

• Calculate objective function “J” to be minimized:

X
c c X
X n
J ðU; c1 ; c2 ; . . . ; cc Þ ¼ Ji ¼ um 2
ij dij
i¼1 i¼1 j¼1

• Calculate membership function “Uij”:

1
uij ¼
Pc  dij 2=ðm1Þ
k¼1 dkj

X
c
uij ¼ 1; 8j ¼ 1; 2; . . .; n
i¼1

dij = Euclidean distance between ith cluster center and jth data point
m = weighting exponent (m > 1)
• De-fuzzify the single vector to get data clusters (Fig. 9).
36 V.M. Mane et al.

Fig. 9 Results of clustering algorithm. a Input image. b K-means clustering. c Fuzzy C-means
clustering

4 Experimental Results

The performance of preprocessing methods is evaluated using publicly available

DRIVE database [15, 16]. The DRIVE database consists of 40 RGB color images
of size 768 × 584 pixels, eight bits per color channel. The image set is divided into a
test and training set and each one contains 20 images. There are two hand-labelling
available for the 20 images of test set made by two different human observers.

4.1 Evaluation of Contrast Enhancement Algorithms

The performance of preprocessing algorithms is evaluated using peak signal to

noise ratio (PSNR) and mean square error (MSE). Ideally, PSNR should be high
and MSE should be low. The presented algorithms for contrast enhancement of
fundus images shows promising results. Adaptive histogram equalization shows a
good contrasted fundus image as seen from Table 1.

Table 1 PSNR and MSE of drive images

Algorithm TEST_10 image TEST_19 image
MSE PSNR MSE PSNR
Contrast stretching 58.5081 30.4586 58.801 30.436
Histogram equalization 42.374 31.859 42.371 31.8600
Adaptive histogram equalization 38.561 32.269 38.259 32.3041
Preprocessing in Early Stage Detection of Diabetic … 37

Table 2 Comparison of blood vessel extraction algorithms

Algorithm TEST_10 image (%) TEST_19 image (%)
Shade correction Sensitivity 65.9 78.5
Specificity 98.33 97.75
Accuracy 95.77 96.15
Top hat transform Sensitivity 63.45 76.83
Specificity 98.3 98.31
Accuracy 95.3 96.53
K-means clustering Sensitivity 70.04 83.02
Specificity 97.38 97.67
Accuracy 94.9 96.45
FCM clustering Sensitivity 54.77 71.5
Specificity 99.4 99.47
Accuracy 95.72 97.15

4.2 Evaluation of Blood Vessels Detection Algorithm

Blood vessels detection algorithms are evaluated in terms of their sensitivity,

specificity and accuracy. It is observed that given a contrast enhanced fundus image
to different blood vessel extraction methods, each gives a promising output in terms
of accuracy. These parameters are evaluated on two images from database as shown
in Table 2. K means clustering and FCM clustering shows good results.

5 Conclusions

While capturing the retinal image using fundus camera, the retina is not illuminated
uniformly because of circular shape of retina. The preprocessing stage in early stage
detection of MAs is necessary to correct the non-uniform illumination and to
enhance the contrast. This paper presents preprocessing step for fundus image
contrast enhancement and blood vessel extraction. Operations like Histogram
stretching, histogram equalization and adaptive histogram equalization are imple-
mented to enhance the contrast of input fundus images. Adaptive histogram
equalization gives better contrast enhanced image which is evaluated in terms of
MSE and PSNR. Blood vessels are extracted using simple thresholding, top hat
transform, K means clustering and fuzzy C means clustering algorithms on retinal
images of publicly available DRIVE database. Blood vessels detection algorithms
are evaluated in terms of their sensitivity, specificity and accuracy. K means
clustering and FCM clustering outperform in terms of accuracy than other methods
presented in this paper. Segmentation of retinal blood vessels is a challenging task
mainly because of the presence of a wide variety of vessel widths, low contrast and
the vessel color is close to that of the background.
38 V.M. Mane et al.

References

1. Chiulla TA, Amador AG, Zinman B (2003) Diabetic retinopathy and diabetic macular edema:
pathophysiology, screening, and novel therapies. Diabetes Care 26(9):2653–2664
2. Frank RN (1995) Diabetic retinopathy. Prog Retin Eye Res 14(2):361–392
3. Klein R, Klein BEK, Moss SE (1994) Visual impairment in diabetes. Ophthalmology 91:1–9
4. Klonoff DC, Schwartz DM (2000) An economic analysis of interventions for diabetes.
Diabetes Care 23(3):390–404
5. Center for Disease Control and Prevention (2011) National diabetes fact sheet: technical
report, U.S.
6. Bresnick GH, Mukamel DB, Dickinson JC, Cole DR (2000) A screening approach to the
surveillance of patients with diabetes for the presence of vision-threatening retinopathy.
Opthalmology 107(1):19–24
7. Susman EJ, Tsiaras WJ, Soper KA (1982) Diagnosis of diabetic eye disease. J Am Med Assoc
247(23):3231–3234
8. Hatanaka Y, Inoue T, Okumura S, Muramatsu C, Fujita S (2012) Automated microaneurysm
detection method based on double-ring filter and feature analysis in retinal fundus images. In:
Proceedings of 25th IEEE international symposium on computer-based medical systems,
paper-150
9. Saleh MD, Eswaran C (2012) An automated decision-support system for non-proliferative
diabetic retinopathy disease based on MAs and HAs detection. Elsevier—Comput Meth
Programs Biomed 108:186–196
10. Marín D, Aquino A, Gegúndez-Arias ME, Bravo JM (2011) A new supervised method for
blood vessel segmentation in retinal images by using gray-level and moment invariants-based
features. IEEE Trans Med Imaging 30(1):146–158
11. El Abbadi NK, Al Saadi EH (2013) Blood vessels extraction using mathematical morphology.
J Comput Sci 9(10):1389–1395
12. Ram K, Joshi GD, Sivaswamy J (2011) A successive clutter-rejection-based approach for early
detection of diabetic retinopathy. IEEE Trans Biomed Eng 58(3)
13. Masroor AM, Mohammad DB (2008) Segmentation of brain MR images for tumor extraction
by combining K means clustering and Perona-Malik anisotropic diffusion model. Int J Image
Proc 2(1)
14. Dey N, Roy AB, Pal M, Das A (2012) FCM based blood vessel segmentation method for
retinal images. Int J Comput Sci Netw (IJCSN) 1(3). ISSN:2277-5420
15. Staal JJ, Abramoff MD, Niemeijer M, Viergever MA, van Ginneken B (2004) Ridge based
vessel segmentation in color images of the retina. IEEE Trans Med Imaging 23:501–509
16. Image Sciences Institute (2001) DRIVE: digital retinal images for vessel extraction.
http://www.isi.uu.nl/Research/Databases/DRIVE
Magnetic Resonance Image Quality
Enhancement Using Transform Based
Hybrid Filtering

Manas K. Nag, Subhranil Koley, Chandan Chakraborty

and Anup Kumar Sadhu

Abstract This paper proposes a novel methodology for improving the quality of
magnetic resonance image (MRI). The presence of noise affects the image analysis
task by degrading the visual contents of image. The proposed methodology inte-
grates transform domain method, discrete wavelet transform with spatial domain
filter, Non local means to smoothed out noisy interferences leading to the
improvement of visual characteristics of MRI insights. The quantitative validation
of the proposed technique has been done and experimental result shows the
effectiveness of this algorithm over anisotropic diffusion, bilateral, trilateral and
wavelet shrinkage filters.

Keywords MRI
Wavelet transform
Non-local means filter
Quality metrics

1 Introduction

MRI is a non invasive radiation free imaging technology for capturing the visual
information of internal body tissues, which provides aid in diagnosing variety of
abnormalities to clinicians. It has been well appreciated in doctor’s community that
magnetic resonance (MR) images are well enough to produce necessary features in
brain tumor diagnosis. It uses a powerful magnetic field, radio frequency waves (RF
waves) and a computer to produce image of internal structure of body. It has been
observed that MR signal fluctuates in random manner because of the presence of
thermal noise. Moreover, impulse noise is very common in medical images, which
appears during the acquisition or transmission of image through channel.

M.K. Nag
S. Koley
C. Chakraborty (&)

School of Medical Science and Technology, Indian Institute of Technology Kharagpur,
Kharagpur 721302, India
e-mail: [email protected]
A.K. Sadhu
EKO CT & MRI Scan Centre Medical College and Hospitals Campus, Kolkata 700073, India

© Springer India 2015 39

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_4
40 M.K. Nag et al.

The appearance of noise severely affects the visual features which are the key
markers of disease recognition from the image. In reality, radiologist does the
analysis on the basis of his ability in investigating visual imaging features apart
from case study. Therefore, the noisy artifact often misleads them to reach the
accurate diagnosis. In continuation, it also makes the next level image processing
tasks i.e. segmentation, feature extraction etc. difficult. It is well known that in case
of teleradiology; where MRI data can be transmitted from one place to another
require high end network set up especially for transmitting images in DICOM
format. The establishment of this kind of infrastructure is costly and not available in
remote area. Under such situation transmission of JPEG images becomes easy and
does not require any extra facility other than internet and hence cost effective. In
such case, noise may appear. After concerning all the issues, this study focuses in
reducing the noisy impact from MR images at the preprocessing stage of an
automated computer assistive brain tumor screening system.
In this section, we address the previous works in reducing the effect of noise
from MRI. Several denoising methods have been developed to enhance the quality
of image. Perona and Malik proposed a new filtering approach based on heat
equation to overcome the drawbacks of spatial filtering by emphasizing on pre-
serving the edge information and this filter is quiet effective for homogeneous
region [1]. Later on Krissian and Aja-Fernandez modified anisotropic diffusion filter
(ADF) and tested over MR images to eliminate rician noise [2]. On the other hand
Tomasi and Manduchi extended the concept of domain filtering through intro-
ducing additional range information. Hence bilateral filter (BLF) [3] adds value in
smoothing along with keeping contour intact. Moreover its non-iterative policy
gives privilege over anisotropic diffusion. Later on Wong et al. [4] introduced a
novel methodology, trilateral filtering (TLF), as an extension of BLF for sup-
pressing noise from medical images. Apart from photometric and geometric simi-
larity of BLF, it makes an addition of structural information. Low pass filter is
employed in homogeneous region, whereas pixel belongs to heterogeneous region
is replaced by the weighted average of those three similarity indexes. They showed
the improved performance over BLF through preserving edges while smoothing.
Transform based technique such as wavelet transform (WT) is popular approach in
image denoising through keeping the characteristics intact [5, 6]. Nowak [7] pro-
posed the wavelet domain filter for denoising MR images where noise follows
rician distribution properties. Baudes et al. [8] developed non local means (NLM)
filter and showed its ability in preserving image structure. After some years,
Manjón et al. [9] introduced the application of NLM filter and also proposed
unbiased NLM for dealing with the rician noise distribution in the magnitude MR
image. They performed the parameterization task of filter for various noise levels.
Later on, Manjon et al. [10] have also developed adaptive NLM for denoising MR
images when noise level in image is spatially varying. In the next year, Erturk et al.
[11] proposed a novel denoising technique based on spectral subtraction for
improving signal-to-noise ratio (SNR).
In this paper, we have introduced a hybrid filtering technique as a combination
of discrete WT and NLM filter.
Magnetic Resonance Image Quality Enhancement … 41

The transform coefficients obtained from discrete WT would have been under-
gone through NLM filter for modification and finally filtered image can be
reconstructed from inverse transformation. This experiment has been performed
without addition (assume presence of noise) and with addition of noise. The pro-
posed algorithm shows its robustness and effectiveness over ADF, BLF, TLF and
wavelet shrinkage (WS) in both cases for enhancing the quality of MR images.

2 Materials and Method

The step wise schematic of proposed filter is shown in Fig. 1. In this section we
discuss the type of MR images and the algorithms employed in this study.

2.1 MR Imaging

The MR imaging were done with 1.5 Tesla MRI scanner and acquired in Digital
Imaging and Communications in Medicine (DICOM) format. The resolution, slice
thickness and flip angle were kept as 512 × 512, 5 mm and 90° respectively. All
total 50 axial slices of MRI from 9 cases of brain tumor have been considered in
this study. The DICOM images are compressed in .jpg (Joint photographic experts
group) for processing. Moreover, the background i.e. non brain part of each indi-
vidual has been reduced through cropping and the resulting image would be con-
sidered for filtering.

2.2 Wavelet Decomposition

Discrete WT is a multiresolution decomposition framework of 1D or 2D signal i.e.

image into low (approximation coefficients) and high frequency components (detail
coefficients) [5]. The high frequency components are precious in nature as they

Fig. 1 The step by step representation of proposed methodology for filtering brain MR images
42 M.K. Nag et al.

carry the edge information. WT is capable of representing edge information in

several directions viz. horizontal, vertical and diagonal. The major advantage of
WT over fourier transform is that this is localised in spatial as well as frequency
domain, whereas fourier is restricted in frequency domain only. The coefficients of
wavelet decomposition provide information which are independent of original
image resolution. In the context of this research, single level wavelet decomposition
has been adopted to generate the coefficients from which approximation coefficients
will be considered for filtering task, whereas detail coefficients will be kept intact.

2.3 Non Local Means

NLM filter replaces the pixel value being filtered with the weighted average of
pixels within a specified region. The weights are computed from the region based
comparison instead of pixel based comparison and this makes it different and
effective from bilateral filtering and also capable of removing thermal noise effi-
ciently by reducing the variation of signal intensity in a region [9].
According to this approach, the filtered value (F) of a pixel i is obtained from
X
F ðI ðiÞÞ ¼ wði; jÞI ð jÞ j ¼ 1; 2; . . .; n: ð1Þ
8j2I

Here, I is an input MR image containing n number of pixels and wði; jÞ denotes the
weight between ith and jth pixel, which has following properties as mentioned below
X
0  wði; jÞ  1 and wði; jÞ ¼ 1
8j2I

The two square neighborhoods Si and Sj centered at ith and jth pixel respectively
are employed to measure the neighborhood based similarity between these two pixels
from which the normalized weight between them can be computed as defined below
   
Dði; jÞ X Dði; jÞ
wði; jÞ ¼ exp  2 = exp  2 ; ð2Þ
h 8j
h

where h is a decaying parameter which is exponential in nature and controls the

amount of smoothing. The Euclidean distance ðDÞ weighted by Gaussian as a
similarity measure between the two neighborhoods is defined as
 2
Dði; jÞ ¼ Gi IðSi Þ  IðSj ÞRs : ð3Þ

In the above expression, Rs represents radius of neighborhood and the normal-

ized Gaussian weighting function is denoted by Gi . The exponential decay
Magnetic Resonance Image Quality Enhancement … 43

parameter can be computed from the standard deviation r of the noise which can be
estimated from the background of the image as
rffiffiffi
l
r ¼ ð4Þ
2

where l is the mean of background extracted by Otsu thresholding [9, 12]. A

special condition in weight calculation is appeared when i ¼ j and under such
situation it is computed as shown below

wði; iÞ ¼ maxðwði; jÞ8j 6¼ iÞ: ð5Þ

The actual process of computing weights by considering all pixels in an image

increases time complexity and makes the technique inefficient. Therefore, a search
window with radius Rsearch can be employed for weight calculation. As per the
suggestion of Manjón et al. [9], a search window of dimension 11 × 11 and a 5 × 5
window for similarity measure along with h ¼ 1:2  r have been considered in this
study to suppress the effect of noise from MR signals.

2.4 Performance Evaluation Metrics

In this study, Mean squared error (MSE) and Peak signal to noise ratio (PSNR) are
considered as quality metrics for the quantitative analysis of filter performance in
terms of image quality [10, 13]. MSE is the accumulative squared error between the
input ðIÞ and filtered image ðFÞ of dimension M  N as expressed below

1 XM X N
MSE ¼ ½Iðx; yÞ  Fðx; yÞ2 : ð6Þ
MN y¼1 x¼1

PSNR is the ratio of the maximum power of the original image to the filtered
image, which is derived from MSE as
 
maxðmaxðIÞ; maxðFÞÞ
PSNR ¼ 20  log10 pffiffiffiffiffiffiffiffiffiffi : ð7Þ
MSE

3 Results and Discussion

The proposed algorithm has been designed in such a way that the major charac-
teristics or features of MR image of brain tumor should be kept intact. The
boundary that separates tumor from neighboring tissues holds the most significant
44 M.K. Nag et al.

characteristics. Blurring due to the smoothing effect of filter reduces the significance
of contour of tumor and affects the analysis task. Hence edge preserving should be
taken into account during the design of filter process. Spatial domain filters are
inefficient of preserving the original characteristics of edges due to the smoothing
effect. Therefore in the proposed framework discrete WT is employed and the low
frequency components fed to NLM for updating through the weighted average of
coefficients under a certain region, while preserving the high frequencies. The
computation of weights plays the pivotal role in NLM. The weights between two
coefficients are derived from the similarity measure and this task is accomplished by
employing two neighbourhoods centered at both coefficients. In the last stage, the
inverse wavelet transformation is applied on modified approximation and preserved
detail coefficients to obtain the final outcome.
We have tested the filtering techniques directly on 50 axial MR images in two
ways; without adding noise and adding external noise on original images. Figure 2
represents the results of proposed methodology along with ADF, BLF, TLF and
WS (Haar wavelet is used with second level decomposition) [14] filter for three
different images. In this case filters have been directly applied to original MR
images. On the other hand, Fig. 3 shows the results of filters when Gaussian noise
with variance 0.01 has been added. The two different quality measures are
employed for quantitative validation of filtering techniques over 50 images. The
lower value of MSE is desirable because higher value signifies large difference
between input and filter image leading to the changes in image characteristics. In
contrast, large PSNR is required as it signifies the comprehensive signal-to-noise

Fig. 2 Filtering outputs of axial MR images of brain tumor, a original MR image, b output of
ADF, c BLF, d TLF, e WS and f proposed method
Magnetic Resonance Image Quality Enhancement … 45

Fig. 3 Filtered outputs of MR image when noise is externally added, a original MR image,
b image after noise addition, c output of ADF, d BLF, e TLF, f WS and g proposed methodology

ratio. The quantitative assessment of different filtering techniques is presented in

Table 1 in terms of mean and standard deviation. From this table we clearly
understand that proposed filtering methodology provides minimum MSE and
maximum PSNR for both cases i.e. without and with adding noise as compare to
ADF, BLF, TLF and WS methods. This evaluation justifies the superiority of
proposed method.
We have plotted the mean and standard deviation of both MSE and PSNR for
proposed method along with four other techniques as presented in Table 1 through
the error bar representation. This graphical presentation has been executed for both
cases i.e. with and without noise for better understanding and analysis of the results
as shown in Fig. 4.

Table 1 Performance evaluation of proposed methodology, ADF, BLF, TLF and WS for
reducing noise from MR images of brain tumor
Filter With noise Without noise
MSE PSNR MSE PSNR
ADF 299.23 ± 12.28 23.36 ± 0.19 35.79 ± 2.47 32.61 ± 0.28
BLF 299.23 ± 14.83 23.36 ± 0.21 69.19 ± 8.32 30.00 ± 0.55
TLF 415.21 ± 13.68 21.94 ± 0.14 48.81 ± 7.3 31.37 ± 0.56
WS 480.54 ± 0.46 20.40 ± 0.08 36.25 ± 0.32 32.16 ± 0.22
Proposed 97.30 ± 0.96 28.24 ± 0.04 27.15 ± 2.25 34.06 ± 0.36
46 M.K. Nag et al.

Fig. 4 Error bar representation of ADF, BLF, TLF, WS and proposed method with respect to
a MSE when external noise is added, b MSE with no external noise, c PSNR when external noise
is added and d PSNR with no external noise

To justify the superiority of proposed algorithm over the existing filters, the
performance (in terms of MSE and PSNR) of all five filtering methods have been
statistically analysed by Fisher’s-F statistic. The performance of this statistical
analysis has been reported in Table 2 by showing F value and p-value [15]. In
Table 2, it is observed that the p-value is less than 0.05 for MSE and PSNR in both
cases (with and without noise). Therefore the results obtained from this analysis
signify that the mean MSE and PSNR of all five filters are different (in both cases)
and hence the quantitative assessment of filter performances with respect to the
quality metrics is statistically significant. After observing the results of proposed
and other four techniques, it is well understood that our methodology reduces the
impact of noise through preserving edges and avoiding the smoothing effect.
Therefore, the internal characteristics are being intact. The overall performance

Table 2 Analysis of variance

One way With noise Without noise
for MSE and PSNR of all five
ANOVA
filters for both the cases (with MSE PSNR MSE PSNR
and without noise) F value 188.900 40.053 67.898 36.274
p-value 0.000 0.000 0.000 0.000
p-value < 0.05 indicates statistical significance
Magnetic Resonance Image Quality Enhancement … 47

analysis also supports this statement. But one thing we observe that, when filtering
is performed on externally added noisy image, the image characteristics change in
output, although proposed method provides efficient result than others. From the
view point of visual interpretation and quantitative evaluation, the WT-NLM based
hybrid filtering approach outperforms ADF, BLF, TLF and WS for improving the
image quality of brain MR images of tumor.

4 Conclusion

A novel noise reduction technique combining transform domain method with

spatial domain filter to enhance the image quality of MRI is presented in this paper.
The proposed method proofs its effectiveness in suppressing noise in both
approaches i.e. with and without adding external noise over four different filters.
From the analysis of visual perception and performance evaluation, altogether we
conclude that our methodology is a reasonable approach in reducing noise from MR
images of brain tumor.

Acknowledgments The authors would like to acknowledge EKO CT & MRI Scan Centre at
Medical College and Hospitals Campus, Kolkata-700073 for providing brain MR images. Authors
would like to acknowledge Board of Research in Nuclear Sciences (BRNS), Dept. of Atomic
Energy for financially supporting the research work under the grant number 2013/36/38-BRNS/
2350 dt. 25-11-2013.

References

1. Perona P, Malik J (1990) Scale-space and edge detection using anisotropic diffusion. IEEE
Trans Pattern Anal Mach Intell 12(7):629–639
2. Krissian K, Aja-Fernández S (2009) Noise-driven anisotropic diffusion filtering of MRI. IEEE
Trans Image Process 18(10):2265–2274
3. Tomasi C, Manduchi R (1998) Bilateral filtering for gray and color images. In: Proceedings of
the sixth international conference on computer vision, pp 839–846
4. Wong WCK, Chung ACS, Yu SCH (2004) Trilateral filtering for biomedical images. IEEE Int
Symp Biomed Imaging: Nano to Macro 1:820–823
5. Xu Y, Weaver JB, Healy DM, Lu J (1994) Wavelet transform domain filters: a spatially
selective noise filtration technique. IEEE Trans Image Process 3(6):747–758
6. Wood JC, Johnson KM (1999) Wavelet packet denoising of magnetic resonance images:
importance of Rician noise at low SNR. Magn Reson Med 41(3):631–635
7. Nowak RD (1999) Wavelet-based Rician noise removal for magnetic resonance imaging.
IEEE Trans Image Process 8(10):1408–1419
8. Buades A, Coll B, Morel J-M (2005) A review of image denoising algorithms, with a new one.
Multiscale Model Simul 4(2):490–530
9. Manjón JV, Caballero JC, Lull JJ, Marti GG, Bonmati LM, Robles M (2008) MRI denoising
using non-local means. Med Image Anal 12(4):514–523
48 M.K. Nag et al.

10. Manjon JV, Coupe P, Bonmati LM, Collins DL, Robles M (2010) Adaptive non-local means
denoising of MR images with spatially varying noise levels. J Magn Reson Imaging 31
(1):192–203
11. Erturk MA, Bottomley PA, El-Sharkawy AM (2013) Denoising MRI using spectral
subtraction. IEEE Trans Biomed Eng 60(6):1556–1562
12. Ng H-F (2006) Automatic thresholding for defect detection. Pattern Recogn Lett 27(14):1644–
1649
13. Starck JL, Candès EJ, Donoho DL (2002) The curvelet transform for image denoising. IEEE
Trans Image Process 11(6):670–684
14. Balster EJ, Zheng YF, Ewing RL (2005) Feature-based wavelet shrinkage algorithm for image
denoising. IEEE Trans Image Process 14(12):2024–2039
15. Zijdenbos A, Forghani R, Evans AC (2002) Automatic “pipeline” analysis of 3-D MRI data for
clinical trials: application to multiple sclerosis. IEEE Trans Med Imaging 21(10):1280–1291
Histogram Based Thresholding
for Automated Nucleus Segmentation
Using Breast Imprint Cytology

Monjoy Saha, Sanjit Agarwal, Indu Arun, Rosina Ahmed,

Sanjoy Chatterjee, Pabitra Mitra and Chandan Chakraborty

Abstract Breast imprint cytology is a well-recognized technique and provides a

magnificent cytological clarity. For imprint cytology slide preparation, tissue
samples from the needle taken out to touch and rolled over glass slide and finally
stained by hematoxilin and eosin (H&E). The aim of this research is to segment
breast imprint cytology nucleus. Images from imprint cytology slides were grabbed
by an optical microscope. The histogram based threshold technique has been used
to segment nucleus. The proposed technique includes pre-processing, segmentation,
post-processing, and final output stage. In pre-processing first image colors was
normalized by white balance technique. Then the green channel was extracted from
the normalized image. In segmentation stage target nucleus was segmented by pixel
intensities. Post-processing stage refers to the clear border nucleus or sharpening
the edges. Finally, the three channels were concatenated to get RGB image. The
proposed technique performs best in imprint cytology nucleus segmentation, and
capable of distinguishing nucleus and non-nucleus objects. Performance of our
proposed algorithm is quite high and much more useful for further analysis.

Keywords Hematoxilin Eosin Histopathology Imprint cytology Morphology




Intensity Nucleus

M. Saha
C. Chakraborty (&)

School of Medical Science and Technology, Indian Institute of Technology,
Kharagpur 721 302, India
e-mail: [email protected]
S. Agarwal
I. Arun
R. Ahmed
S. Chatterjee
Tata Medical Center, New Town, Rajarhat, Kolkata 700 156, India
P. Mitra
Computer Science and Engineering, Indian Institute of Technology,
Kharagpur 721 302, India

© Springer India 2015 49

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_5
50 M. Saha et al.

1 Introduction

Microscopic observation for cancer detection is one of the most preferred tech-
niques. By microscopic visual inspection pathologist conclude grading and severity
of cancer. The most prominent and effective example of microscopic image analysis
is in the breast cancer grading system. Breast cancer considered a very common
cancer among women [1–4].
Core biopsy confirmation of cancers is often time-consuming and entails biopsy,
fixation, staining with hematoxilin and eosin (H & E) before visual examination
under the microscope to report the diagnosis and characterize (grading, type) the
cancer. This leads to increase patient discomfort apart from adding to the diagnostic
delay. So, it is required to develop automated early screening protocol especially in
rural areas to provide an efficient diagnostic prediction.
In view of this, breast imprint cytology is a well-recognized simple technique
and provides an excellent cytological clarity [5, 6]. During the last few years,
researchers investigated imprint cytology to screen cancer cells. This cytological
approach involves taking an imprint of the core biopsy specimen, based on such
imprints a result can be predicted instantly before detailed characterization in the
core biopsy report. Trained technician takes breast imprint easily, and it does not
cause distortion of the biopsy specimen architecture. This process could allow one
needle biopsy that gives an instant report of cancer versus no-cancer as well as
tumor characterization (as reported on core biopsy). The main challenge here lies in
developing robust and efficient image processing algorithms for automated char-
acterization of breast cancer nucleus cells from imprint cytological images (grabbed
by optical microscope); because, there is a higher chance of having imprecision and
ambiguities in these cytology images. Nuclei are the important component in breast
imprint cytology. Here, we proposed to develop automated breast imprint cytology
nucleus detection by intensity based image segmentation technique.
This paper is divided into four sections. First, section presents an introduction of
breast imprint cytology and cancer. The second section describes materials and
methods. Third section describes experimental results. Fourth section presents the
conclusion.

2 Materials and Methods

2.1 Breast Imprint Cytology Slide Preparation and Image

Acquisition

Ethical rules and patient’s consent has been taken from every patient in our
research. All the breast imprint cytology slides were prepared and maintained at
Tata Medical Center, Kolkata by the approved cytologists and cytopathologists.
Local anesthesia and 0.5 % epinephrine are injected around the lesions. Lidocaine
Histogram Based Thresholding for Automated Nucleus … 51

hydrochloride is mixed along with 0.5 % epinephrine during anesthesia [9]. Tissue
samples from the needle taken out to touch and rolled over glass slide. Then the
tissue samples are fixed with 95 % ethanol [9]. Finally, H&E stain is used to stain
the slides.
Breast imprints cytology images were grabbed in Leica DM750 microscope with
Leica ICC50 HD camera, from Tata Medical Center, Kolkata and BMI Lab, School
of Medical Science and Technology, Indian Institute of Technology, Kharagpur,
West Bengal.

2.2 Proposed Methodology

The proposed methodology divided into four parts i.e., image pre-processing,
segmentation, post-processing and final output. The method is summarized in Fig. 1
using the flow diagram, and Fig. 2 describes resulted image of each step of Fig. 1.
All the images were grabbed at a constant brightness, and contrast. For proper
visibility and enhancement of the target (nucleus) region pre-processing step is
applied. The preprocessing step is sub-divided into white balance adjustment, G-
channel extraction, and image intensity adjustment.
Before going to further analysis of breast imprint cytology images, it is neces-
sary to reduce the different color response error in the image. These types of color
response error may occur due to microscopic light. Even some digital microscopic
cameras sometimes give inconsistent outputs. So, white balance adjustment is
necessary to normalize the unrealistic colors present in the image. Color correction
model of white balance can be represented by the following matrix equation [7].
2 3 2 32 3
Rco Rref =Rmeg 0 0 Ro
4 Gco 5 ¼ 4 0 Gref =Gmeg 0 5:4 Go 5
Bco 0 0 Bref =Bmeg Bo

Here, Ro , Go , and Bo corrected color coordinates Rref , Gref , and Bref reference color
coordinates, Rmeg , Gmeg , and Bmeg measured coordinates and Rco , Gco , Bco corrected
color output to a standard white illuminant. Breast imprint cytology white balance
adjusted images were then separated into red (R), green (G), and blue (B) intensity
levels. Our proposed methodology has been used for analyzing the images.

Fig. 1 Schematic of the proposed breast imprint cytology malignant nucleus segmentation
technique
52 M. Saha et al.

Fig. 2 Steps of Breast imprint cytology malignant nucleus segmentation. a Original breast imprint
cytology image at 100X; b white balance adjustment; c G-Channel image; d image intensity
adjustment; e histogram; f segmented nucleus by proposed method; g morphological operation
image; h clear border image; i masked RGB image

We tested all images in RGB, HSV, L*a*b etc. color levels. Only green (G)
channel images showed good results. G-channel extracted images have high con-
trast between nucleus and background. The next step is an image intensity
adjustment. Image intensity adjustment maps the image intensity into 0–255 range
and increases the contrast of the image.
Furthermore, intensity adjusted images go to the next segmentation process.
Segmentation process is sub-divided into the histogram profile, intensity based
image segmentation, and morphological operations. A histogram is a graphical
representation of intensity and pixel count data. Here in the proposed method whole
image is marked by gray level color intensities. Using histogram multi-threshold
technique a threshold value is generated. The threshold range varies between 0–255
and it is automatically detected through our algorithm. This threshold values sep-
arate nucleus from the background. Mathematically, we can define the threshold as:
Histogram Based Thresholding for Automated Nucleus … 53

If f ði; jÞ  Ta then f ði; jÞ ¼ 255

else if Ta \f ði; jÞ\Tb then f ði; jÞ ¼ 128
else f ði; jÞ ¼ 0
Here, Ta and Tb are threshold values, and ‘i’ and ‘j’ are co-ordinates. In the
morphological operation image erode and dilate concept used. Erosion and dilation
do with a structuring element. The grayscale erosion is [8]:

ðp H qÞðr; sÞ ¼ minfpðr þ r 0 ; s þ s0 Þ  qðr 0 ; s0 Þjðr 0 ; s0 Þ 2 Vq g

Here, ðp H qÞ denotes binary erosion of p and q. Structuring element q’s domain

is Vq . pðr; sÞ ¼ þ a.
If, qðr; sÞ ¼ 0, we’ll get flat structuring element.
Now the gray scale erosion will be [8]:

ðp H qÞðr; sÞ ¼ minfpðr þ r0 ; s þ s0 Þ  jðr 0 ; s0 Þ 2 Vq g

For dilation equation will be [8]:

qðr; sÞ = 0;

ðp  qÞðr; sÞ ¼ minfpðr þ r0 ; s þ s0 Þ  jðr 0 ; s0 Þ 2 Vq g

In post-processing stage border image has been cleared. This stage is required
because if a nucleus is connected at the border, it’s very tough to get full mor-
phological information of nucleus. It may give incorrect information in nucleus
counting. Nucleus edges have been shaped by image sharpening algorithm. Finally,
three channels have been concatenated to get final RGB nucleus.

3 Results and Discussion

The proposed algorithm is tested on breast imprint cytology images. Figure 3,

shows the result of three R, G, B channels. From the output images clearly R and G
channel images are much more visible than B channel image. Again, between R and
G images, G channel image has much more contrast and brightness than R channel
image. So, G channel image extraction is very much suitable for our analysis.
Figure 4, shows the original image and ground truth image where the nucleus
has been marked by specialists at Tata Medical center using red color.
In Fig. 5, pre-processing, segmentation, and final output results have shown in
three different breast imprint cytology images. Table 1, describes the comparison
table of manual and automated nucleus count of total 9 frames.
54 M. Saha et al.

Fig. 3 Contrast visualization

in RGB space. a Original
breast imprint cytology image
at 100X; b R-channel image;
c G-channel image;
d B-channel image

Fig. 4 Ground truth RGB

image. a Original breast
imprint cytology image at
100X; b ground truth RGB
image (red circle)
Histogram Based Thresholding for Automated Nucleus … 55

Fig. 5 Results of the proposed methodology. a Original cytology image at 100X; b pre-processed
image; c segmented binary image; d final segmented nucleus

Table 1 Comparison table of

nucleus count Image Manual nucleus Automated nucleus
index count count
1 5 5
2 11 11
3 6 6
4 6 7
5 4 4
6 10 10
7 9 9
8 5 5
9 6 6

Figure 6, shows a regression curve between manual and automated nucleus

count. Here, almost all nucleus segmented normally.
The graph clearly shows that the proposed method is perfectly segmenting
nucleus as machine generated nucleus count provides exact equivalent results in
parallel to the ground truth, counted manually.
56 M. Saha et al.

Fig. 6 The graph shows the

strength of the proposed
method for automated nucleus
segmentation in comparison
with ground truth

4 Conclusion

Nucleus identification is important in breast cancer detection in breast imprint

cytology. It is concluded that this algorithm performs very well on breast imprint
cytology images for nucleus segmentation. We have successfully eliminated other
background objects except nucleus from the images. Our proposed method gives
higher count accuracy (97 %) result in nucleus detection. Further research will be
focused on the breast imprint overlapped or connected nucleus.

Acknowledgments First author acknowledges the Department of Science and Technology

(DST), Govt. of India, under INSPIRE fellowship.
All other authors acknowledge the Ministry of Human Resource Development (MHRD), Govt.
of India, for financial support under grant no: 4-23/2014 -T.S.I. date: 14-02-2014.

References

1. Niwas SI, Palanisamy P, Sujathan K, Bengtsson E (2013) Analysis of nuclei textures of fine
needle aspirated cytology images for breast cancer diagnosis using complex Daubechies
wavelets. Sig Process 93:2828–2837
2. Kowal M, Korbicz J (2010) Segmentation of breast cancer fine needle biopsy cytological
images using Fuzzy clustering. In Koronacki J, Raś Z, Wierzchoń S, Kacprzyk J (eds)
Advances in machine learning I, vol 262. Springer, Berlin, pp 405–417
3. Kamangar F, Dores GM, Anderson WF (2006) Patterns of cancer incidence, mortality, and
prevalence across five continents: defining priorities to reduce cancer disparities in different
geographic regions of the world. J Clin Oncol 24:2137–2150
4. Niwas SI, Palanisamy P, Sujathan K (2010) Wavelet based feature extraction method for breast
cancer cytology images. In: IEEE symposium on industrial electronics & applications (ISIEA),
2010. pp 686–690
5. Suen K, Wood W, Syed A, Quenville N, Clement P (1978) Role of imprint cytology in
intraoperative diagnosis: value and limitations. J Clin Pathol 31:328–337
Histogram Based Thresholding for Automated Nucleus … 57

6. Bell Z, Cameron I, Dace JS (2010) Imprint cytology predicts axillary node status in sentinel
lymph node biopsy. Ulster Med J 79:119–122
7. Wannous H, Lucas Y, Treuillet S, Mansouri A, Voisin Y (2012) Improving color correction
across camera and illumination changes by contextual sample selection. J Electron Imaging
21:023015-1–023015-14
8. Gonzalez RC, Woods RE, Eddins SL (2004) Digital image processing using MATLAB.
Prentice Hall, Upper Saddle River
9. Kashiwagi S, Onoda N, Asano Y, Noda S, Kawajiri H, Takashima T et al (2013) Adjunctive
imprint cytology of core needle biopsy specimens improved diagnostic accuracy for breast
cancer. SpringerPlus 2:1–7
Separation of Touching and Overlapped
Human Chromosome Images

V. Sri Balaji and S. Vidhya

Abstract Chromosomes are generally thread-like structures present in the nucleus of

each living cells. There are twenty three pairs of chromosome in human beings. The
additional in chromosome number or missing chromosome will cause chromosome
abnormality i.e. chromosome anomaly in human beings. This is mainly occurring due
to accident or error while sperm or egg progress in growing. The abnormality in
chromosome will cause birth defects, genetic disorders and cancer to the human
beings. The twenty three pairs of chromosome can be classified into twenty four
different classes by using karyotyping process. In this paper, the main objective is to
give idea about how to diagnose the genetic disorder. The genetic disorder can lead to
cancer in human. It may happen due to touching or overlapped chromosomes in
human beings. In order to overcome these genetic disorders and cancer in human, the
touching and overlapped chromosomes were separated. This separation process will
be easier to undergo karyotyping analysis and to be handled by a cytogeneticist.

Keywords Nucleus
Chromosome abnormality
Birth defects
Cancer

Karyotyping

1 Introduction

Chromosomes are generally thread-like structures found in the nucleus of each

living cells. It carries genetic information [1]. They cannot be viewed by naked eye,
but can be viewed through a microscope. There are several bands available in the
chromosome. To view the image clearly under microscope, G-band metaphase or
metaspread images is shown in Fig. 1 are taken for analysis [2].

V.S. Balaji
Biomedical Engineering, VIT University, Vellore 632014, Tamil Nadu, India
e-mail: [email protected]
S. Vidhya (&)
Biomedical Engineering, VIT University, Vellore 632014, Tamil Nadu, India
e-mail: [email protected]

© Springer India 2015 59

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_6
60 V.S. Balaji and S. Vidhya

Fig. 1 The G-band

metaphase human
chromosomes image before
karyotyping

There are twenty three pairs of chromosome in human beings. The first twenty
two pairs are called as autosomes and twenty third pair is called as sex chromosome
i.e. two X in female and one X and one Y in male [3]. The additional in chro-
mosome number i.e. forty seven instead of forty six or missing chromosome will
cause chromosome abnormality in human beings [4].
The anomaly in chromosome will cause birth defects which affect new born
babies and lead to mental or physical disabilities, improper body function and can
even be fatal sometimes [5, 6]. The leukemia, aneuploidy, deletion, duplication,
inversion and translocation will occur due to genetic disorders or any defects in
chromosome [7]. The study of cancer and diagnosing the defects in chromosome
images i.e. genetic disorders are important process. After separating the chromo-
some, the classification is followed. For karyotyping process as shown in Fig. 2 is
the human chromosomes are classified into twenty four different classes [8]. The
karyotyping process will include segmentation and classification [4].

2 Methodology, Results and Discussion

2.1 The Steps to be Considered While Segmenting an Image

(i) Get an overlapped or touching Chromosome images as input image.

(ii) Apply Binary approximation to input image.
(iii) Obtain Binary and Thresholding output.
(iv) Apply Contour to the threshold image.
(v) Apply a suitable algorithm to obtain the segmented output.
(vi) After segmentation, classification is followed.
Separation of Touching and Overlapped … 61

Fig. 2 The G-band

metaphase human
chromosomes image after
karyotyping

2.2 Description and Results

The abnormal chromosomes will cause chromosome anomaly in human begins this
can be identified by analysis G-band metaphase images. To overcome this problem,
the suitable segmentation algorithm should be followed for the approach [5, 6].
Before segmenting, the pre-processing steps to be considered (i, ii, iii, iv) is shown
in Fig. 3. In some cases the pre-processing steps will be included in algorithm itself.
The first four steps contribute to the pre-processing of the image. Initially, an
overlapped or touching chromosome as an input image was taken and binary
approximation was applied to the input image. To get the complete and fulfilled
information about an image Otsu thresholding was applied, followed by applying
contour [5, 6, 9–12].
By finding either the intersecting points (concave points) [5, 6] or the cut points
[10–12], or by the Voronoi diagrams and Delaunay triangulations [13] or Semi-
automatic Segmentation [14] the touching and overlapped chromosome can be
separated easily. After drawing a circle in Fig. 1, the image was investigated to see
that the number 1 in Fig. 4 represents the overlapped chromosomes and number 2 in
Fig. 4 represents the touching chromosomes in G-band image respectively.
62 V.S. Balaji and S. Vidhya

Fig. 3 The block diagram

The abnormal chromosomes in Fig. 5 will cause cancer, birth defects, fatal etc. in
human beings. This can be overcome by early diagnosis using appropriate seg-
mentation, which follows the steps in Fig. 3.
The separation of touching chromosomes in Fig. 7 is easier than the separation
of overlapped chromosomes in Fig. 6. This segmentation was done by semi-
automatic technique and currently by automatic technique also to be followed.
After segmentation, the chromosomes are classified into twenty four different
classes. This segmentation and classification of chromosome together follow the
karyotyping process [4–6]. This process is lengthy, quite difficult and repetitive
[15]. So the automatic segmentation of chromosome was followed. By segmenting
a chromosome image, it would be helpful for karyotyping analysis. The genetic
disorders in human can be diagnosed using segmentation method in Figs. 6 and 7
followed by karyotyping process in Fig. 2 [10–12].
Separation of Touching and Overlapped … 63

Fig. 4 The overlapped and touching chromosomes in G-band image shown in Fig. 1

Fig. 5 The sample overlapped and touching chromosomes (abnormal chromosomes) in Fig. 4

Fig. 6 The separation of

overlapped chromosomes
image
64 V.S. Balaji and S. Vidhya

Fig. 7 The separation of

touching chromosomes image

3 Conclusion

By identify touching and overlapped chromosomes in G-band metaphase image

under light microscope; one can separate the touching and overlapped chromo-
somes. This separation method is helpful for diagnosing the genetic disorders using
karyotyping method. Thus we may overcome cancer, genetic disease in human by
diagnosing this disorder.

Acknowledgments The authors wish to thank Ms. Nirmala Madian working as an Assistant
Professor and Phd research scholar in K.S.R College of Technology and Dr. Suresh, Director,
Birth Registry of India for providing the G-band chromosome images for the work.

References

1. Legrand B, Chang CS, Ong SH, Soek-Ying N, Nallasivam P (2008) Chromosome

classification using dynamic time warping. Pattern Recogn Lett 29:215–222
2. Cai N, Hu K, Xiong H, Li S, Su W, Zhu F (2004) Image segmentation of G bands of Triticum
monococcum chromosomes based on the model-based neural network. Pattern Recogn Lett
25:319–329
3. Piper J (1995) Genetic algorithm for applying constraints in chromosome classification.
Pattern Recogn Lett 16:857–864
4. Loganathan E, Anuja MR, Nirmala M (2013) Analysis of human chromosome images for the
identification of centromere position and length. IEEE Point Care Healthc Technol, pp 314–317
5. Madian N, Jayanthi KB (2002) Overlapped chromosome segmentation and separation of touching
chromosome for automated chromosome classification. In: IEEE EMBS, pp 5392–5395
6. Madian N, Jayanthi KB (2014) Analysis of human chromosome classification using centromere
position. Measurement 47:287–295
7. Roshtkhari MJ, Setarehdan SK (2008) A novel algorithm for straightening highly curved
images of human chromosome. Pattern Recogn Lett 29:1208–1217
8. Sampat MP, Bovik AC, Aggarwal JK, Castleman KR (2005) Supervised parametric and non-
parametric classification of chromosome images. Pattern Recognit 38:1209–1223
9. Gao H, Wenbo X, Sun J, Tang Y (2010) Multilevel thresholding for image segmentation
through an improved quantum-behaved particle swarm algorithm. IEEE Trans Instrum Meas
59:934–946
Separation of Touching and Overlapped … 65

10. Somasundaram D, Vijay Kumar VR (2014) Separation of overlapped chromosomes and

pairing of similar chromosomes for karyotyping analysis. Measurement 48:274–281
11. Somasundaram D, Vijay Kumar VR (2014) Straightening of highly curved human
chromosome for cytogenetic analysis. Measurement 47:880–892
12. Somasundaram D, Palaniswami S, Vijayabhasker R, Venkatesakumar V (2014) G-Band
chromosome segmentation, overlapped chromosome separation and visible band calculation.
Int J Hum Genet 14:73–81
13. Wacharapong S, Krisanadej J, Mullica J (2006) Segmentation of overlapping chromosome
images using computational geometry. Walailak J Sci Tech 3:181–194
14. Munot MV, Joshi MA, Mandhawkar P (2012) Semi automated segmentation of chromosomes
in metaphase cells. IET Conference on Image Processing, pp 1–6
15. Grisan E, Poletti E, Ruggeri A (2009) Automatic segmentation and disentangling of
chromosomes in Q-Band prometaphase images. IEEE Trans Inf Technol Biomed 13:575–581
Combination of CT Scan
and Radioimmunoscintigraphy
in Diagnosis and Prognosis
of Colorectal Cancer

Sutapa Biswas Majee, Narayan Chandra Majee and Gopa Roy Biswas

Abstract Staging of colorectal cancer constitutes an important part of its diagnosis

and prognosis. However, both invasive and non-invasive techniques prevail with
their own advantages and disadvantages. The present review focuses on the exis-
tence of complementarity between the information obtained from computerized
tomography and radioimmunoscintigraphy in the study of hepatic and extra-hepatic
lesions of significance and relevant to colorectal cancer. The latter technique utilizes
different monoclonal antibodies which are tagged with radioisotopes and imaging
done by gamma-camera. For complete diagnosis of recurrent carcinoma or
metastases, knowledge form both the pre-surgical procedures is an absolute
necessity to choose the correct therapeutic modality.

1 Introduction

Colorectal cancer is the commonest and the most preventable form of cancer where
survival rate can be improved markedly by selection of proper diagnosis of its
stages. Progression to adeno-carcinoma from adenomatous polyps takes place over
a span of time and through various stages as proposed by American Joint Com-
mittee on Cancer/Union Internationale Contre le Cancer (AJCC/UICC). The system
is referred to as TNM system which is used for both clinical and pathological
classification, where T denotes local extent of untreated primary tumor, N indicates
tumor involvement of regional lymph nodes and lymphatic system and finally M
refers to metastatic disease.

S.B. Majee (&)
G.R. Biswas

NSHM College of Pharmaceutical Technology, NSHM Knowledge Campus,
Kolkata-Group of Institution, 124 B L Saha Road, Kolkata 700053, India
e-mail: [email protected]
N.C. Majee
Department of Mathematics, Jadavpur University, Kolkata 700053, India

© Springer India 2015 67

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_7
68 S.B. Majee et al.

The prognosis of colorectal cancer depends on the stage at which examination is

being carried out which helps in determining the extent of tumor as well as
detecting regional and distant metastases. Invasive methods of staging colorectal
cancers like sigmoidoscopy, optical and screening computerized tomography (CT)
enabled colonoscopy involve surgical operations and have several drawbacks
including the failure to identify tumor deposits in lymph nodes, soft tissues in the
abdomen and retro peritoneal areas. However, any less invasive method, e.g. CT
colonography or virtual colonoscopy, capable of diagnosing and staging cancer
prior to surgery would always be more preferable. This method is suitable mainly
for screening of patients with polyps of larger size (>1 cm in diameter) which may
grow into tumors. Recent development in non-invasive pre-surgical cancer staging
is related to targeting of monoclonal antibodies to the antigenic markers of colo-
rectal cancer cells and radiolabeling with gamma-isotopes. Once administered, the
distribution of the radioisotope which is studied by gamma-immunoscintigraphy
gives a precise indication about the location, locoregional and metastatic spread,
size and severity of the tumor. Radioimmunoscintigraphy with 123I has been rou-
tinely used by various scientists with different fragments of IMMU-4[F(ab′)2 and F
(ab′)]in confirming suspected tumors and disclosing occult lesions along with CT.
Similarly, immunoscintigraphy with anti-CEA monoclonal antibody coupled with
single photon emission computerized tomography (SPECT) reduces the time delay
between diagnosis and treatment since it permits early diagnosis of recurrent car-
cinoma or metastasis of colorectal cancer [1–3].

2 Role of Monoclonal Antibodies (mAbs)

in Radioimmunoscintigraphy (RIS)

Monoclonal antibodies with high affinity are designed for specific tumor markers or
antigens expressed on the cell surface due to alterations in cell DNA. Ideally,
antigen should be produced in abundance (5,000 epitopes per cell) only from tumor
cells and not from normal cells or during other pathological condition. Moreover, it
should be expressed from tumors at various stages of differentiation. An ideal
monoclonal antibody should recognize only tumor cells and should possess limited
reactivity with the non-malignant cells. The most widely used monoclonal anti-
bodies used in diagnosis of colorectal cancer are IMMU-4 and PRIA-3, both of
which are targeted against carcinoembryonic antigen (CEA), an onco-fetal antigen
arising from gastrointestinal epithelium. Monoclonal murine antibody B72.3 can be
exploited for radioimmunoscintigraphy as TAG-72, its tumor-associated cell-sur-
face glycoprotein target can interact specifically with majority of the mucin-pro-
ducing colon adenocarcinomas. After isolation and purification, monoclonal
antibody can be site specifically conjugated with satumomab pendetide to form the
immunoconjugate which can then be labeled with either radioisotopes of Techne-
tium, Iodine, Indium or Rhenium. These isotopes are used because of the ease of
Combination of CT Scan and Radioimmunoscintigraphy … 69

labeling with them, physical characteristics, high percentage of intake per gram of
tumor tissue. 99mTc emits optimal energy gamma rays detectable by gamma camera
possesses a short half-life enabling imaging to be completed within the same day.
Although, uptake by liver and marrow is less compared to those of 111Indium,
higher percentage of urinary excretion leads to accumulation in bladder thereby
causing a hindrance to pelvis imaging. Moreover, due to its long half-life, imaging
should be continued for 48–72 h after administration. Radioisotope of Indium also
fails to provide good image of hepatic tissues, where colorectal cancer is known to
spread commonly during metastasis. 125I produces low radiation energy and long
half-life. Labeling of antibody fragment of anti-CEA, IMMU-4 with Technetium
has proved satisfactory in diagnosis of occult metastatic cancer which could not be
detected by abdominal and pelvic CT scans in patients with elevated CEA levels.
Stability of rhenium-labeled antibodies can be improved by chelating with tetra-
fluorophenyl-activated ester derivative of triamide thiolate. Intravenous adminis-
tration of the B 72.3 conjugate followed by imaging with gamma camera between
second and 7 day and separated by an interval of not less than 24 h, has been able to
detect cases of both primary and local recurrence of colorectal cancer including
occult lesions and incidences of liver metastases successfully and with high sen-
sitivity. However, there are reports of non-specific uptake in spleen and bone
marrow as well as gastrointestinal and genitourinary systems. Monoclonal antibody
PRIA-3 possesses high selectivity for CEA and could detect recurrent colorectal
cancers with high degree of accuracy [4–9].

3 Role of Monoclonal Antibody Fragments (Fabs)

in Radioimmunoscintigraphy (RIS)

Antibodies are characterized by two heavy chains and two light chains linked
together by disulphide bonds to form a Y-shaped configuration. The stalk of the Y
is the Fc portion constitutes the stalk of the Y and Fab portions represent the arms.
The tip of the Fab portion is responsible for reaction with antigen. Whole murine
monoclonal antibodies may induce an immune response in the form of allergic
reaction in 5–40 % of patients, due to the formation of human anti-murine anti-
bodies (HAMA) which are targeted against the Fc portion of the antibody. This can
be potentially avoided by use of Fab portion. There are also other reasons which
favour the use of antibody fragments instead of whole antibodies in radioimmu-
noscintigraphy. Use of antibody fragments accelerates the blood clearance com-
pared to intact form of monoclonal antibodies thereby reducing high background
noises. Moreover, cocktail of antibody fragments helps in recognition of different
epitopes, otherwise not recognizable by individual fractions due to heterogeneity of
tumor. An important monoclonal antibody which found use in detection of occult
metastatic colorectal cancer was radiolabeled and stabilized F(ab′)2 fragments of
ZCE-025, an anti-CEA monoclonal antibody and investigated by single photon
70 S.B. Majee et al.

emission CT (SPECT). It exhibited higher sensitivity at smaller protein doses than

the intact immunoglobulin antibody. Moreover, cancer could be detected in patients
who showed negative results with laparotomies and presence of tumor as observed
from positive scans could be confirmed by exploratory surgery or by diagnostic
biopsy. Indium-labelled ZCE-025 has been found to be quite useful in pre-operative
staging, distinguishing recurrent tumors from post-operative or post-radiation
alterations manifested on CT scans or any other radiographic images. Radioim-
munoscintigraphy and detection of colorectal cancer was also conducted with
radiolabelled monoclonal antibody fragments [F(ab′)2 BW 431/31] against CEA.
However, efforts should be taken to improve and optimize tumor affinity and
specificity. IMACIS 1 is a cocktail of mAb19–9F (ab′)2 and mAb anti-CEA F (ab′)
2 which is labeled with 131I. Another related antibody is Indimacis 19–9 containing
19–9F(ab′)2/DTPA fragments of mAbs.
There are several factors governing targeting and imaging modality with
monoclonal antibodies like stability of the monoclonal antibody fragment, the
labeling chemistry, modification of critical residues, the number of antigens
expressed on the cell surface, recycling rate of the target and ability of the target to
internalize upon binding [10–13].

4 CT Scan and RIS: Comparative Assessment in Diagnosis

of Colorectal Cancer

CT is an important imaging method for detection of lymph node metastases, where an

idea about size, number and sometimes morphology can be obtained. A significant
drawback of the method is that it possesses sensitivity varying from 22–73 % and may
provide incorrect information about malignancy if the size of the tumor is less than
1 cm. It has been reported that clusters of drainage nodes of smaller size located near
the primary tumor may be considered as malignant although it diminishes the extent
of specificity. On the other hand, radioimmunoscintigraphy has been found to locate
tumor growth in lymph nodes with size less than 1 cm and also strands of tumor
within the peritoneum. However, the technique seems to fail in detecting extrahepatic
metastases which present themselves as localized areas of increased uptake within the
abdomen. The main disadvantage with RIS is its low degree of spatial resolution. The
type of the radionuclide used alters the appearance of hepatic metastases which
appear as areas of increased isotope activity or photopenic areas. This problem seems
more acute with Indium-labelled antibodies. In scintiscans using technetium-labelled
antibodies against CEA-derived antigens hepatic metastases are observed as either
areas of increased radioactivity with a high uptake boundary, most probably due to
central necrosis in the core. In CT scan, accuracy in detection of hepatic metastases is
comparatively greater.
CT alone does not provide accurate information in the early stage of recurrent
carcinoma as the local anatomy is distorted after surgery. It has been observed that
Combination of CT Scan and Radioimmunoscintigraphy … 71

if recurrence were suspected from CT and RIS scans, the patients actually dem-
onstrated recurrence and reduced the need for biopsy. Any false positive inter-
pretation due to presence of fecal matter and bladder activity can be confirmed by
correlating the observations from both CT and RIS investigations. In some other
instances, Technetium-labeled anti-CEA scan could detect local disease and distant
metastases with improved sensitivity. Elevated levels of CEA in blood of post-
operative patients indicated signs of recurrence which could only be confirmed by
combining the favorable features of both RIS and CT study.
Moreover, following therapy with radiation, it becomes difficult to differentiate
fibrotic areas from the viable tissue. Introduction of multi-detector CT scanning
(MDCT) and improved processing software has enhanced the accuracy rates of
stage detection primarily for T patients but not so much for N patients. Therefore, it
must be realized that information from both the pre-surgical diagnostic procedures
is required for complete evaluation of different anatomic regions of human gastro-
intestinal tract and thus they provide complementary tools in the diagnosis and
prognosis of colorectal cancer, especially in those cases where it is known or
suspected to extend beyond the bowel. CT colonography shows promise in
assessment of synchronous lesions and metastasis [14, 15].

5 Radioimmunoscintigraphy Adds a New Milestone

Radioimmunoscintigraphy has paved the way for radioimmunoguided surgery

(RIGS) as a surgical intervention in management of colorectal cancer. In the pro-
cedure, an antitumor antibody is injected intravenously before surgery which is
followed during the operation with the help of a hand-held gamma camera probe, in
order to locate tumor selectively by counting of radioactivity in the operative field.
But prolonged blood clearance time of whole antibodies causes a delay prior to
surgical resection of tumor. Therefore, here also, single chain Fv antibodies have
been used with better tumor penetration and faster blood clearance enabling iden-
tification of small areas of increased uptake which would otherwise be invisible and
therefore difficult to localize. The success rate of the technique depends highly on
the availability of good and specific antibodies and of appropriate nuclides [16–18].

6 Conclusion

Tumor imaging with radiolabeled antibodies can be considered useful as only

neoplastic cells are specifically targeted without being localized in non-cancerous
cells. However, for complete anatomical profile of the patient with proven or
suspected incidence of colonic adenocarcinoma, combination of both CT scan and
radioimmunoscintigraphy is mandatory to avoid any biased-ness or false-negative
results and finally for better therapeutic management.
72 S.B. Majee et al.

References

1. Artiko V, Petrovic M, Jankovic Z, Jaukovic L, Sobic-Saranovic D, Grozdic I, Odalovic S,

Pavlovic S, Jaksic E, Zuvela M, Ajdinovic B, Matic S, Obradovic V (2012) Scintigraphic
detection of colon carcinomas with iodinated monoclonal antibodies. J BUON 17:695–699
2. Artiko V, Koljevic Marković A, Šobić-Šaranović D, Petrović M, Antić A, Stojković M, Žuvela
M, Šaranović D, Stojković M, Radovanović N, Galun D, Milovanović A, Milovanović J, Bobić-
Radovanović A, Krivokapic Z, Obradović V (2011) Monoclonal immunoscintigraphy for
detection of metastasis and recurrence of colorectal cancer. World J Gastroenterol 17:2424–
2430
3. Doerr RJ, Abdel-Nabi H, Krag D, Mitchell E (1991) Radiolabeled antibody imaging in the
management of colorectal cancer results of a multicenter clinical study. Ann Surg 214:118–124
4. Rodriguez-Bigas MA, Bakshi S, Stomper P, Blumenson LE, Petrelli NJ (1992) 99mTc—IMMU-
4 monoclonal antibody scan in colorectal cancer: a prospective study. Arch Surg 127:1321–1324
5. Patt YZ, Podoloff DA, Curley S, Kasi L, Smith R, Bhadkamkar V, Charnsangavej C (1994)
Technetium 99m-labeled IMMU-4, a monoclonal antibody against carcinoembryonic antigen,
for imaging of occult recurrent colorectal cancer in patients with rising serum carcinoembryonic
antigen levels. J Clin Oncol 12:489–495
6. Durbin H, Young S, Stewart LM, Wrba F, Rowan AJ, Snary D, Bodmer WF (1994) An epitope
on carcinoembryonic antigen defined by the clinically relevant antibody PR1A3. PNAS
91:4313–4317
7. Hosono MN, Hosono M, Zamara PO, Guhlke S, Haberberger T, Bender H, Russ Knapp FF,
Biersack HJ (1998) Localisation of colorectal carcinoma by rhenium-188-labelled B 72.3
antibody in xenografted mice. Ann Nucl Med 12:83–88
8. Doerr RJ, Abdel-Nabi H, Merchant B (1990) Indium 111 ZCE-025 Immunoscintigraphy in
occult recurrent colorectal cancer with elevated carcinoembryonic antigen level. Arch Surg
125:226–229
9. Lamki LM, Patt YZ, Rosenblum MG, Shanken LJ, Thompson LB, Schweighardt SA, Frincke
JM, Murray JL (1990) Metastatic colorectal cancer: radioimmunoscintigraphy with a stabilized
In-111-labeled F(ab′)2 fragment of an anti-CEA monoclonal antibody. Radiol 174:147–151
10. Patt YZ, Lamki LM, Shanken J, Jessup JM, Charnsangavej C, Levin B, Merchant C,
Halverson C, Murray JL (1990) Imaging with indium111-labeled anticarcinoembryonic
antigen monoclonal antibody ZCE-025 of recurrent colorectal or carcinoembryonic antigen-
producing cancer in patients with rising serum carcinoembryonic antigen levels and occult
metastases. J Clin Oncol 8:1246–1254
11. Abdel-Nabi HH, Schwartz AN, Higano CS, Wechter DG, Unger MW (1987) Colorectal
carcinoma: detection with indium-111 anticarcinoembryonic-antigen monoclonal antibody
ZCE-025. Radiol 164:617–621
12. Bares R, Fass J, Truong S, Buell U, Schumpelick V (1989) Radioimmunoscintigraphy with
111In labelled monoclonal antibody fragments (F(ab′)2 BW 431/31) against CEA:
radiolabelling, antibody kinetics and distribution, findings in tumour and non-tumour
patients. Nucl Med Commun 10:627–641
13. Oymada H, Uchida I, Nomura E, Yamada Y, Ohta H, Kasumi F, Yoshomoto M (1992) Clinical
trials of radioimmunoimaging with 111Indium-ZCE-025 and IMACIS-1. Radioisotopes 41:14–22
14. Saunders TH, Mendes Ribeiro HK, Gleeson FV (2002) New techniques for imaging colorectal
cancer: the use of MRI, PET and radioimmunoscintigraphy for primary staging and follow-
up. Brit Med Bull 64:81–99
15. Corbisiero RM, Yamauchi DM, Williams LE et al (1991) Comparison of immunoscintigraphy
and computerized tomography in identifying colorectal cancer: individual lesion analysis.
Cancer Res 51:5704–5711
Combination of CT Scan and Radioimmunoscintigraphy … 73

16. Kim JC, Kim WS, Ryu JS et al (2000) Applicability of carcinoembryonic antigen-specific
monoclonal antibodies to radioimmunoguided surgery for human colorectal carcinoma.
Cancer Res 60:4825–4829
17. Mayer A, Tsiompanou E, O’Malley D et al (2000) Radioimmunoguided surgery in colorectal
cancer using a genetically engineered anti-CEA single-chain Fv antibody. Clin Cancer Res
6:1711–1719
18. Arnold MW, Schneebaum S, Martin EW Jr (1999) Radioimmunoguided surgery in the
treatment and evaluation of rectal cancer patients. Cancer Cont:JMCC 3:42–45
Enhanced Color Image Segmentation
by Graph Cut Method in General
and Medical Images

B. Basavaprasad and M. Ravi

Abstract Segmentation of color images is a tricky task. Regaining the segments in

an image using the content image is a difficult and significant problem. Medical
imaging is the most alive research topic from the past two decades. From the medical
diagnosis of patients suffering from various diseases, abnormal regions in the organs
can be easily identified, which is a greatest achievement. It is experimentally proved
that graph based segmentation methods are better than the other segmentation
techniques, especially when combined with statistical methods. We have proposed
color image segmentation by adaptive graph cut method in this paper. It consists of
two important stages. During the first phase we enhance the input color image using
transformation technique as the image may contain noise, may be of low contrast and
missing some color statistics. Then this enhanced color image is processed under
graph cut technique to get better results, especially for the analysis of medical and
general images. The proposed method contributes to medical imaging by means of
image segmentation and also to other general image analysis. Our experimental
results are found to very good in segmenting color images.

Keywords Medical

Image enhancement

Fast fourier transform
Image



segmentation Graph Maximum flow Minimum cut

1 Introduction

In computer visualization, image segmentation is the process of dividing an image

into several segments which consists of sets of pixels. There set of pixels are also
called as super pixels. The objective of image segmentation is to make image

B. Basavaprasad (&)
M. Ravi

Research and Development Centre, Bharathiar University, Coimbatore 641046, India
e-mail: [email protected]
M. Ravi
e-mail: [email protected]

© Springer India 2015 75

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_8
76 B. Basavaprasad and M. Ravi

simpler and modify the presentation into further meaningful to understand and
easier to analyze [1–3]. The outcome of image segmentation is a collection of
segments that together cover the whole image, or a collection of regions taken out
from the image [4]. Every pixel in a segment is identical by means of to some
futuristic or calculated property, such as intensity, color or texture. Neighboring
regions also called as segments are considerably dissimilar by means of the same
properties or characteristics.
Graphs theoretical techniques can effectively be used for image segmentation.
Though many segmentation techniques exist but graph theoretical methods are
proven more efficient and accurate [5]. Generally a pixel or a collection of pixels are
vertices and edges define the dissimilarity among the neighborhood pixels. Graph
cut is one among the segmentation algorithms. Mincut/Maxflow [6] is one of the
techniques used to get cut on the graph. In this paper we have proposed enhanced
graph cut method for image segmentation consists of two main phases. During the
first phase the input color image is enhanced using the frequency domain methods.
The resultant enhanced image is then segmented using the conventional graph cut
technique. The proposed method can be used in various applications of image
processing and pattern recognition systems such as medical imaging, in particular
analysis of endoscopic images, tumor images, defected organs etc. It can be used in
other applications such as face recognition, biometric, finger print recognition,
detection of satellite images and much more. The paper is organized as follows.
Proposed method is described in Sect. 2. Results and discussion are explained in
Sect. 3. And finally experimental results and comparison with other segmentation
methods are discussed in Sects. 3 and 4.

2 Proposed Method

A method which is a combination of two techniques is presented in this paper and

hence contains two phases. During the first phase we are enhancing the input color
image using transform technique [7]. During the second and final phase the
enhanced image is processed under graph cut to get the quality segmented image [8].

2.1 Image Enhancement

The main goal of this technique is to get an enhanced image to protect the natu-
ralness of an image. This method can be applied for all kinds of image; also a dark
image can be brightening to visualize the dark image information. This method
analyses the Bilog Transform, Fast Fourier Transform and NTSC space color
enhancement [7] as shown in the following Fig. 1.
Enhanced Color Image Segmentation by Graph Cut Method … 77

Fig. 1 Image enhancement

using transform technique

2.1.1 Apply Fast Fourier Transform

The input image is a color image and this color image consists of three primary
spectral components of red, green and blue i.e., RGB color panels. This RGB color
Panels has to be segregated to grayscale images i.e., binary images. HSV. FFT filter
is used to filter the pixels based on intensity. Then the low frequency components
representing zeros are to be shifted to the center of array matrix in frequency
domain. Inverse operation is performed to reconstruct the image in frequency
domain to its spatial domain so as to picturize the image whose dimension is N  N.
The equation for FFT is given by

1XN 1
Pðk; bÞel2pN
lb
F ðk; lÞ ¼ ð1Þ
N b¼0
78 B. Basavaprasad and M. Ravi

1XN 1
f ða; bÞel2p N
ka
Pðk; bÞ ¼ ð2Þ
N a¼0

The equation for inverse Fourier Transform is given by

1 X N 1 X
N 1
F ðk; lÞel2pð n þ N Þ
ka lb
f ða; bÞ ¼ ð3Þ
N 2 k¼0 l¼0

where f ða; bÞ the image in the spatial domain is, 1

N2 is the normalization term in
inverse FT, F ðk; lÞ is the point in Fourier space.

2.1.2 Apply Bilog Transformation

Still there may be some presence of negative frequency components (zero fre-
quency components). Bilog transformation is made use of here to perform action on
low frequency information. The region near zeros is to be highlighted for the
enhancement and brightness preservation. Hence, after the application of this
transform, the region around zeros is enhanced. This is followed by grouping of
pixels, where clustering is done to increase the high resolution pixels. At this stage,
the image pixels are converted back to RGB color model and pixels highlighted to a
certain level.

2.1.3 NTSC Color Space Enhancement

Further enhancement is done using NTSC format. The NTSC stands for national
television system committee. This color space is used in televisions in the United
States. In this format the RGB color panels are converted to YIQ and converting
back YIQ panels to RGB color model to process the gray scale and color infor-
mation present in the image hence, the resulting image is an enhanced image.
The algorithm steps are as follows:
• Read the input image from the file and output the image. Transform RGB color
panels to binary images i.e., to HSV images.
• Image matrix is initialized as KLAP and other variables. Then rewrite the values
of matrix by comparing it with the original or actual image matrix.
• Use FFT filter to perform FAST FOURIER TRANSFORM, as FFT filter is used
to filter the pixels based on intensity i.e., low resolution pixels and high reso-
lution pixels.
• Use Envelope function to convert low resolution pixels into high resolution
pixels. Perform Inverse FFT wherever needed to reconstruct the image.
Enhanced Color Image Segmentation by Graph Cut Method … 79

• Perform BILOG TRANSFORM, as image does not contain same pixels some
might be very large or small and regroup Pixels using envelope check function
to increase high resolution pixels in image.
• Further enhance the image using NTSC format. Cost function performs the
NTSC color space enhancement.
• Enhance the image by Performing L:  R and Convert back HSV into RGB
image to view image in RGB color model.

2.2 Image Segmentation Using Graph Cuts

Let G ¼ ðV; EÞ be graph with V as, set containing all the vertices of a graph G and
E is an edge set which contains all the edges of a graph G. A cut is a set of edges
C  E such that the two terminals become separated on the induced graph G0 ¼
ðV; EnC Þ Denoting a source terminal as s and a sink terminal as t, a cut ðS; T Þ of is a
partition of V into S and T ¼ VnS, such that t 2 T and s 2 S: A flow network is
defined as a directed graph where an edge has a nonnegative capacity. A flow in
G is a real-valued (often integer) function that satisfies the following three
properties:

Capacity Constraint: 8u; v 2 V; f ðu; vÞ  cðu; vÞ ð4Þ

Skew Symmetry: 8u; v 2 V; f ðu; vÞ ¼ f ðu; vÞ ð5Þ

!
X
Flow Conservation: u 2 V fs; tg; f ðu; vÞ ¼ 0 ð6Þ
v2V

• If f is a flow, then the net flow across the cut ðS; T Þ is defined to be f ðS; T Þ,
which is the sum of all edge capacities from S to T subtracted by the sum of all
edge capacities from T to S.
• The capacity of the cut ðS; T Þ is CðS; T Þ, which is the sum of the capacities of all
edge from S to T.
• A minimum cut is a cut whose capacity is the minimum over all cuts of G.

2.2.1 Finding the Min-Cut

After the max-flow is found, the minimum cut is determined by S ¼ fAll vertices
reachable from sg:
Finding the cut with minimal cost is solvable in polynomial time as shown in the
Fig. 2. A directed graph shown below in Fig. 3a with positive edge weights and two
special vertices: A source s with only outgoing edges and a sink T with only
80 B. Basavaprasad and M. Ravi

Fig. 2 Finding the min-cut of a graph

Fig. 3 A graph G and its corresponding cut

incoming edges. On this graph a cut (shown Fig. 3b) is a binary partition of the
vertices into a set S around the source and a set T around the sink. The cost of the
cut is the sum of the weights of all the edges inducing flow from source to sink. Cut
edges that induce flow in the opposite direction. Binary labeling is equivalent to
partitioning, so construct a directed graph. All edges in the graph are assigned some
weight or cost. A cost of a directed edge ðp; qÞ may differ from the cost of the
reverse edge ðq; pÞ. In fact, ability to assign different edge weights for ðp; qÞ and
ðq; pÞ is important for many graph-based applications in vision. Normally, there are
two types of edges in the graph: N-links and T-links. N-links connect pairs of
neighboring pixels or voxels. Thus, they represent a neighborhood system in the
image. Cost of N-links corresponds to a penalty for discontinuity between the
pixels. T-links connect pixels with terminals (labels). The cost of a T-link con-
necting a pixel and a terminal corresponds to a penalty for assigning the corre-
sponding label to the pixel.
Enhanced Color Image Segmentation by Graph Cut Method … 81

2.2.2 Maxflow Algorithm

The max-flow algorithm presented here belongs to the group of algorithms. Figure 4
is an example of the search trees S (red nodes) and T (blue nodes) at the end of the
growth stage when a path (yellow line) from the source s to the sink t is found.
Active and passive nodes are labeled by letters A and P, correspondingly. Free nodes
appear in black on augmenting paths. It builds search trees for detecting augmenting
paths. Two trees search is constructed; first tree is called source and the second is
called as the sink. The other difference is that reuse of these trees is done and never
start building them from scratch. This method has one drawback i.e., paths of the
augmentation created are not essentially need to be shortest; thus the time com-
plexity of the shortest augmenting path is no longer valid. The trivial upper bound on
the number of augmentations for our algorithm is the cost of the minimum cut jCj,
which results in the worst case complexity Oðmn2jC jÞ.

S  V; s 2 S; T  V; t 2 T; S \ T ¼ ; ð7Þ

Figure 4 illustrates the basic terminology. Two non-overlapping search trees ‘S’
and ‘T’ with roots at the source s and the sink t, correspondingly. The nodes which
do not exist in S or T are called as free nodes. These nodes can be either active or
passive. The nodes which are active stand for the external border in every tree at the
same time the nodes which are passive represents internal. One thing can be notices
that the nodes which are active permits the trees to grow by attaining children which
are new beside non-saturated edges from a group of nodes which are free. The
nodes which are passive are totally blocked by remaining nodes from the tree and
hence cannot grow. It is also vital that active nodes are able to come in touch by
way of the nodes from the outside tree.
There are three important stages of this algorithm which iteratively repeats.
• Growth: Look for trees S and T grow until they touch giving a path s ! t.
• Augmentation: In this phase the path is augmented and the tree searches will
split into forests.
• Adoption: Here the both the S and T trees are regained.
During the stage of growth the trees get bigger. The active nodes travel around the
neighboring edges which are non-saturated and obtain fresh children from a group of
nodes which are free. The recently acquired nodes would be the active members of
the consequent trees. Every adjacent active node which would travel around these

Fig. 4 Path from source to sink

82 B. Basavaprasad and M. Ravi

nodes will be converted into passive nodes. Suppose if the nodes which are active
come across a adjacent node belonging to the opposite tree, then growth stage ends.
Therefore a path is found from source to sink (Fig. 4). Because the push is gutter the
largest flow possible, as a result a few of the edges during the path get soaked.
As soon as the stage of the adoption is finishes the algorithm proceeds back to
the stage of growth. The algorithm ends when there is no further growth of the
search trees S and T. As a result the trees are separated from each other by edges
which are saturated. This indicates that a flow is reached to its maximum.

3 Experimental Results

The experimental results using the proposed method are shown in Fig. 5. First
column images represent the original input images. The second column contains the
enhanced images using the frequency domain method. The third and final column

Fig. 5 Different image segmentation using the proposed method

Enhanced Color Image Segmentation by Graph Cut Method … 83

Table 1 Comparison of
Sl. No. Method Performance in second(s)
execution time in seconds of
different algorithms with the 1 Graph cut 0.6
proposed method 2 Contour 3
3 Region growing 11
4 K-means 20
5 Pattern Matching 120

represents the segmented color images using graph cuts. We have taken both
general and medical images for our experimentation. Over 200 images are tested
using the proposed method. The results are very encouraging and useful in both
medical and general image analysis such as tumor image analysis, MRI scanned
images, endoscopic images, general images such as natural images, building ima-
ges, flowers and so on and so forth.
There exist many image segmentation algorithms. Among them graph based
image segmentation techniques are proved as the efficient and accurate once.
Table 1 shows different segmentation approaches compared with graph cut method.
We observed that methods such as region growing, k-means clustering, contour are
very slow in their segmentation tasks. Whereas graph cut methods are very fast in
segmenting the images. We have improved the quality of the segmentation by
enhancing the original input images using frequency domain methods. This will
improve the image analysis with respect to human perception is get bettered on
output images.

4 Conclusion

A color image segmentation using adaptive graph cut method for medical and
general has been presented in this paper. The input color image is enhanced using
transform method and then processed under graph cut. As all we know that graph
based image segmentation techniques yield better results especially for medical
images to detect the abnormality caused by diseases in human organs. This tech-
nique enhances the detection of abnormal region of the organ caused by various
diseases with respect to human perception. Here we tried to improve the graph cut
method by giving the enhanced image as input for further improved results. The
experimental results we got are tested over 200 images which consist of medical
and general images and are found to be very encouraging. The powerful min-cut or
max-flow algorithm from combinatorial optimization can be used to minimize
certain important energy functions in computer vision. The proposed method can be
used in both medical and general image analysis.
84 B. Basavaprasad and M. Ravi

References

1. Basavaprasad B, Hegadi RS (2012) Graph theoretical approaches for image segmentation.

Aviskar–Solapur Univ Res J 2:7–13
2. Shi J, Malik J (1998) Normalized cuts and image segmentation. In: IEEE Trans PAMI, vol 22,
no 8, Aug 2000S
3. Metev M, Veiko VP, Osgood RM Jr (eds) (1998) Laser assisted microtechnology, 2nd ed.
Springer, Berlin, Germany
4. Bhambri P, Kaur A (2013) Novel technique for robust image segmentation: new technique of
segmentation in digital image processing. Lambert Academic Publishing, Germany
5. Basavaprasad B, Hegadi RS (2014) A survey on traditional and graph theoretical techniques for
image segmentation. IJCA Proc Natl Conf Recent Adv Inf Technnol (NCRAIT) (1):38–46
6. Boykov Y, Kolmogorov V (2004) An experimental comparison of min-cut/max-flow algorithms
for energy mimization in vision. IEEE Trans Pattern Anal Mach Intell 26(9):1124–1137
7. Umamageswari D, Sivasakthi L, Vani R (2014) Quality analysis and image enhancement using
transformation techniques. In: International conference on signal processing, embedded system
and communication technologies and their applications for sustainable and renewable energy
(ICSECSRE ’14), vol 3
8. Boykov Y, Veksler O, Zabih R (2001) Fast approximate energy minimization via graph cuts.
IEEE Trans Pattern Anal Mach Intell 23(11):1222–1239
A New Approach for Color Distorted
Region Removal in Diabetic Retinopathy
Detection

Nilarun Mukherjee and Himadri Sekhar Dutta

Abstract Automatic detection of Diabetic Retinopathy (DR) abnormalities in

fundus retinal images can assist in early diagnosis and timely treatment of DR, to
avoid further deterioration of vision. Many Fundus Retinal images contain color
distorted regions originated due to noise, extremely uneven or poor illumination
and improper exposure of fundus camera. These regions are required to be removed
to avoid poor results for feature extraction and erroneous DR abnormality detec-
tions, as they introduce high amount of false positive detections. In this paper, we
have proposed a totally automatic method for segmentation and removal of the
color distorted regions in retinal fundus images, using modified Valley Emphasized
Automatic thresholding method and morphological operations. The proposed
algorithm accurately defines the well illuminated color undistorted retinal region
inside the input fundus image, from which both the normal and disease features can
be successfully detected. The proposed method has yield an average accuracy of
more than 95 % when tested over around 700 fundus images from diaretdb0,
diaretdb1, STARE, HRFDB and DRIVE databases.

Keywords Diabetic retinopathy Retinal fundus image Automatic thresholding

Color distorted region segmentation

1 Introduction

Diabetic Retinopathy (DR) is one of the most harmful effects of Diabetes, leading to
blindness. Diabetic retinopathy (DR) can be defined as the damage of the micro-
vascular system in the retina, due to prolonged Hyperglycemia [1]. Blockages or

N. Mukherjee (&)
Bengal Institute of Technology, Kolkata, West Bengal, India
e-mail: [email protected]
H.S. Dutta
IEEE, Kalyani Government Engineering College, Kalyani, West Bengal, India
e-mail: [email protected]

© Springer India 2015 85

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_9
86 N. Mukherjee and H.S. Dutta

clots are formed, as blood containing a high level of glucose flows through the
small blood vessels in the retina. This in effect raptures the wall of those weak
vessels due to high pressure. The leakage of blood on surface of retina leads to
blurred vision and can cause complete blindness, known as Diabetic Retinopathy
[1]. Study showed that the major systemic risk factors for onset and progression of
DR are duration of diabetes, degree of glycemic control and hyper-lipidemia. DR is
a vascular disorder affecting the microvasculature of the retina. It is estimated that
diabetes mellitus affects 4 % of the world’s population, almost half of whom have
some degree of DR at any given time [2]. DR occurs both in type 1 and type 2
diabetes mellitus. Earlier epidemiological studies has shown that nearly 100 % of
type 1 and 75 % of type 2 diabetes develop DR, after 15 year duration of diabetes
[3, 4]. In India, with epidemical increase in type II diabetes mellitus, diabetic
retinopathy is fast becoming an important cause of visual disability, as reported by
the World Health Organization (WHO) [5]. However, with early diagnosis and
timely treatment Diabetic Retinopathy can be well treated.
The regular screening of diabetic retinopathy produces a large number of retinal
images, to be examined by the ophthalmologists. The cost and time of manual
examination of large number of retinal images becomes quite high. An automated
screening system for retinal images has become esteem need for early and in time
detection of DR [1]. The automated screening system should be able to differentiate
between normal and abnormal retinal images. It should also be able to detect
different effects of Diabetic Retinopathy such as Exudates, Micro Aneurysms and
Hemorrhages in the retinal images with adequate accuracy. Such a system will
greatly facilitate and accelerate the DR detection process and will reduce the
workload for the ophthalmologists. Healthy Retina contains Blood Vessels, Optic
Disc, Macula and Fovea as main components, as depicted in Fig. 1. An automated
system for screening and diagnosis of DR should be able to identify and eliminate
all these normal features [6–8] prior to automatically detecting all signs of Diabetic
Retinopathy such as Micro-Aneurysms [9–11] and Edema and Hemorrhages [9]

Fig. 1 Illustration of various

features on a typical image of
retina
A New Approach for Color Distorted Region Removal … 87

and Exudates and Cotton-wool Spots [12, 13]. Accurate detection of DR depends
highly on the quality of the retinal fundus images, which practically varies widely
due to noise and uneven illumination.
Detection of these abnormalities requires perfect separation of the regions con-
tained in those retinal images, which are outside the retina and belong to the back-
ground. In many cases, due to noise, uneven and poor illumination and degradation of
illumination away from the center and improper exposure of the fundus camera,
certain regions inside the retina become totally unrecoverable or unusable due to color
distortion. These color distorted retinal regions are required to be removed during the
preprocessing steps, to extract and define the actual Region of Interest (ROI) before
applying any DR detection algorithm. Otherwise it may lead to poor results for feature
extraction and erroneous abnormality detections. These preprocessing steps are
applied to enhance the quality of the retinal images to make them suitable for reliable
detection of DR abnormalities by applying any DR detection algorithms.
In this paper, we have proposed a new fully automated algorithm for color
distorted unrecoverable retinal region segmentation for retinal fundus images. The
paper has been organized as follows: In Sect. 2 we have discussed all the related
works. Then in Sect. 3, the proposed method has been depicted along with actual
output images corresponding to each processing step, from its software imple-
mentation. In Sect. 4, we have provided both subjective and analytical accuracy and
performance analysis of the proposed method with supporting experimental results
and sensitivity-specificity analysis.

2 Related Work

Design and development of automated and accurate Diabetic Retinopathy detection

systems, has gained significant research interests in recent times. Many contribu-
tions has been made till date, on preprocessing and background segmentation mask
creation for retinal images. Which are necessary for accurate detection of DR. The
aim of preprocessing is to increase the quality of an image by reducing the amount
of noise appearing in the image and highlighting features that are used in image
segmentation. Chaudhuri et al. has used a fixed global threshold on the I channel of
the HSI format of the retinal images [8]. Two typical techniques used in prepro-
cessing are filtering and contrast enhancing. Lee et al., Goldbaum et al. and Osareh
et al. have applied standard contrast stretching techniques for segmentation and
noise reduction [9, 14, 15]. Usher et al., Sinthanayothin et al. and Firdausy et al.
have used local contrast enhancement method for equalizing uneven illumination in
the intensity channel of retinal images [16–18]. A large mean filter and a large
median filter are collectively used on the intensity channel values to detect the dark
regions from retinal images. Thresholding is also an important and widely used
technique in image segmentation [19], because thresholding is effective and simple
to implement. In thresholding, pixels having gray level intensity values within a
defined intensity range are selected as belonging to the foreground objects, whereas
88 N. Mukherjee and H.S. Dutta

pixels having gray levels outside that range are rejected to be the background [19].
Hoover et al. and Goldbaum et al. have used thresholding on the Mahalanabis
Distance over a neighborhood for each pixel, for background estimation [20, 21].
Jamal et al. has used thresholding on the Standard Deviation over a neighborhood
for each pixel for background estimation and removes noise using HSI color space
[22]. The threshold is chosen in an empirical basis. Kuivalainen et al. has thres-
holded the Intensity (I) channel in the HSI converted retinal fundus images using a
sufficiently low I value, to form the background segmentation mask [23]. The I
channel threshold is experimentally or empirically chosen. From the training image
set, Kuivalainen et al. has found that the regions of distorted color due to inadequate
illumination, have high hue values (H) and relatively low intensity values (I) in the
HSI color system. Thus, regions having distorted color was found by, first dividing
the hue channel by the intensity channel and then thresholding the result with a
preset threshold, to form the distorted region segmentation mask [23]. The preset
threshold is also empirically chosen.
Although, there have been a lot of work in background and color distorted region
segmentation in retinal fundus images, but all of them have used some empirically
selected threshold values to create the masks, which requires manual intervention.
This is turn, restricts the entire process of DR abnormality detection from becoming
totally automatic. In this paper, we have proposed a totally automated and dynamic
threshold selection method to create the color distorted region segmentation mask
for any given retinal fundus image.

3 Proposed Method

In this paper, an intuitive and fully automatic technique for detection and removal
of the color distorted unrecoverable retinal regions has being proposed, which takes
advantage of the bimodal nature of the red channel histograms of the input retinal
images. It has been found through rigorous testing on the retinal images from
STARE [20, 21], DRIVE [24], diaretdb0 [25], diaretdb1 [26] and HRFDB [27]
databases, that the red channel histograms of most of the retinal fundus images
exhibits clear bimodal nature, having a clear separation i.e. valley region between
the background and object regions. Moreover, it is established that the red channel
does not contain much information regarding the retinal features and abnormalities.
It only contains the illumination difference information of the retinal region and the
background, whereas green channel or the intensity (I) channel for HSI converted
retinal images, contains most of the retinal component information. This opens the
opportunity to apply a modified version of Valley Emphasis Method [28] for
Automatic Threshold Selection, to segment the color distorted object regions.
Therefore, the red channel threshold determined by the modified Valley Empha-
sized Method [28] is used to tune the I channel threshold, determined by the same
method, to determine the optimized threshold level in the red channel for color
distorted region segmentation.
A New Approach for Color Distorted Region Removal … 89

In Valley Emphasis Method [28] for Automatic Threshold Selection, an image is

represented by a 2D gray level intensity function f(x, y). The value of f(x, y) is the
gray level intensity value, ranging from 0 to L − 1, where L is the number of distinct
gray levels. Let the number of pixels with gray-level i be ni and n be the total number
of pixels in a given image, the probability of occurrence of gray-level i is defined as:
ni
pi ¼ ð1Þ
n

The average gray-level of the entire image is computed as:

X
L1
lT ¼ ipi ð2Þ
i¼0

In the case of single thresholding, the pixels of an image are divided into two
classes C1 = {0, 1,…, t} and C2 = {t + 1, t + 2,…, L − 1}, where t is the threshold
value. C1 and C2 normally correspond to the foreground (objects of interest) and the
background. Probabilities of the two classes are:

X
t X
L1
k1 ¼ pi and k2 ¼ pi ð3Þ
i¼0 i¼tþ1

The mean gray level values of the two classes are computed as:

Xt
ipi X
L1
ipi
l1 ðtÞ ¼ and l2 ðtÞ ¼ ð4Þ
k ðt Þ
i¼0 1
k ðt Þ
i¼tþ1 2

In the proposed algorithm, a modified version of the Valley Emphasis Method

[28] is used on the Red and Intensity Channel of the fundus retinal images. The
average intensity level of the original red and intensity channel image, have been
incorporated in the formulation for the Valley Emphasis Method [28]. The modified
Valley Emphasis threshold selection formula used in the proposed method is given
below. The optimal threshold τ can be determined by maximizing the between-class
variance; that is:
n  
s ¼ ArgMax ð1  pt Þ k1 ðtÞðl1 ðtÞ  lT Þ2
0  t\L
o ð5Þ
þ k2 ðtÞðl2 ðtÞ  lT Þ2

The key to this formulation is the application of a weight, (1 − pt). The smaller
the pt value (low probability of occurrence), the larger the weight will be. This
weight ensures that the result threshold will always be a value that resides at the
valley or bottom rim of the gray-level distribution. The objective of automatic
90 N. Mukherjee and H.S. Dutta

thresholding is to find the valley in the histogram that separates the foreground from
the background. For single thresholding, such threshold value exists at the valley of
the two peaks (bimodal), or at the bottom rim of a single peak (unimodal). The
modified Valley Emphasis Method exploits this observation to select a threshold
value that has a small probability of occurrence and also maximize the between
group variance.
Morphological operations such as Erosion, Dilation, Opening and Closing are
used to remove the boundary by inclusion of it inside the masked background
region and to minimize and remove small and medium sized white islands and
black holes both from the object and background regions. Morphological Erosion,
Dilation, Opening and Closing operations [19] in binary image is defined as:

Morphological Erosion: A ⊖ B = {x | (B)x ⊆ A}

Morphological Dilation: A ⊕ B = {x b x ∩ A ∉ Ø}
| ( B)
Morphological Opening: A ○ B = (A ⊖ B) ⊕ B
Morphological Closing: A ● B = (A ⊕ B) ⊖ B

where A is the input image and B is the structuring element. Then connected
component extraction is used on the resultant image to retain the largest connected
object, removing any white patches left inside the background. Then a second
connected component extraction is used on the negative of the resultant image to
retain the largest connected object, removing any black holes left inside the object.
Final color distorted unrecoverable retinal region segmentation mask is obtained by
smoothing the boundary of the negative of the resultant image.
The Proposed Method:
Step 1: The Red channel from the original RGB fundus image and the Intensity
(I) channel from the HSI [19] converted original fundus image are extracted.
Step 2: Noise Removal: A low pass filtering is done in spatial domain on the
resultant Red and I channel images using 7 × 7 Median filter [19], to remove any
sand paper noise present in the image.
Step 3: Threshold Selection: The modified version of the Valley Emphasis
Method depicted in Eq. 5 is applied on both of the resultant images to get the Red
and I channel threshold intensity levels. The final threshold is obtained by aver-
aging the Red and I channel threshold levels, which is used to threshold the Red
channel image. Figure 2 shows, the red channel (Black Bar) and I channel (Magenta
Bar) thresholds determined by the modified Valley Emphasized method, drawn
over the Red channel histogram of a sample fundus image from the image dat-
abases. The resultant binary masks contain the well illuminated retinal boundary
inside the object region. Figure 3 shows, sample original color fundus images from
diarectdb0 database in Column (a) and Corresponding Thresholded Red Channel
Images in Column (b).
Step 4: It has been observed through thorough examination that, the maximum
expansion of the retinal border approximately ranges to 15–17 pixels. The retinal
A New Approach for Color Distorted Region Removal … 91

Fig. 2 The red channel

(Black bar) and I channel
(Magenta bar) thresholds
determined by the modified
valley emphasized method
drawn over the red channel
histogram of a sample fundus
image

Fig. 3 Threshold selections

for fundus images with
bimodal distribution in red
channel: column a original
color fundus images from
database; column
b thresholded images

boundary inside the object region is eliminated and included inside the masked
background region using morphological Erosion operation [19] with a disc shaped
structuring element of size 17 × 17. This also helps to isolate or separate the object
regions corresponding to the high intensity color distorted regions inside the retina,
from the object regions corresponding to the color undistorted well-illuminated
regions inside the retina, in certain thresholded output images from step 3.
Step 5: The resultant mask output may also contain white islands in the back-
ground regions originated due to noise, improper exposure of fundus camera and
uneven illumination and black holes in the object region originated due to noise,
uneven illumination and dark abnormalities such as lesions and hemorrhages inside
retina. Small and medium sized white islands and black holes are removed using
morphological Opening and Closing operations [19], with a disc shaped structuring
element of size 15 × 15, as shown in Fig. 4.
92 N. Mukherjee and H.S. Dutta

Fig. 4 Result of morphological opening—closing operation

Step 6: Large sized white islands in the background regions originated due to
noise, improper exposure of fundus camera and uneven illumination, which are still
present in the resultant mask image are removed using connected component
labeling and extraction algorithm [19].
It is evident, that the largest connected component in the thresholded, opened
and closed mask image, will be the actual ROI i.e. the most prominent, well
illuminated, color undistorted useful region of the retina. All other smaller con-
nected components present, will either be noise or disjoint color distorted part of the
retinal region caused by extreme uneven illumination or/and abnormal exposure.
Exploiting this observation, all the connected components are extracted and labeled
accordingly. Only the largest connected component is preserved in the resultant
image, removing all remaining white islands in the background region, as shown in
Fig. 5a, b. 8–connectivity [19] among pixels have been considered in the connected
component labeling and extraction algorithm.
Step 7: After removing the white islands in the background region, a single large
connected component is retained. It may contain large sized black holes, originated
due to noise, uneven or poor illumination and dark abnormalities such as lesions
and hemorrhages. To remove them, the resultant image from step 6 is negated.

Fig. 5 Result of white

island removal is depicted in
(a and b) and result of black
hole removal is depicted
in (c and d)
A New Approach for Color Distorted Region Removal … 93

Fig. 6 Final color distorted unrecoverable retinal region segmentation masks

In the negative image, original background becomes the largest spotless connected
component and the original object region becomes the background. The large sized
black holes inside the original object region in the previous image become smaller
connected components. Exploiting this observation, all the connected components
in the negative image are extracted and labeled accordingly, using connected
component labeling and extraction algorithm [19]. Only the largest connected
component is preserved in the resultant image, thus removing all the remaining
large sized black holes in the original object region, as shown in Fig. 5c, d.
Step 8: In the resultant mask image, every black pixel having white pixels either
at both left and right in the same row or at both top and bottom in the same column
is replaced by a white pixel, to get the final boundary smoothed background seg-
mentation mask, as shown in Fig. 6. These final segmentation masks are intersected
with the original images to get the color distorted region segmented images.

4 Experimental Results

In medical image processing, it is crucial to verify the validity and evaluate any
newly proposed algorithm contributing to the automated diagnosis of a disease. We
have used five standard retinal image databases i.e. STARE [20, 21], DRIVE [24],
Diaretdb0 [25], Diaretdb1 [26] and HRFDB [27] as depicted in Table 1, to
extensively verify and validate the proposed method for color distorted unrecov-
erable retinal region segmentation. These databases contain both normal and DR
affected retinal images with different qualities in terms of noise and illumination.

Table 1 Retinal fundus

image databases Retinal image database No. of images Resolution
STARE [20, 21] 402 700 × 605
DRIVE [24] 40 565 × 585
DIARETDB0 [25] 130 1,500 × 1,152
DIARETDB1 [26] 89 1,500 × 1,152
HRFDB [27] 44 3,504 × 2,336
94 N. Mukherjee and H.S. Dutta

Figure 7, depicts the subjective validity of proposed method, it well represent the
diverse characteristics of the test images. Figure 7 shows the manually created
segmentation masks provided along with the Diaretdb0 database and the corre-
sponding automatically created segmentation masks by the proposed algorithm.
These results support the validity of the proposed technique and show that the
proposed automatic segmentation mask creation technique gives considerably
acceptable results for both well illuminated good quality fundus images and poorly
illuminated, noisy and color distorted fundus images. The quantitative accuracy
analysis of the proposed background and color distorted unrecoverable retinal
region segmentation algorithm is performed on the images from DRIVE, Diaretdb0
and HRFDB databases, for which manually labeled masks are available. The
manually labeled segmentation masks provided along with these image databases,
serve as ground truths and are used to calculate the accuracy of the proposed
algorithm.
For each of the fundus image in the set, the following metrics are calculated by
pixel by pixel comparison between the mask created by the proposed method and
the corresponding manually created mask provided with the respective image
databases:

Sensitivity = TPR (True Positive Rate) = ðTPTP

þFNÞ

Specificity = TNR (True Negative Rate) = ðTN TN

þFPÞ

FPR (False Positive Rate) = 1 − Specificity = ðFPFP

þTNÞ

Accuracy = Sensitivity  T þ Specificity  T ¼ þTTN

P N TP

Here, TP = The total number of those Pixels, which are detected as Object
Pixels, by the proposed automatic mask creation method, which are also Object

Fig. 7 Subjective validity of

the proposed algorithm: a and
c depict the final color
distorted region segmentation
masks created by the
proposed algorithm and b and
d depict the manually created
masks
A New Approach for Color Distorted Region Removal … 95

Table 2 Accuracy of the

Image database No. of images Accuracy
proposed automatic color
distorted region segmentation HRFDB [27] 44 99.93
method DRIVE [24] 40 99.49
DIARETDB0 [25] 130 94.98

Pixels in the manually created masks. TN = The total number of those Pixels, which
are detected as Background Pixels, by the proposed automatic mask creation
method, which are also Background Pixels in the manually created masks.
FN = The total number of those Pixels, which are detected as Background Pixels,
by the proposed automatic mask creation method, but which are Object Pixels in the
manually created masks. FP = The total number of those Pixels, which are detected
as Object Pixels, by the proposed automatic mask creation method, but which are
Background Pixels in the manually created masks. T = Total number of Pixels in
the Image. P = Total number of Object Pixels in the manually created mask.
N = Total number of Background Pixels in the manually created mask.
The Sensitivity, Specificity and thereafter Accuracy is calculated for each
individual automatically created mask by the proposed method. The overall accu-
racy of the proposed method for the three image databases i.e. DRIVE, Diaretdb0
and HRFDB databases are calculated separately, by taking the average accuracy of
all the images belonging to a particular database, which is depicted in Table 2.
It is evident from Table 2, that our proposed algorithm has worked quite effi-
ciently and correctly for the fundus images in all the three databases. The proposed
technique has failed for only one image in Diarectdb0 database and for some
Diarectdb0 fundus images it has resulted larger object area than the corresponding
manual masks. Although, it has been found that the automatic masks created by the
proposed algorithm have successfully rejected the really color distorted or poorly
illuminated portions inside the retinal regions for those particular Diarectdb0 fundus
images. However, the automated masks created by the proposed technique, have
considered some more regions inside the retina as object region than their corre-
sponding manual masks. These regions are found to contain certain important DR
abnormality information with adequate illumination, which have been unneces-
sarily rejected in the corresponding manual masks. A Java based standalone
application has been built to implement the proposed Color Distorted Unrecover-
able Retinal Region segmentation technique. All the outputs and histograms shown
in this paper are captured from the running instance of the application.

5 Conclusion

In this paper, a fully automated technique for segmentation of the color distorted
regions in retinal fundus images have been proposed. These color distorted retinal
regions originate due to noise and extremely uneven and poor illumination and
improper exposure of the fundus camera. These regions are required to be removed
96 N. Mukherjee and H.S. Dutta

to avoid poor results for feature extraction and erroneous DR abnormality detec-
tions, as they introduce high amount of false positive detections. Although, there
have been a lot of contributions in background and color distorted region seg-
mentation in retinal fundus images, but all of them have used some empirically
selected threshold values to create the masks, which requires manual intervention.
This is turn, restricted the entire process of DR abnormality detection from
becoming totally automatic. In this paper, we have overcame that limitation by
developing a totally automated and dynamic threshold selection method to create
the color distorted region segmentation mask for any given retinal fundus image.
The proposed algorithm accurately defines the well illuminated and color undis-
torted retinal region inside the input fundus image, from which both the normal and
disease features can be successfully detected. The proposed method yields satis-
factory results to segment color distorted retinal regions, when tested over around
700 images from diaretdb0, diaretdb1, STARE, HRFDB and DRIVE retinal image
databases, with an average accuracy of more than 95 %. Our technique may further
be combined with some automated learning methods to achieve better results, for
the few fundus images, for which the proposed algorithm has failed to produce
accurate segmentation masks.

References

1. Sussman EJ, Tsiaras WG, Soper KA (1982) Diagnosis of diabetic eye disease. JAMA
Ophthalmol 247(23):3231–3234
2. Rema M, Pradeepa R (2007) Diabetic retinopathy: an Indian perspective. Indian J Med Res
125:297–310
3. Klein R, Klein BE, Moss SE, Davis MD, DeMets DL (1984) The Wisconsin epidemiologic
study of diabetic retinopathy II. Prevalence and risk of diabetic retinopathy when age at
diagnosis is less than 30 years. Arch Ophthalmol 102:520–526
4. Klein R, Klein BE, Moss SE, Davis MD, DeMets DL (1984) The Wisconsin epidemiologic
study of diabetic retinopathy III. Prevalence and risk of diabetic retinopathy when age at
diagnosis is 30 or more years. Arch Ophthalmol 102:527–532
5. Wild S, Roglic G, Green A, Sicree R, King H (2004) Global prevalence of diabetes, estimates
for the year 2000 and projections for 2030. Diab Care 27:1047–1053
6. Sinthanayothin C, Boyce JF, Cook HL, Williamson TH (1999) Automated localization of the
optic disc, fovea and retinal blood vessels from digital color fundus images. Br J Ophthalmol
83(8):231–238
7. Foracchia M, Grisan E, Ruggeri A (2004) Detection of optic disc in retinal images by means of
a geometrical model of vessel structure. IEEE Trans Med Imaging 23(10):1189–1195
8. Chaudhuri S, Chatterjee S, Katz N, Nelson M, Goldbaum M (1989) Detection of blood vessels in
retinal images using two dimensional matched filters. IEEE Trans Med Imaging 8(3):263–269
9. Lee SC, Lee ET, Kingsley RM, Wang Y, Russell D, Klein R, Warner A (2001) Comparison of
diagnosis of early retinal lesions of diabetic retinopathy between a computer system and
human experts. Graefe’s Arch Clin Exp Ophthalmol 119(4):509–515
10. Spencer T, Phillips RP, Sharp PF, Forrester JV (1991) Automated detection and quantification
of micro-aneurysms in fluoresce in angiograms. Graefe’s Arch Clin Exp Ophthalmol 230
(1):36–41
A New Approach for Color Distorted Region Removal … 97

11. Frame AJ, Undill PE, Cree MJ, Olson JA, McHardy KC, Sharp PF, Forrester JF (1998) A
comparison of computer based classification methods applied to the detection of micro
aneurysms in ophthalmic fluoresce in angiograms. Comput Biol Med 28(3):225–238
12. Osareh A, Mirmehdi M, Thomas B, Markham R (2001) Automatic recognition of exudative
maculopathy using fuzzy c-means clustering and neural networks. In: Proceedings of
conference on medical image understanding analysis, pp 49–52
13. Phillips R, Forrester J, Sharp P (1993) Automated detection and quantification of retinal
exudates. Graefe’s Arch Clin Exp Ophthalmol 231(2):90–94
14. Goldbaum MH, Katz NP, Chaudhuri S, Nelson M, Kube P (1990) Digital image processing
for ocular fundus images. Ophthalmol Clin N Am 3(3):447–466
15. Osareh A, Mirmehdi M, Thomas B, Markham R, Classification and localization of diabetic-
related eye disease. In: Proceedings of 7th european conference on computer vision, vol 2353.
Springer LNCS, Copenhagen, Denmark, pp 502–516
16. Usher D, Dumskyj M, Himaga M, Williamson TH, Nussey S, Boyce J (2003) Automated
detection of diabetic retinopathy in digital retinal images: a tool for diabetic retinopathy
screening, diabetes UK. Diab Med 21(1):84–90
17. Sinthanayothin C, Kongbunkiat V, Ruenchanachain SP, Singlavanija A (2003) Automated
screening system for diabetic retinopathy. In: Proceedings of the 3rd international symposium
on image and signal processing and analysis, pp 915–920
18. Firdausy K, Sutikno T, Prasetyo E (2007) Image enhancement using contrast stretching on
RGB and IHS digital image. TELKOMNIKA 5(1):45–50
19. Gonzalez RC, Woods RE (2002) Digital image processing, 2nd edn. Prentice Hall, New Jersey
20. Hoover A, Kouznetsova V, Goldbaum M (2000) Locating blood vessels in retinal images by
piece-wise threshold probing of a matched filter response. IEEE Trans Med Imaging 19
(3):203–210
21. Hoover A, Goldbaum M (2003) locating the optic nerve in a retinal image using the fuzzy
convergence of the blood vessels. Trans Med Imaging 22(8):951–958
22. Jamal I, Akram MU, Tariq A (2012) Retinal image preprocessing: background and noise
segmentation. TELKOMNIKA 10(3):537–544
23. Kuivalainen M (2005) Retinal image analysis using machine vision, Master’s Thesis, 6 June
2005, pp 48–54
24. Staal JJ, Abramoff MD, Niemeijer M, Viergever MA, Ginneken BV (2004) Ridge based
vessel segmentation in color images of the retina. IEEE Trans Med Imaging 23:501–509
25. Kauppi T, Kamarainen V, Lensu JK, Sorri L, Uusitalo I, Kälviäinen H, Pietilä J (2006)
DIARETDB0, evaluation database and methodology for diabetic retinopathy algorithms,
Technical Report
26. Kauppi T, Kamarainen V, Lensu JK, Sorri L, Raninen A, Voutilainen R, Uusitalo I,
Kälviäinen H, Pietilä HJ (2007) DIARETDB1, diabetic retinopathy database and evaluation
protocol, Technical Report
27. Köhler T, Budai A, Kraus M, Odstrcilik J, Michelson G, Hornegger J (2013) Automatic
no-reference quality assessment for retinal fundus images using vessel segmentation. In: 26th
IEEE international symposium on computer-based medical systems, Porto
28. Hui-Fuang N (2006) Automatic thresholding for defect detection. Pattern Recogn Lett 27
(15):1644–1649
 Part II
Biomedical Instrumentation
and Measurements
A New Heat Treatment Topology
for Reheating of Blood Tissues After Open
Heart Surgery

Palash Pal, Pradip Kumar Sadhu, Nitai Pal and Prabir Bhowmik

Abstract This paper presents the human blood reheating technique for the blood
tissues after surgery as 37–51 °C temperature is required during surgery with high
frequency induction heating system. The surgeon opens the chest by dividing the
breastbone (sternum) and connects with the heart-lung machine to operate on the
heart. This machine allows the surgeon to operate directly on the heart by per-
forming the functions of the heart and lungs. The length of the operation will
depend on the type of surgery that is required for patient. Most surgeries take at
least 4–5 h. The preparation for surgery, which requires approximately 45–60 min
included in this time. After this operation the patients has required high temperature
blood for the continuation of blood flow to the heart as the human body temperature
is decreases after operation. Here high frequency converter technique can give
better topology for reheating the blood after open heart surgery by taking least time
than conventional system.

Keywords Open heart surgery Blood reheating Modified half bridge inverter



Induction heating MATLAB

P. Pal (&)
P.K. Sadhu
N. Pal

Department of Electrical Engineering, Indian School of Mines
(Under MHRD Government of India), Dhanbad, Jharkhand, India
e-mail: [email protected]
P.K. Sadhu
e-mail: [email protected]
N. Pal
e-mail: [email protected]
P. Bhowmik
AMRI Mukundapur, Kolkata 700099, India
e-mail: [email protected]

© Springer India 2015 101

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_10
102 P. Pal et al.

1 Introduction

In now a day’s for clean heat production high frequency induction heating are more
efficiently applicable for increasing quality of industrial equipments, domestic
equipments as well as medical purposes. The general purpose of a heat treatment is
to enhance human blood flow before transfused of blood to the human body [1, 2].
Basically superficial and deep heat treatment processes are used in medical system.
Superficial heat treatments apply heat to the outside of the body. Deep heat treat-
ments direct heat toward specific inner tissues through ultrasound technology and
by electric current. High frequency heat treatment is beneficial for reheating of
blood before transfusion to human body [3, 4].
Now the fact is that at the time of open heart surgery heart-lung machine is used
[5, 6]. This machine does the work for the heart and lungs to oxygenate and circulate
the blood through the body while allowing the surgical team to perform the detailed
operation on a still, non-beating heart. In that time the body temperature is too much
decreased which, is near about 10–15 °C [7–9]. If the reheated blood is inject to the
human body after open heart surgery, then the heart beat will be recovered and
corresponding all organs of human body will be well functioning [10–14].
There are different ways to convey heat such as conduction is the transfer of heat
between two objects in direct contact with each other, conversion is the transition of
one form of energy to heat, radiation involves the transmission and absorption of
electromagnetic waves to produce a heating effect, convection occurs when a liquid
or gas moves past a body part creating heat are used for heat treatment purposes [15].
Prior to the development of induction heating, microwave provided the prime
means of heating human blood [16]. Induction heating offers a number of advan-
tages over that heating such as quick heating, uniform heat distribution, Smooth and
easy temperature control, good compactness and high reliability, high energy
density. Moreover the high frequency induction heating provides other advantages
such as easy of automation and control, requirement of less maintenance, safe and
clean working conditions [17].

2 Methodology

High frequency induction heating is incorporated for human blood reheating.

Basically, Induction heating involves applying an ac electric signal to a coil placed
near specific place in the heating loop and the metallic object will be heated [18].
The alternating current creates an alternating magnetic flux within the metal to be
heated in the loop. Eddy emf is induced in the metal by the electromagnetic flux and
heats up the material. Fundamental theory of induction heating is similar as
transformer operation, where primary coil is treated as heating coils and the current
induced in secondary is directly proportional to primary current according to turn
ratio [17, 19].
A New Heat Treatment Topology … 103

Fig. 1 Diagram of blood

heating through non-metallic
pipe line

In this proposed induction heating topology blood is taking as a secondary

element of blood itself and secondary current will be flow through the human blood,
there by eddy current will be generated and blood will be heated as with required
temperature. Figure 1 shows the diagram of blood heating through non-metallic
pipe line.
Here the human blood is excited by the high frequency alternating current. The
simulation shows that the heating area can be effectively controlled by using the
cylindrical shield with adjustable space and thereby the temperature will be con-
trolled [20]. Figure 2 depicts that the proposed topology of blood reheating in the
non-metallic pipe line. If the small heating area is needed, it may require longer
treatment time. However, the efficiency of heat can be increased by varying the
radius size of cylinder there by more flux will be passes and more eddy emf is
induced. As a result the eddy current flows through blood cell and heat will be
generated within the blood cell due to presence of iron particle in the blood [21].
Generally the rate of change in total body heat (TBH) is calculated as follows,
The rate of change in total body heat = (total body heat at the end of reheating) −
(total body heat before reheating)/time taken for reheating.
In this topology the temperature can be controlled before transfusion of blood to
the human body which is required for operating all organs [22, 23]. Here the
inductive coil is a primary working coil of induction heating system which is

Fig. 2 Proposed blood

reheating topology through
the non-metallic pipe line
104 P. Pal et al.

energized by modified half bridge high frequency inverter and the human blood will
be working as secondary element of this high frequency inverter. The heating area
can be effectively controlled by using the cylindrical shield with adjustable space.
However, the efficiency of heat can be increased by varying the radius size of
cylinder thereby more flux appears and more eddy emf is induced. As a result the
eddy current flows through blood cell and blood is reheated [19].

3 Proposed High Frequency Modified Half Bridge

Inverter for Blood Reheating

In the circuit operation has been discussed in detail. Here human blood is con-
sidered secondary coil of heating element which can be passes through the vessel or
placed in the vessel thereby it can be reheated with this proposed inverter [19, 24].
The exact circuit diagram of the Modified Half Bridge inverter is shown in Fig. 3.
Modified half bridge circuit is normally used for higher power output. Four solid
state switches are used and two switches are triggered simultaneously. Here
MOSFETs (BF1107) are used as solid state switches because it can be exist at high
frequency applications. Anti-parallel diodes D1 and D2 are connected with the
switches S1 and S2 respectively that allows the current to flow when the main switch
is turned OFF. According to Table 1, when there is no signal at S1 and S2, capacitors
C1 and C2 are charged to a voltage of Vi/2 each. The Gate pulse appears at the gate
G1 to turn S1 ON. Capacitor C1 discharges through the path NOPTN. At the same
time capacitor C2 charges through the path MNOPTSYM. The discharging current
of C1 and the charging current of C2 simultaneously flow from P to T. In the next
slit of the gate pulse, S1 and S2 remain OFF and the capacitors charge to a voltage

Fig. 3 Proposed modified half-bridge inverter circuit diagram

Table 1 Switching ON–OFF

chart of MOSFETs (BF1107) S1 S2 Vout
ON OFF +Vi/2
OFF ON −Vi/2
A New Heat Treatment Topology … 105

Vi/2 each again. The Gate pulse appears at the gate G2, so turning on S2. The
capacitor C2 discharges through the path TPQST and the charging path for capacitor
C1 is MNTPQSYM. The discharging current of C2 and the charging current of C1
simultaneously flow from T to P. The both switches must operate alternatively
otherwise there may be a chance of short circuiting. In case of resistive load, the
current waveform follows the voltage waveform but not in case of reactive load. The
feedback diode operates for the reactive load when the voltage and current are of
opposite polarities.
In new topology, the modified half bridge inverter is operating at high frequency
(above 30 MHz) and feeding the blood for reheating. The high frequency alter-
nating current is created by switching two MOSFETs sequentially by an appropriate
logic circuit which keeps track of the frequency. The frequency can be varied by
varying the pulse rate of the logic train. The load is represented as a series com-
bination of resistance and inductance. Both these parameters vary along with
temperature rise. An inductance is placed in series with the rectifier output to
smooth out the ripples as far as possible so as to realize a current source.
Here the proposed high frequency induction heating topology can give clean
heated blood without damaging the blood cell components [17]. This is required
after surgery as human body temperature is decreased after open heart surgery as
37–51 °C temperature is required during surgery [25]. As the blood temperature is
decreased about 10–15 °C after open heart surgery so, this proposed scheme will be
too much for blood reheating which is required for human body to functioning all
organs after open heart surgery.

4 Simulation Results and Discussion

Here the proposed modified half-bridge high frequency inverter for blood reheating
after open heart surgery is simulated with MATLAB for getting output voltage and
current waveforms. Thereby the output harmonic current will be get and blood will
be reheated due to high harmonic current.
Figure 4 shows the wave-form of the output voltage for the proposed modified
half-bridge high frequency inverter. The rms value of output voltage is 309.38 V
across T and P point of Fig. 3. Figure 5 depicts that the output current wave form
for the same. The rms value of output current is 20.31 ampere through load working
coil, which is taken in the platform of PSIM.
The Fig. 6 depicts that the harmonic content of output that is for the load current.
The 3rd, 5th, 7th and 9th harmonics have a magnitude of 151, 119.7, 40 and 19.8 %
respectively of the fundamental component.
It is observed from the wave-shapes that the harmonics of the output current is
very high. In this proposed high frequency inverter high eddy current will be
generated at the output due to generation of high harmonics at the output. And the
blood temperature will be reached to 37 °C or above as per requirement. It is
required for blood reheating with a short period of time without damaging the blood
106 P. Pal et al.

Fig. 4 Output voltage waveform

Fig. 5 Output current waveform

Fig. 6 Output harmonic

content

cell components after open heart surgery. This temperature will be controlled by
adjusting space of cylindrical vessel. It is clear that the temperature field follows the
heat-source distribution quite well. That is, near the projection the heat source is
strong, which leads to high temperatures and the blood manages to keep the tissue
A New Heat Treatment Topology … 107

at normal body temperature after surgery. The eddy currents in a conductive cyl-
inder produce heat. Here, the ohmic losses and temperature distribute in the vessel,
the heat transfer and electric field simulations must be carried out simultaneously.
Finally the results are in good conformity with this proposed inverter scheme.

5 Conclusion

Hence this proposed high frequency induction heating topology will be more
suitable for blood reheating after open heart surgery. As these heat treatments have
the potential of human blood reaming without damaging blood composition from
excessive temperatures without creating any hazards due to easy control can be
possible where as other system can damage the blood cells. Again during heat
treatment results in easy, pollution free and clean heat production for the present of
power electronics converter. However proposed modified half bridge inverter will
give new setup in medical sciences for blood reheating before blood transfusion to
human body after open heart surgery.

Acknowledgments Authors are thankful to the UNIVERSITY GRANTS COMMISSION, Ba-

hadurshah Zafar Marg, New Delhi, India for granting financial support under Major Research
Project entitled “Simulation of high-frequency mirror inverter for energy efficient induction heated
cooking oven” and also grateful to the Under Secretary and Joint Secretary of UGC, India for their
active co-operation.

References

1. Sharkey A, Gulden RH, Lipton JM, Giesecke AH (1993) Effect of radiant heat on the
metabolic cost of postoperative shivering. Br J Anaesth 70:449–450
2. Sessler DI, Moayeri A (1990) Skin-surface warming: heat flux and central temperature.
Anesthesiology 73:218–224
3. Shander A, Hofmann A, Gombotz H, Theusinger OM, Spahn DR (2007) Estimating the cost
of blood: past, present, and future directions. Best Pract Res Clin Anaesthesiol 21:271–289
4. Burns JM, Yang X, Forouzan O, Sosa JM, Shevkoplyas SS (2012) Artificial micro vascular
network: a new tool for measuring rheologic properties of stored red blood cells. Transfusion
52(5):1010–1023
5. Van Beekvelt MC, Colier WN, Wevers RA, van Engelen BG (2001) Performance of near-
infrared spectroscopy in measuring local O2 consumption and blood flow in skeletal muscle.
J Appl Physiol 90(2):511–519. PMID 11160049
6. Sinha D, Sadhu PK, Pal N (2012) Design of an induction heating unit used in Hyperthermia
treatment- advances in therapeutic engineering, CRC Press, Taylor and Francis Group,
ISBN:978-1-4398-7173-7, pp 215–266 (Chapter 11)
7. Gersh BJ, Sliwa K, Mayosi BM, Yusuf S (2010) Novel therapeutic concepts: the epidemic of
cardiovascular disease in the developing world: global implications. Eur Heart J 31(6):642–648
8. Logmani L, Jariani AL, Borhani F (2006) Effect of preoperative instruction on postoperative
depression in patients undergoing open heart surgery. Daneshvar pezeshki 14(67):33–42.
[Persion]
108 P. Pal et al.

9. McAlister FA, Man J, Bistritz L, Amad H, Tandon P (2003) Diabetes and coronary artery
bypass surgery: an examination of perioperative glycemic control and outcomes. Diabetes
Care 26:1518–1524
10. Padmanaban P, Toora B (2011) Hemoglobin: emerging marker in stable coronary artery
disease. Chron Young Sci 2(2):109. doi:10.4103/2229-5186.82971
11. Handin RI, Lux SE, Stossel B, Thomas PB (2003) Principles and practice of hematology.
Lippincott Williams and Wilkins. ISBN:0781719933
12. Minic Z, Hervé G (2004) Biochemical and enzymological aspects of the symbiosis between
the deep-sea tubeworm Riftia pachyptila and its bacterial endosymbiont. Eur J Biochem 271
(15):3093–3102. doi:10.1111/j.1432-1033.2004.04248.x. PMID 15265029
13. Newton DA, Rao KM, Dluhy RA, Baatz JE (2006) Hemoglobin is expressed by alveolar
epithelial cells. J Biol Chem 281(9):5668–5676. doi:10.1074/jbc.M509314200. PMID 16407281
14. Marik PE, Corwin HL (2008) Efficacy of red blood cell transfusion in the critically ill: a
systematic review of the literature. Crit Care Med 36:2667–2674
15. Charles P, Elliot P (1995) Handbook of biological effects of electromagnetic fields. CRC
Press, New York
16. Sang V (2010) The use of the mechanical fragility test in evaluating sublethal RBC injury
during storage 99(4):325–331
17. Sadhu PK, Mukherjee SK, Chakrabarti RN, Chowdhury SP, Karan BM, Gupta RK, Reddy
CVSC (2002) High efficient contamination free clean heat production. Indian J Eng Mater Sci
9:172–176
18. Sadhu PK, Pal N, Bhattacharya A (2013) Design of working coil using Litz Wire for industrial
induction heater. Lap Lambert Academic Publishing, ISBN:978-3-659 -35853-1, pp 1–65
19. Inayathullaah MA, Anita R et al (2010) Single phase high frequency Ac converter for
induction heating application. Int J Eng Sci Technol 2(12):7191–7197
20. Kotsuka Y, Hankui E, Shigematsu E (1996) Development of ferrite core applicator system for
deep-induction Hyperthermia. IEEE Trans Microw Theor Tech 44(10):1803–1810
21. Chen H, Ikeda-Saito M, Shaik S (2008) Nature of the Fe-O2 bonding in oxy-myoglobin: effect of
the protein. J Am Chem Soc 130(44):14778–14790. doi:10.1021/ja805434m. PMID 18847206
22. Hohn L, Schweizer A, Kalangos A et al (1998) Benefits of intraoperative skin surface
warming in cardiac surgical patients. Br J Anaesth 80:318–323
23. Kotsuka Y et al (2000) Development of inductive regional heating system for breast
Hyperthermia. IEEE Trans Microw Theor Tech 48(2):1807, 1813
24. Burdio JM, Fernando M, Garcia JR, Barragan LA, Abelardo M (2005) A two-output series-
resonant inverter for induction-heating cooking appliances. IEEE Trans Power Electron 20
(4):815–822
25. Rajek A, Lenhardt R, Sessler DI et al (1998) Tissue heat content and distribution during and
after cardiopulmonary bypass at 31 °C and 27 °C. Anesthesiology 88:1511–1518
Real Time Monitoring of Arterial Pulse
Waveform Parameters Using Low Cost,
Non-invasive Force Transducer

S. Aditya and V. Harish

Abstract Cardiovascular disease is currently the biggest single cause of mortality in

the developed world (Alty, IEEE Trans Biomed Eng 54:2268–2275, 2007) [1],
(Clerk Maxwell, A treatise on electricity and magnetism, Clarendon, 1892) [2].
Hence, the early detection of its onset is vital for effective prevention. Aortic stiffness
as measured by aortic Pulse wave velocity (PWV) has been shown to be an important
predictor of Cardiovascular disease. However, the measurement of the same is
complex and time consuming (Alty, IEEE Trans Biomed Eng 54:2268–2275, 2007)
[1], (Clerk Maxwell, A treatise on electricity and magnetism, Clarendon, 1892) [2].
This paper presents a simple, low-cost, speedy and non-invasive method using Force
Sensing Resistors (FSR) strategically placed over the Carotid and Radial arteries to
evaluate various arterial wave pulse parameters like heart-rate, Stiffness Index (SI),
Reflectivity Index (RI) and Pulse Wave velocity. The pulse rate and shape is used as
an estimate of heart-rate. This is used for diagnosis of arrhythmias, tachycardia and
bradycardia. The proposed method could be employed as a cheap and effective
Cardiovascular disease screening technique, to be later integrated with small wrist
watch-like monitors for suitable commercial purposes.

Keywords Cardiovascular disease (CVD) Pulse wave velocity (PWV) Stiffness





index (SI) Reflectivity index (RI) Arterial wave pulse Force sensing resistor
(FSR)

S. Aditya (&)
Department of Electronics, Electrical and Instrumentation, BITS Pilani, Pilani, Goa, India
e-mail: [email protected]
V. Harish
Department of Electronics and Instrumentation, Madras Institute of Technology,
Chennai, India
e-mail: [email protected]

© Springer India 2015 109

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_11
110 S. Aditya and V. Harish

Radial Artery In human anatomy, the radial artery is the main artery of
the lateral aspect of the forearm
Carotid Artery In human anatomy, the left and right carotid arteries the
head and neck with oxygenated blood
Arteriosclerosis The thickening, hardening and loss of elasticity of the
walls of arteries
Atherosclerosis A specific form of arteriosclerosis in which an artery wall
thickens as a result of invasion and accumulation of white
blood cells
Hemodynamic Relating to the flow of blood within the organs and tissues
of the body
Systolic Hypertension Refers to elevated systolic blood pressure

1 Introduction

Cardiovascular Disease (CVD) is the leading cause of mortality in the developed

world. An estimated 17 million people die every year from CVD (mainly from
myocardial infarction and stroke; source: World Health Organization). Stiffness Index
and Pulse Wave velocity are an important parameters used in the assessment of risk of
CVD. Arterial stiffening leads to systolic hypertension and increased load on the heart.
Arterial stiffness is not only a marker of the effects of atherosclerosis/arteriosclerosis
on the arterial wall, but in itself lead to adverse hemodynamic effects that increase
CVD risk [1, 2]. Studies performed in [3, 4] indicated large artery stiffness, as mea-
sured by PWV has proven to be a powerful predictor of CVD events [4]. Pulse wave
velocity was also recognized by the European Society of Hypertension as integral to
the diagnosis and treatment of hypertension [5]. Thus it is important to estimate these
biological parameters for early diagnosis and prevention of CVD.
Through the proposed method the Arterial Pulse Waveform and critical
parameters related to it such as heart rate, SI and PWV can be extracted. The current
method of measuring the arterial pulse waveform parameters is through Photopl-
ethysmography and analysis of the obtained Digital Volume Pulse (DVP) [1, 2].
Other methods of measuring PWV use invasive catheters [3] or mechanical
tonometers [6] to measure the transit time between pressure waves at two different
points. Some of these sensors are not only invasive but also expensive. The pro-
posed method would greatly simplify measurement of the arterial wave pulse and
its parameters using a low-cost and non-invasive transducer.
10 Consenting test subjects participated in the study. 2 Healthy males in the age
group of 20–30 and two males in the age group of 40–50 one healthy and the other
a patient and two males in the age group of 70–80 one healthy and the other a
patient. 4 females participated in the study 2 healthy in the age group of 40–50 and
2 in the age group of 70–80, one healthy and the other a patient.
Real Time Monitoring of Arterial Pulse Waveform … 111

Fig. 1 Block diagram of proposed system

Firstly, from the arterial wave pulse, Heart Rate of a patient can be estimated
through the proposed method. The count of number of pulses with a given time
interval gives heart rate information. Diagnosis of Tachycardia and Bradycardia can
also be performed using heart-rate information. Through the shape of the wave
pulse, Stiffness Index and Reflectivity index are estimated. Pulse Wave Velocity is
measured using 2 FSRs placed appropriately over the Carotid artery. The integrity
of the system is verified through comparison with PPG Analysis of the same test
subject. The variation of PWV and SI estimated using the proposed method with
age has also been measured. The working of the system is as described in the
diagram below (Fig. 1).

2 Methodology

2.1 Signal Conversion and Conditioning

2.1.1 Signal Sensing

During these measurements the patient is requested to remain still to avoid errors
due to motion artifact although a compensation has been provided for these in the
proposed algorithm later on. Force Sensing Resistors, are robust polymer thick film
devices that exhibit a decrease in resistance with increase in force applied to the
surface of the sensor. A standard Interlink Electronics FSR 402 sensor, a round
sensor 18.28 mm in diameter is used to sense the bio signal. The FSR is placed over
112 S. Aditya and V. Harish

Fig. 2 Signal conversion circuit with V+ = 10 V, RM = 120 KΩ

the carotid (over the wrist) or radial artery (over the neck region). The FSR’s
terminals are connected to a circuit which performs signal conversion. The circuit in
Fig. 2 Performs this function. As the force experienced by the FSR increases, its
resistance decreases consequently the voltage across RM increases and since the
amplifier is connected in a buffer configuration, its output voltage increases.
Voltage V+ was chosen to be 10 V and split power supply of ±15 V is provided to
the amplifier circuit.

2.1.2 Signal Conditioning

The typical human heart beats at about a minimum 60 up to a maximum of

300 beats/min which corresponds to a frequency range of 1–5 Hz. The output
voltage signal from the circuit in Fig. 2 is required to be conditioned to extract the
signal of desired frequency. This is done using a Simple filter circuit as shown in
Fig. 3. This signal is filtered using a band pass filter of 1–5 Hz. Although this
filtering should give us the signal in the desired frequency, we also use a Notch
Filter as a precaution to eliminate power line noise. Interference from power lines
(50 Hz) is the largest source of extraneous noise in bio-electric signals. Since the
device is to be used in environments such as homes and hospitals where power line
interference is unavoidable, there is further a need to eliminate distortions due it.
Thus the signal is also passed through a notch filter of Bandwidth 40–60 Hz to
specially eliminate power line 50 Hz noise. The frequency response of the filter

Fig. 3 Signal conditioning circuit with band pass: LPF R1 = 1.5 KΩ C1 = 22 µF and HPF
R2 = 8.3 KΩ C2 = 22 µF and Notch: LPF R3 = 180 Ω C3 = 22 µF and HPF R4 = 120 Ω
C4 = 22 µF and R7 = R8 = R9 = R10 = R11 = 10 KΩ
Real Time Monitoring of Arterial Pulse Waveform … 113

Fig. 4 Frequency response of filter as simulated using Pspice

Fig. 5 Smoothened
waveform at the output of the
circuit as seen on oscilloscope

circuit is as shown in Fig. 4. Y axis is magnitude response in dB and X axis

frequencies from 0 to 60 Hz. Simulations were carried out in Pspice.
From the frequency response plot of the signal it in the desired frequency range
(1–5 Hz) is attenuated by only 2.3 dB as indicated in Fig. 4 (yellow lines) while
power line noise frequency (50 Hz) is attenuated by about 23 db (Red Line) hence
being able to extract the signal of desired frequency. The output of the circuit is
observed using an oscilloscope and after smoothing a waveform is obtained as
shown in Fig. 5.

2.2 Sampling and Data Acquisition

The analog output of this circuit is required to be converted to digital and the
voltage values are required to be sampled to process the waveform data. This is
done using the NI- My DAQ Data acquisition system. The output of the signal
conditioning circuit Fig. 3. Is connected to analog input channel 1 of the My DAQ
and its USB interface with a laptop allows data to be acquired. The DAQ assistant
in LabVIEW is then configured to acquire the data. A correctly acquired heart beat
signal can have a maximum frequency of 5 Hz and hence by Nyquist Criterion,
114 S. Aditya and V. Harish

Fig. 6 Noisy waveform as acquired using DAQ assistant

sampling should be done at at least 10 Hz. A large sampling rate of 1 K sample/s is

chosen to acquire large number of samples and for a more accurate digital recon-
struction of the waveform. The reconstructed waveform acquired through DAQ
assistant can be seen in Fig. 6.

2.3 Assessment of Heart-Rate

For calculation of parameters, the noisy waveform obtained as shown in Fig. 6 is

difficult to work with. Thus Digital Signal Processing is required. First 6 Hz
low pass filtering of this data is done using signal processing tool in LabVIEW.
This filtered output is given as an input to the Math-Script Node in LabVIEW. The
MATLAB code processes the filtered waveform at the input and extracts the
necessary information. Now that a smoother waveform was obtained, it is required
to count the number of pulses. A simple method to do this would be counting the
number of peaks within a given time interval. The following counting method is
followed. The derivative of the digitally reconstructed signal is taken to find out the
maxima points. Now since the arterial wave pulse by nature is a double-peaking
pulse, i.e. it has a end of systolic peak and end of systole peak. Let the number of
maxima be N. Thus the number of pulses can be found as

No of Pulses ¼ N=2

where N is the number of maxima points.

The heart rate in BPM would be = N/2 * 60/5 = 6 N (bpm)
In the waveform in Fig. 7, the number of maximas = 11. Thus heart
rate = 66 bpm which is the correct heart rate. In some cases however due to artifacts
Real Time Monitoring of Arterial Pulse Waveform … 115

Fig. 7 Arterial pulse waveform as seen MATLAB

Fig. 8 Erroneous waveform obtained as seen in MATLAB

waveforms like those in Fig. 8 are obtained. In Fig. 8 there are more than 2 maxima
per pulse. This will lead to incorrect estimation of heart rate. The same code when
run for the waveform in Fig. 8 counts 16 maxima, i.e. 6 × 16 = 96 bpm. However
the actual heart rate is 72 bpm.
A compensation for such artifacts is provided by checking closeness of maxima
points in time. If two maxima points are closer than 100 ms apart in time, then there
is highly likely to be an error and the code provides a compensation for the same,
by ignoring such maxima while counting. Hence miscounting of maxima is avoided
using this method.
116 S. Aditya and V. Harish

Artifacts due to Power Line interference have been removed by the notch filtering
performed earlier. Even after compensation some errors due to patient movements
and irregular breathing may still exist as shown in Fig. 8. High frequency artifacts
caused by sudden movements or spike in circuit voltage won’t cause a problem as
the signal is already pre-filtered. Detection and removal of artifacts requires further
analysis. Although artifact detection and compensations is not covered in the paper,
this can also be performed through extensive data collection, signal processing,
feature extraction and learning machines such as SVMs or ANNs.

2.4 Assessment of Stiffness Index

Stiffness is defined as the ratio of the height of a subject (h) to the time difference (ΔT)
or Peak to Peak time (PPT) between systolic peak and end of systole peak [1, 2].
This information is usually obtained from the Digital Volume Pulse (DVP) measured
using Photoplethysmography. However a simpler and low-cost method to obtain the
same information is from the waveform obtained using an FSR.
To find arterial stiffness index the output of the DAQ assistant is 6 Hz low-pass
filtered in LabVIEW. The data points corresponding to maxima and minima are found.
Then the difference between consecutive maxima points corresponding to the Systole
peak and End of Systole peak is found. This gives us ΔT or PPT. This done for all the
pulses in the 5 s interval and the average value is recorded. The height of the peaks ‘a’
and ‘b’ are calculated with respect to the closest previous minima. The height of the
subject is recorded previous to the experiment using a simple height- measuring scale.
After inputting the subject’s height, the arterial stiffness can be calculated.
Reflectivity index is as defined the ratio of height of End of Systole peak to
Systole Peak or b/a, as is seen from Fig. 9. To obtain this information, the same
code previously used is modified to obtain the values of the waveform peaks and
the bottom. Then the ratio of the difference is found and calculated.

2.5 Assessment of Pulse Wave Velocity

Pulse Wave velocity is also another important indicator of arterial stiffness. For
calculation of Pulse wave velocity, the arterial pulse waveform is measured at two
points about 5 cm apart over the carotid artery simultaneously. Two FSR’s were
used and analog and digitally conditioned in the same manner as done previously.
The outputs of the circuit were given to channel 0 and channel 1 of the Ni-My
DAQ. The data was acquired at the same sampling rate of 1 K sample/s and 5 s. The
signals are low pass filtered at 6 Hz in LabVIEW and then fed to a math script node.
The Algorithm devised basically extracts the data samples, corresponding to the
systolic maxima (upper-peak) of the waveform. The time value of data sample
corresponding to the first peak that occurs of each waveform is stored. Let us
Real Time Monitoring of Arterial Pulse Waveform … 117

Fig. 9 Estimation of SI and

RI from arterial pulse wave

assume that measurements start from time t = 0 and the first maximas occur at time
t1 and t2 respectively. Thus |t1–t2| would give us the time difference. Now that the
time difference is known and the distance between two points is known,

PWV ¼ Distance (5 cm)=Time Difference.

Although the above measurements could be performed by placing the sensor

over two points over the radial artery, in case of the radial artery, the sensitivity was
found to be far less than when placed over the carotid artery.

3 Results and Statistical Analysis

The table below presents the average value of different biological parameters- heart
rate, Stiffness Index and Reflectivity Index obtained for different test subjects in 10
estimation trials.
From the data in Table 1, a graph of Stiffness Index versus Age is plotted. This
can be seen in Fig. 10.
The best fit line for the plot gives linear equation SI = 0.08805 * Age + 4.2997.
Here we observe a positive correlation between Stiffness Index and age, i.e. as the
age of the test subject increases, the stiffness index increases. This is in close
agreement to the results presented in [7]. Mean Arterial Pressure (MAP) is another
variable which affects the Stiffness Index, measuring Blood pressure is not in the
scope of the proposed hence is neglected while performing regression analysis.
118 S. Aditya and V. Harish

Table 1 Estimated value of Biological Parameters for test subjects using the proposed method
Test Gender Age Heart rate Stiffness index Reflectivity Pulse wave velocity
subject (bpm) (SI) (m/s) index (RI) (PWV) (m/s)
I Male 21 74.2 5.93 0.47 7.41
II Male 22 68.3 6.22 0.53 7.50
III Male 45 80.5 7.97 0.38 9.32
IV Male 48 92.4 8.69 0.54 9.65
V Male 67 84.7 10.15 0.46 11.30
VI Male 65 71.8 9.83 0.56 11.43
VII Female 42 76.6 8.32 0.72 9.28
VIII Female 45 83.1 8.54 0.45 9.36
IX Female 63 80.8 9.84 0.84 10.76
X Female 68 70.7 10.28 0.67 11.25

Mean of the Stiffness Index = 8.57 m/s and Standard Deviation (SD) = 1.547 m/s.
The mean heart rate for the 10 test subjects was found to be 78.3 bpm with
SD = 7.4058 bpm. Range of heart rate = 24.1 bpm.
Table 1 also presents Pulse Wave velocity of different test subjects. Regression
analysis of Pulse Wave velocity versus age obtained using the proposed method,
yields best fine line y = 0.0843 * Age + 0.527. This is also in agreement with results
obtained in [7]. The mean PWV was found to be 9.726 m/s and SD = 1.475 m/s. A
better sensitivity was observed when the FSR was placed over the carotid artery
than the radial artery during Pulse Wave Velocity Measurement.

Fig. 10 A 6 Hz low pass filtered waveform used for estimation of Reflectivity index (RI) and
Stiffness in dex (SI)
Real Time Monitoring of Arterial Pulse Waveform … 119

Table 2 Comparison of
Biological parameters from different sensors
Biological Parameters
obtained from proposed Parameter FSR sensor PPG signal
method and Photop- Heart rate 78.3 BPM 78.3 BPM
lethsymography Peak to peak time 376 ms 364 ms
Stiffness index 8.57 m/s 8.85 m/s
Reflectivity index 0.562 0.546

4 Photoplethysmography Analysis

To validate the results obtained using the existing sensor, Photoplethysmogram

measurements were performed simultaneously on the same test subject. The stan-
dard PPG probe was placed on the finger tip of the test subject and the arterial pulse
waveform was obtained from PPG signal acquired in LabVIEW and imported. This
data is processed using the same algorithm as before after making some changes
with respect to sampling rate.
A comparison of the average value parameters obtained for the 10 test subjects
from PPG analysis of all test subjects and the proposed method is presented in the
Table 2.
Analysis shows that there is a 3.3 % error in values of Peak to Peak Time and
Stiffness Index, and 2.84 % error in values of reflectivity index obtained using FSR
sensor and from PPG signal processing. Hence the parameters are in agreement and
can be estimated with reasonably high accuracy.

5 Discussion: Hardware and Software Integration

After processing the signal and finding techniques to accurately extract information
about these biological parameters, a Real time monitoring System was built in
LabVIEW which monitors these parameters every 5 s. Further based on the value of
these parameters, and a diagnosis of medical conditions such as arrhythmias,
Tachycardia and Bradycardia and risk of cardiovascular disease based on Pulse
Wave Velocity and Stiffness Index is done. Hear rate in the range of 60–100 bpm
are classified as normal. Heart rates below 60 bpm are classified as bradycardia and
those over 100 bpm are classified as tachycardia. Pulse wave velocities above 10 m/s
are classified as high risk of CVD. The front-panel View of the integrated system is
shown in Fig. 11. The final system comprises of a simple User friendly GUI in
LabVIEW where the user simply has to enter his/her enter age and height based on
which, his/her heart rate, pulse wave velocity, stiffness index and reflectivity will be
displayed every 5 s. The transducer is also fast in response and negligible delays are
observed during measurements An averaging feature is also introduced in the system
which calculates, the average heart rate per minute. The user can know about his/her
risk of CVD or Arrhythmias. Alarms are also introduced in case of abnormally high
heart rates or PWVs (Figs. 12 and 13).
120 S. Aditya and V. Harish

Fig. 11 Arterial pulse waveforms obtained 5 cm apart above the carotid artery

Fig. 12 Arterial pulse waveforms obtained 5 cm apart above the carotid artery

Fig. 13 Variation of Stiffness Index (left) and Pulse Wave Velocity (right) obtained using the
proposed method with age
Real Time Monitoring of Arterial Pulse Waveform … 121

Fig. 14 Acquisition of PPG signal

Fig. 15 Front panel view of integrated system

6 Calibration

Since the proposed system is interested in measurement of parameters (hear rate, SI,
PWV) that are sensitive to the time variation of the signal as opposed to amplitude,
thus a calibration of amplitudes is not required. Even RI remain unaffected as it is
122 S. Aditya and V. Harish

simply an amplitude ratio. However an Omp amp power supply of ±15 V and a
10 V for V+ supply to the signal conditioning circuit will yield proper results.
Power supply voltage can be increased, for bettering the sensitivity of the circuit
however it should be kept within safe limits as IC 741 cannot withstand very high
input voltages. Also the height of the test subject measured accurately at the time of
measurement must be given as input for correct estimation of Stiffness Index
(Figs. 14 and 15).

7 Conclusion

Accurate estimation of heart-rate, Stiffness Index and Pulse Wave velocity was
done using a low-cost, speedy and non invasive FSR. Variation of Stiffness Index
and Pulse Wave Velocity with age, estimated using the proposed method is also
performed and the results were found to be in agreement to those obtained using
previous methods. Offline monitoring of these parameters can be performed using
TDMS logging option available in the DAQ assistant and the same algorithm can
be implemented again to extract these parameters. Although the system proposed
here uses a lot of hardware, it can easily be integrated into small wrist watch like
monitor with FSR sensors strapped to the bottom of the device. Small DSP Pro-
cessors/Microcontrollers can be used replaced MyDAQ and LabVIEW to perform
the processing functions and power supplies for hardware can easily be made
portable through batteries. We also hope that the affordability and simplicity of the
proposed system will encourage more people to use the instrument and help them
lead a safe and healthy life.

References

1. Alty SR, Angarita-Jaimes N, Millasseau SC, Chowienczyk PJ (2007) Predicting arterial

stiffness from the digital volume pulse waveform. IEEE Trans Biomed Eng 54(12):2268–2275
2. Clerk Maxwell J (1892) A treatise on electricity and magnetism, 3rd edn, vol 2. Clarendon,
Oxford, pp 68–73
3. Blacher J, Asmar R, Djane S, Gerard M (1999) Aortic pulse wave velocity as a marker of
cardiovascular risk in hypertensive patients. Hypertension 33:1111–1117
4. Boutouyrie P et al (2002) Aortic stiffness is an independent predictor of primary coronary
events in hypertensive patients: a longitudinal study. Hypertension 39:10–15
5. Mancia G, de Backer G, Dominiczak A et al (2007) Guidelines for the management of arterial
Hypertension: the task force for the management of arterial Hypertension of the European
Society of Hypertension (ESH) and of the European Society of Cardiology (ESC). J Hypertens
25(6):1105–1187
Real Time Monitoring of Arterial Pulse Waveform … 123

6. Salvia P, Liob G, Labatc C, Riccid E, Panniere B, Benetosc A (2004) Validation of a new non-
invasive portable tonometer for determining arterial pressure wave and pulse wave velocity: the
Pulse Pen device. J Hypertens 22:2285–2293
7. Millasseau SC, Kelly RP, Ritter JM, Chowienczyk PJ (2002) Determination of age-related
increases in large artery stiffness by digital pulse contour analysis. Department of Clinical
Pharmacology, St. Thomas’ Hospital, Centre for Cardiovascular Biology and Medicine, King’s
College London SE1 7EH, UK, Clinical Science
Selection of Relevant Features
from Cognitive EEG Signals Using ReliefF
and MRMR Algorithm

Ankita Mazumder, Poulami Ghosh, Anwesha Khasnobish,

Saugat Bhattacharyya and D.N. Tibarewala

Abstract Cognition may be defined as a set of mental activities or processes which

deals with knowledge, attention, memory and working memory, reasoning and
computation, and judgement and evaluation. In this paper, we aim to study two
distinctive cognitive processes dealing with evaluation of two similar stimuli and
reasoning and computation of some mathematical problem. Here, we have used
Wavelet Transforms and Distance Likelihood Ratio Test for feature extraction and
classification respectively. We have also used two feature selection algorithm: Re-
liefF and Minimum Redundancy Maximum Relevance to select only the best relevant
features for classification. The results show a 15 % improvement on accuracy when
feature selection algorithms are used in the process. The results also suggests that the
brain activation is dominant at the frontal, parietal and temporal region.

Keywords Cognition Electroencephalography ReliefF Minimum redundancy


maximum relevance Feature selection

A. Mazumder (&)
P. Ghosh
A. Khasnobish
S. Bhattacharyya
D.N. Tibarewala

School of Bioscience and Engineering, Jadavpur University, Kolkata, India
e-mail: [email protected]
P. Ghosh
e-mail: [email protected]
A. Khasnobish
e-mail: [email protected]
S. Bhattacharyya
e-mail: [email protected]
D.N. Tibarewala
e-mail: [email protected]

© Springer India 2015 125

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_12
126 A. Mazumder et al.

1 Introduction

The entire human brain is divided into number of regions and each of these regions
has separate sets of functions. Cognition is one of these functions which is basically
related to tasks like decision making, memorization, perception, consciousness and
likewise [1]. This study is mainly focused on the cognitive capabilities, or to be
more precise, on the problem-solving capabilities of the human brain. It has been
confirmed in a number of previous studies that the bio-signals generated while
performing the cognitive tasks fall within the alpha (α) and beta (β) frequency bands
[2], which originates from the parietal and temporal regions of the brain. It is also
be concluded from literature that the frontal lobe takes a major part in the process of
cognition [3]. So, it can be said that the activation will take place mostly in these
three regions while performing any cognitive task.
The ability of the human brain can get deteriorated due to the advent of number
of diseases such as Parkinson’s disease, Alzheimer’s disease, stroke, multiple
sclerosis, Lupus, severe brain injury and many more [4, 5]. Through advances in
neuroscience and computing, it is now possible to provide such patients with
rehabilitative treatments. A number of different methods for cognitive rehabilitation
have been devised over the years in form of Brain Computer Interface (BCI)
technologies. The main aim of the BCI system is to decode the signals acquired
from human brain and then decode those into signals which can be used for con-
trolling a device (for example a rehabilitative aid and likewise). The principal
components of the BCI technologies are feature extraction [6], selection of the
features [7] of interest and finally classification [8] of those signals.
The outputs of the BCI system can be optimized by selecting a proper combination
of these algorithms. The feature extraction and classification stages are mandatory for
any BCI application. The feature selection stage is optional but it is important for
obtaining precise and error free outputs for a number of applications. Sometimes, it is
noticed that the performance of the BCI system is affected because of the high
dimensionality of the feature vector. Because of the presence of large number of
irrelevant features (features which are not discriminable among classes), the effect of
relevant features are negated which reduces the performance of the BCI [9].
In this paper we aim to study the performance of ReliefF [10], and Minimum
Redundancy Maximum Relevance (MRMR) [11] for feature selection of the best
relevant features in a cognitive task experiment and its effect on the performance of
the classifier. The cognitive task experiment comprises two separate assignments.
First, the subject performs a task where he/she has to spot the difference between
two similar pictures. This corresponds to the evaluation related cognitive process.
The second task presents a mathematical puzzle to the subject, who must solve it in
a given time period. This corresponds to the reasoning and computation cognition.
The rest of the paper is organized as follows: Sect. 2 provides a description on the
experimental methods employed in this study. It also gives a brief description on the
feature extractor, feature selector and classifier algorithm. Section 3 presents the results
produced by this experiment and the concluding remarks are mentioned in Sect. 4.
Selection of Relevant Features … 127

2 Materials and Methods

In this experiment, the subjects are instructed to spot the difference between two
sets of ‘look alike’ pictures, and solve some mathematical tasks (as shown in
Fig. 1). Through these tasks, we aim to understand the underlying processes taking
place in the brain when the subjects perform these tasks. 7 healthy subjects (4
female and 3 male) in the age group of 22–28 years have participated in this
experiment. The EEG signals from the subjects were recorded using a 19 channel
EEG amplifier (NeuroWin, Make-NASAN). Based on the nature of the experiment,
we have selected the electrodes: F3, F4, Fz, P3, P4, T3, and T4 for our study as
because these electrode locations coincide with those regions of the brain which are
responsible for such cognitive tasks, which are the parietal, frontal and temporal
lobe. After that these data are filtered using a band-pass filter. The next step is the
extraction of features which is performed by Wavelet Transform [12]. Features
pertaining to wavelet transforms yields high dimensional features which is suitable
for this paper. The next step is the feature selection step which uses ReliefF and
Minimum Redundancy Maximum Relevance (MRMR) to select the best N features
from the original feature set. After that the selected features are classified using
Distance Likelihood Ratio Test (DLRT) [13]. The same steps are performed again
but on the second trial the feature selection step is excluded. The results suggest that
the performance of the classifier is better while incorporating the feature selection
step than without it.

Fig. 1 A example of a portion of the timing sequence of the visual stimuli

128 A. Mazumder et al.

2.1 Design of the Visual Stimuli

The comprehensive format for the visual cue is given below: For the first 5 s of a
session, the subject is made to relax during which the baseline EEG of the subject is
recorded. Then for 30 s a set of two ‘look alike’ pictures which has only one single
difference, are shown to the subject and he is required to identify that particular
difference. For the next 40 s a mathematical puzzle appears in front of the subject
and he is asked to solve that. Then again a set of picture appears for 30 s and so on.
In between each of the subsequent task slides there is a blank slide of 5 s during
which the subject is asked to answer. This collective set of 80 s (30 + 5 + 40 + 5) is
repeated for 5 times in the experiment with different pictures and mathematical
puzzles in each set. A visual cue for a single set is shown in Fig. 1.

2.2 Filtering the EEG Signal

It is known from standard literature that cognitive signals are dominant in the α (8–
12 Hz) and β (16–30 Hz) band. Thus, for this study, we have designed an IIR
elliptical filter of bandwidth 8–30 Hz to filter the EEG signals acquired from the
amplifier. The selection of an elliptical filter was made on the basis that it possesses
good frequency domain characteristics with sharp roll off and also has good
attenuation of the pass- and stop-band ripples.

2.3 Feature Extraction: Wavelet Transforms

In this experiment the preprocessing of the raw data consists of three stages, of
which the first one is feature extraction. Various different algorithms can be applied
on the raw data set to perform feature extraction. Here, in this experiment we have
used discrete wavelet transform algorithm.
Wavelet transform is basically a time-frequency domain technique and hence is
found to have more advantages over other time domain techniques or frequency
domain techniques, as unlike wavelet transform, these techniques lack the in-
formation about the other domain. Besides, frequency based techniques like Fourier
transform, are not good at dealing EEG signals, as they are non-stationary signals.
But Discrete Wavelet Transform (DWT) can combine information of both time and
frequency domains and at a given instant of time, can also provide localized
information related to frequency domain [12]. The energy distribution [14] equation
for discrete wavelet transform is given as
!
1X 2 1 X 2 XJ
1 X 2
j f ðtÞj ¼ jaJ ðkÞj þ jdJ ðkÞj ð1Þ
N t NJ k J¼1
NJ k
Selection of Relevant Features … 129

Fig. 2 The decomposition process in a discrete wavelet transform

Using (1) the features of power distribution of signals can be extracted. In DWT
features are extracted by decomposing the input signals into two halves. This is
done at every single level using two digital filters, a low pass and a high pass, as
shown in Fig. 2. In this experiment a sampling frequency of 250 Hz is used and our
required frequency band from which the signals are extracted, is of range 8–30 Hz.
Hence, in order to achieve this particular frequency band, the signal has to be
decomposes for 5 levels. As a single level passes by, the signal got divided into CAi
(coarse approximation) and DIi (detailed information). The CAi obtained from the
low pass filter are further decomposed to get the subsequent levels. There are
different mother (base) wavelets available, from which the Daubechies wavelet
(db4) of fourth order is chosen. The outputs of levels 4 and 5 are selected post
decomposition, as the desired frequency band (8–30 Hz) lies in these levels. The
final dimensions of the feature vector is 7 electrodes × 35 features = 245 features.

2.4 Feature Selection

In the world of BCI, feature selection is considered to be an important step,

especially while dealing with high dimensional features. The main goal is the
reduction of the feature space dimension so as to get the most effective features. It
so happens often that huge datasets contain a lot of irrelevant and ambiguous
feature data that ultimately result in unnecessary computation complexity and
130 A. Mazumder et al.

lowering of classification accuracy. In this experiment, we have used two feature

selection algorithms. The first one is ReliefF and the second one is Maximum
Relevance Minimum Redundancy (MRMR).

2.4.1 ReliefF Feature Selector

The ReliefF algorithm [10] is a modification of the Relief algorithm. In this

algorithm, the qualities of the attributes are estimated depending on how accurately
they can distinguish between the values lying in their vicinity. In order to achieve
that, the algorithm has to undergo two steps. Let R be an instance. In both of the
steps the algorithms tries to find out the k nearest neighbors of R but the main
difference is that, at the first step, the search is performed within the same class and
in the next step, the it is done by taking all the classes into consideration. The
nearest neighbors found in both the first and the second steps are known as nearHits
and nearMisses respectively. Suppose, X is an attribute and its quality estimate is
given by W [X]. Then W [X] will be updated depending on both nearHits and
nearMisses. This entire process is repeated I times so that a good estimate of the
weights can be obtained. The algorithm of the complete process is given below.

2.4.2 MRMR Feature Selector

In an unsupervised situation where the classifiers are not specified, minimal error
requires the maximum statistical dependency of the target class c on the data
distribution in the subspace Rm of m features. This scheme is called the maximal
dependency (Max-Dependency). The most popular approach to realize Max-
Dependency is maximal relevance (Max-Relevance). Some researchers have also
Selection of Relevant Features … 131

employed minimum redundancy (Min-Redundancy) to reduce the redundancy of

the feature set.
In terms of mutual information, feature selection algorithm aims to find a feature
set S with m features {xi}, which jointly have the largest dependency on the target
class c,

maxDðS; cÞ ¼ Iðxi ; i ¼ 1; . . .; m; cÞ ð2Þ

Max-Relevance searches features satisfying (3), which approximates D(S, c) in

(2) with the mean value of all mutual information values between individual fea-
tures xi and c:

1 X
maxDðS; cÞ; D ¼ Iðxi ; cÞ ð3Þ
j Sj x 2 S
i

It is likely that the features selected would have rich redundancy (dependency
among the features are large). Thus, when two features are highly dependent on
each other, the respective class discriminative power would not change much if one
of them were removed. Therefore, Min-Redundancy is applied to select mutually
exclusive features:

1 X
maxRðSÞ; R ¼ Iðxi ; xj Þ ð4Þ
jSj2 xi ;xj 2 S

The criterion combining the above two constraints is called “minimal-redundancy-

maximal-relevance” (MRMR).
Thus, the condition for any search algorithm applying this method is

1 X
max½Iðxj ; cÞ  Iðxj ; xi Þ ð5Þ
m  1 x 2S
i m1

The computational complexity of this incremental search method is O(|S|.M),

where S is the dimension of the feature set and M is the number of features. Also, a
theorem exist which states that the first order incremental search, mRmR is
equivalent to Max-Dependency.

2.5 Classifier: Distance Likelihood Ratio Test (DLRT)

The DLRT algorithm is a statistical tool for classification and quite efficiently used
for BCI applications. This classification algorithm is a modification of the k-NN
classifier and particularly suitable for the datasets whose feature distributions under
the different classes are well defined and known to the user. If the probability
132 A. Mazumder et al.

distribution of the features is not well defined then application of DLRT may render
a significant amount of error in the result. Non-parametric probability distributions
are used in order to avoid such situations. This classifier first estimates the class
conditional probability vector for the feature vector which is obtained from the
feature extraction stage. The estimate is obtained using the following formula

^pðxjHi Þ ¼ frack Ni Aðk; xÞ ð6Þ

where, k denotes the number of neighbors of x, Ni denotes the number of training

points and A denotes the volume of the feature space containing the nearest
^
neighbors (k). This equation is then used to calculate the likelihood ratio ðkðxÞÞ
which is given by,
     
nH0 Mkð0Þ nH0
log D ¼ log þ D logðlogðMkð0ÞÞ  logðMkð1ÞÞ ð7Þ
nH1 Mkð1Þ nH1

where logðMkð1Þ Þ and logðMkð0Þ Þ denote the log of the distances to the kth neighbor
and D denotes the dimensionality of the feature vector. This estimated logarithmic
ratio is then compared to the actual likelihood ratio λ(x) to verify whether the
outputs of the DLRT are clustered mostly around the true ratio, i.e. if the algorithm
has rendered satisfactory results [13].

3 Results and Discussions

In this study, the subject performed two separate cognitive tasks, viz, (i) spotting the
difference from two sets of look-alike pictures, and (ii) solving a mathematical
puzzle. EEG signals from 7 electrode locations are acquired for further analysis.
Seven subjects performed the experiments in 3 sessions and each session consists of
10 sets of cognitive tasks (5 spotting the difference and 5 mathematical tasks).
Wavelet Transform is used for feature extraction and the size of the original feature
vector is 245. After introducing ReliefF and MRMR algorithm, the size of the
feature vector is reduced to 5 best features. Following that, the reduced feature
vector is fed as input to the DLRT classifier and the results are given below.
First, we illustrate the activity of the brain when the subject performs cognitive
experiment. From Fig. 3 it is observed that the activation (as shown in red) occur
mostly in the frontal (component 2, 3, 4 and 5), parietal (component 5 and 6) and
temporal (component 7) regions. It is previously discussed, while performing any
cognitive task, the activation of a human brain takes place mostly in the frontal,
parietal and temporal lobes. This result, thus, validates our claim that the cognitive
tasks are dominant in the three regions of the brain.
This study also reveals that the use of feature selection method in signal pro-
cessing of EEG data, improves the results obtained. As it can be seen from Fig. 4
the accuracy increased almost 15 % on average after using feature selection. Same
Selection of Relevant Features … 133

Fig. 3 An example of the activation map of the brain during the cognitive task performed by
Subject 3. 7 samples or components (denoted by numbers: 1, 2, …, 7 in the figure) are considered
during the performance of the cognitive task. Red marks the maximum brain activation and blue
marks the minimum brain activation

Fig. 4 Classification Accuracy (CA) of seven subjects without using any feature selection
algorithm, ReliefF algorithm and MRMR algorithm
134 A. Mazumder et al.

Fig. 5 Area under the curve (AUC) of seven subjects without using any feature selection
algorithm, ReliefF algorithm and MRMR algorithm

is the result for area under the curve (AUC). In Fig. 5 it is evident that the AUC is
almost 20 % less when feature selection is not used.
The performance measures of ReliefF and MRMR methods of feature selection
are further validated by means of Friedmans Test [15]. We compare the perfor-
mance, in terms of classification accuracy and AUC, of the two algorithms with
other standard algorithms: Correlation Based Feature Selection (CFS), Principal
Component Analysis (PCA) and Minimal Redundancy (MR) algorithms. Friedman
test compares the relative performance of the two algorithms with 4 more feature
selection algorithms. The null of the algorithm states that the rank of all the
algorithms should be equal as all the algorithms. The Friedman statistic is given by
Eq. (8).
" #
12N X kðk þ 1Þ
v2F ¼ R 
2
ð8Þ
kðk þ 1Þ i i 4

It is distributed accordingly to v2F with 1 degrees of freedom, where k is the

number of algorithms to be compared and N is the number of parameters used for
comparison. In this study, the mean of the two metric: classification accuracy (CA)
and area under the curve (AUC), have been selected for evaluation, thus, k = 5 and
N = 2 and Table 1 is the ranking table for all the algorithms.
From Table 1, the value of Rj is calculated, which is further used to obtain
v2F ¼ 8 [ v25;0:95 . Hence it can be seen that the proposition of null hypothesis that all
the algorithms are equivalent is wrong and so the algorithm performances are
determined from their ranks itself. It is clear from the table that the rank of MRMR
is 1 and ReliefF is 2, claiming ReliefF and MRMR yields better results than its
competitors. Also, MRMR works slightly better than ReliefF.
Selection of Relevant Features … 135

Table 1 Comparison and ranking of five feature selection algorithms

Algorithms Mean CA Rank Mean AUC Rank Average rank
(%) CA (%) AUC (Rj)
ReliefF 85.39 2 78.12 2 2
MRMR 85.77 1 80.60 1 1
CFS 81.23 3 77.56 3 3
PCA 63.45 5 60.00 5 5
MD 78.67 4 75.00 4 5

4 Conclusion

The work presented here aims to study the different brain processes during two
different kinds of mental tasks, (i) while spotting the difference between two similar
pictures, and (ii) while performing some a mathematical puzzle. For this purpose,
we have used Wavelet Transforms for feature extraction and Distance Likelihood
Ratio Test as classifier. We have also explored the change in performance of the
classifier by introducing a feature selection step in between the feature extraction
step and classification step. Here, we have used ReliefF and MRMR algorithm for
this purpose. It is observed from the results that the accuracy of the classifiers
increased by II. 15 % on inclusion of the feature selection step. From Friedman
Test, it is observed that MRMR performs slightly better than ReliefF but both the
algorithms are much higher as compared to other algorithms, which are, CFS, PCA
and MD. From the brain activation maps, shown in results, it is noted that the
parietal, frontal and temporal regions are most active.

Acknowledgments The authors would like to thank Council of Scientific and Industrial
Research, India for their financial assistance.

References

1. Milner B, Squire LR, Kandel ER (1998) Cognitive neuroscience and the study of memory.
Neuron 20:445–468
2. Davis CE, Hauf JD, Wu DQ, Everhart DE (2011) Brain functionwith complex decision
making using electroencephalography. Int J Psychophysiol 79:175–183
3. Ramsey NF, van de Heuvel MP, Kho KH, Leijten FSS (2006) Towards human BCI
applications based on cognitive brain systems: an investigation of neura signals recorded from
the dorsolateral prefrontal cortex. IEEE Trans Neural Syst Rehabil Eng 14(2):214–217
4. Dauwels J, Vialatte F, Cichocki A (2010) Diagnosis of Alzheimer’s disease from EEG signals:
where are we standing? Curr Alzheimer Res 7(6):487–505
5. Giles GM, Radomski MV, Champagne T et al (2013) Cognition, cognitive rehabilitation, and
occupational performance. Am J Occup Ther 67:S9–S31
6. Cososchi S, Strungaru R, Ungureanu A, Ungureanu M (2006) EEG features extraction for motor
imagery. In: Proceedings of the 28th annual international conference IEEE engineering medicine
and biology society EMBS ’06, New York, USA, pp 1142–1145, 30 Aug–3 Sept, 2006
136 A. Mazumder et al.

7. Schroder M, Bogdan M, Hinterberger T, Birbaumer N (2003) Automated EEG feature

selection for brain computer interfaces. In: Proceedings of the 1st international IEEE EMBS
conference on neural engineering, pp 626–629
8. Lotte F, Congedo M, L’ecuyer A, Lamarche F, Arnaldi B (2007) A review of classification
Algorithms for EEG-based brain-computer interfaces. J Neural Eng 4(2):R1–R13
9. Koprinska I (2010) Feature selection for brain-computer interfaces. New Front Appl Data Min,
Lect Notes Comput Sci 5669:106–117
10. Megchelenbrik W (2010) Relief-Based feature selection in bioinformatics: detecting functional
specificity residues from multiple sequence Alignments. Radboud University, Nijmegen
11. Estevez PA, Tesmer M, Perez CA, Zurada JM (2009) Normalized mutual information feature
selection. IEEE Trans Neural Netw 20(2):189–201
12. Darvishi S, Al-Ani A (2007) Brain-computer interface analysis using continuous wavelet
transform and adaptive neuro-fuzzy classifier. In: Proceedings of the 29th international annual
conference in IEEE engineering, medicine and biology society, pp 3220–3223
13. Remus JJ, Morton KD, Torrione PA, Tantum SL, Collins LM (2008) Comparison of a
distance-based likelihood ratio test and k-nearest neighbor classification methods. In: IEEE
workshop on machine learning for signal processing MLSP 2008, pp 362–367
14. Bhattacharyya S, Rakshit P, Konar A, Tibarewala DN, Janarthanan R (2013) Feature selection
of motor imagery EEG signals using firefly temporal difference Q-Learning and support vector
machine. In: Panigrahi B, Suganthan PN, Das S, Dash SS (eds) Swarm, evolutionary, and
memetic, computing, lecture notes in computer science, vol 8298. Springer International
Publishing, Switzerland, pp 534–545
15. Bhattacharyya S, Konar A, Tibarewala DN (2014) A differential evolution based energy
trajectory planner for artificial limb control using motor imagery EEG signal. Biomed Signal
Process Control 11:107–113
Generalised Orthogonal Partial Directed
Coherence as a Measure of Neural
Information Flow During Meditation

Laxmi Shaw, Subodh Mishra and Aurobinda Routray

Abstract Neural information flow in brain during meditation, can be addressed by

brain connectivity studies. This work aims to obtain neural connectivity measures
based on a strictly causal time varying Multi-Variate Auto-Regressive (MVAR)
model, fitted to EEG signals obtained during meditation. The time varying Granger
Causality based connectivity estimators as PDC (Partial Directed Coherence), g-PDC
(generalized Partial Directed Coherence), OPDC (Orthogonalized Partial Directed
Coherence) and g-OPDC (generalized Orthogonalized Partial Directed Coherence)
are calculated using the adaptive autoregressive MVAR parameters. The MVAR
model parameters have been estimated by Kalman Filter algorithm. In this work
g-PDC and g-OPDC have been used to make the connectivity measures scale
invariant. These connectivity estimators quantify the neural information flow
between Electroencephalograph (EEG) channels. In addition g-OPDC is also immune
to volume conduction artifact and gives better result compared to g-PDC. Finally,
surrogate data statistics has been used to check the significance of the above con-
nectivity estimators.

Keyword Coherence EEG

Time varying auto regressive model
Connectivity

measures Kalman filter

L. Shaw (&)
Silicon Institute of Technology, Bhubaneswar, India
e-mail: [email protected]
S. Mishra
Pune, India
A. Routray
Indian Institute of Technology, Kharagpur, Kharagpur, India
e-mail: [email protected]

© Springer India 2015 137

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_13
138 L. Shaw et al.

1 Introduction

Frequency domain study of brain connectivity has assumed great importance in the
recent years. There are many conventional frequency domain estimators of
connectivity such as coherence and partial coherence for calculating coupling and
direct coupling respectively [1]. Coherence and partial coherence being symmetric,
give no information regarding the direction of information flow. The direction of
neural information flow helps in determining causality.
A time series (e.g.: a single EEG channel data) can be said to cause another series
when the information in the past of the former series can help in predicting the present
value of the latter series. This is based on the famous Granger Causality principle
which has been found to be very successful in econometric causality analysis.
Applications of Granger Causality (GC) are profoundly used in neuroscience.
Granger Causality based brain connectivity measures are Directed Transfer Func-
tion (DTF), Partial Directed Coherence (PDC) and generalized-Partial Directed
Coherence (g-PDC). They have also been established in [2–4], The generalized
PDC (or g-PDC) is a scale invariant version of PDC and it is immune to static gain.
The connectivity measures discussed above are essentially derived from a
strictly causal MVAR (Multi-Variate Auto-Regressive) model fitted to the multi-
channel EEG data [1–4].
The estimators are generally defined for stationary, time invariant linear signals.
This makes their application to EEG signals a challenging task because of EEG’s
nonlinearity and non-stationarity [5, 6]. To overcome this problem of non-stationa-
rity, a strictly causal time varying MVAR model has been used in [7−9]. In this paper,
we have considered only lagged effects; hence the proposed model is strictly causal
MVAR model. The effects of instantaneous or zero lag causation can be modeled
using an extended MVAR model [10]. PDC is non-zero only when a direct causal link
between two channels exist, while DTF is non-zero if any causal pathway (both direct
and indirect (or cascaded)) occurs between two channels. Mathematically, PDC is a
result of the factorization of the partial coherence (PCoh) function [1]. DTF is a
particularization of another causality measure i.e. Directed Coherence (DC) [1]. The
PDC combines the qualities of partial coherence (PC) and direct coherence (DC) [1].
Major problem in estimating true connectivity measures are: (1) Effects of volume
conduction and, (2) Different artifacts which includes electro galvanic signals (slow
artifacts, movement artifacts and frequency artifacts) associated with brain signals.
Volume conduction is the effect of a single source. The effects of volume conduction
are seen in most of the electrodes [11]. Hence two or more electrodes may appear to be
connected due to the underlying effect of volume conduction but such a connection is
not a true indicator of interaction among electrodes.
An optimum estimator of brain connectivity should mitigate the effects of the
volume conduction. But the MVAR model parameters that we measure are sensitive
to volume conduction [12]. This problem can be overcome by connectivity analysis
at the source level, which requires highly reliable source localization techniques
[13, 14]. Spurious connectivity patterns which occur due to volume conduction can
Generalised Orthogonal Partial Directed … 139

be overcome by orthogonalizing signal powers [15]. In the imaginary part of

coherence between two channels is found to be free ofvolume conduction artifact
[11]. A recent study combines orthogonalization and imaginary part of coherence to
obtain an orthogonalized version of classical PDC called the Orthogonalized Partial
Directed Coherence (OPDC). The OPDC is not affected by volume conduction
artifacts. Hence it has the potential to be a true estimator of brain connectivity among
EEG channels. Analogous to the definition of gPDC, the gOPDC can be formulated
from OPDC [16].
This paper is organized as follows. In Sect. 1 the data collection method has been
discussed. Then in Sect. 2 clear description of the EEG datasets and the prepro-
cessing techniques are given. Section 3 throws some light on the MVAR model
fitting on the EEG data and various connectivity estimators are briefly discussed.
Section 4 then explains about the surrogate data method for statistical validation of
obtained results. In Sect. 5 summarizes the results of this connectivity study. The
paper concludes in Sect. 6.

2 Data Description and EEG Signal Preprocessing

EEG data has been collected from 23 meditating subjects. The data is collected at a
sampling frequency of 256 Hz. To include all the five bands of brain waves, data
from each channel is band pass filtered within 0–64 Hz. A notch filter has been used
to remove the 50 Hz power-line interference frequency component from the EEG
signal. The baseline has also been removed from the data. The five bands of brain
waves are delta (0.4–4 Hz), theta (4–8 Hz), alpha (8–16 Hz), beta (16–32 Hz) and
gamma (>32 Hz) bands [17]. Artifacts from the EEG data has been removed by
using the technique of wavelet thresholding [18] using the ‘db4’ mother wavelet
and scaling function. The db4 mother wavelet has been used owing to its structural
similarity with the rhythmic EEG. The general block diagram of preprocessing
technique has been given as in Fig. 1.

3 MVAR Model Fitting and Time Varying Connectivity

Measures

Consider a multichannel EEG data with M channels and N data points. The strictly
causal MVAR model that we fit into the above data set is of the form:

X
p
y ð nÞ ¼ AðkÞyðn  kÞ þ wðnÞ ð1Þ
k¼1
140 L. Shaw et al.

Fig. 1 General Block diagram of preprocessing techniques and followed by connectivity

measures using g-PDC and g-OPDC

Here, y(n) = [y1(n), y2(n),…, yM(n)]T is a M × 1 vector of present values of the

model, y(n − k)) = [y1(n − k),y2(n − k),…..yM(n − k)]T is a M × 1 vector of values
of y(n) at lag k, A(k) is a M × M coefficient matrix, p is the model order, which
denotes the maximum lag of the model, w(n) = [w1(n), w2(n), …, wM(n)]T is a
M × 1 vector of uncorrelated white noise, w(n) is also called innovation process.
Since it is a matrix of uncorrelated noise, the correlation matrix given by Rw = E[w
(n)wT(n − k)] is zero for all positive lags except for k = 0, where it equals the
covariance matrix ∑ = cov(w(n)). ∑ is a diagonal matrix containing the variances of
the innovation process as the diagonal elements. The diagonality of ∑ ensures that
instantaneous effects are absent in the MVAR model described above [10].
AðkÞ, the coefficient matrix for a lag k is of the form:
2 3
a11 ðk Þ


a1M ðk Þ
6 .. .. .. 7
AðkÞ ¼ 4 . . . 5 ð2Þ
aM1 ðk Þ


aMM ðk Þ

The real parameters aij(k) of A(k) matrix find the relation between time series
i and j at a lag k. If aij(k) is non-zero for at leas tone lag k, then series j is said to
cause series i.
For all k varying from 1 to p we have different values of A(k). Hence, the total
number of parameters to be estimated is M × Mp.
For reliable estimation of model parameters that give a good estimation of the
actual data the number of data points MN must be significantly larger than number
of parameters to be estimated [19], i.e.

MN  M 2 p ð3Þ

This implies N ≥ MP. The optimum model order p can be chosen using different
information theory based criteria, the AIC (Akaike Information Criterion) and SBC
(Schwarz Bayesian Information Criterion), to name a few. In [20] the SBC is found
Generalised Orthogonal Partial Directed … 141

to outperform AIC for time series analysis. The model order must be high enough to
account for all the delays and fluctuations in the original time series and low enough
to allow authentic model identification from the measured data [19].
“Equation (1)” is the strictly causal MVAR model for time invariant systems.
The time varying form of (1) can be written as:

X
p
y ð nÞ ¼ Aðn; kÞyðn  k Þ þ wðnÞ ð4Þ
k¼1

Here in (4) the coefficients in the parameter matrix A(n, k) are time varying for
all lag k. The time varying parameters take care of the non-linearity of the EEG
channels. The time varying MVAR model given in (4) is fit into the data by using
Adaptive Auto-Regressive (AAR) modeling algorithm. This algorithm uses Kalman
filtering for estimating the time varying model parameters [20]. The kalman filtering
based parameter estimation has been done using the ‘mvaar’ module of the BIOSIG
toolbox. The model order p has been estimated by using the ARFIT module [21].
The model order is kept constant throughout the analysis. Optimal model order
depends on the sampling rate and higher sampling rate often requires high model
order.
The frequency domain representation of (4) is given as:

X
p
 
Yð f Þ ¼ Aðn; k Þei2pfk Y ð f Þ þ W ð f Þ ð5Þ
k¼1

Let

X
p
Aðn; f Þ ¼ Aðn; kÞei2pfk ð6Þ
k¼1

A(n, f) is a M × M matrix with each element Aij(n, f) given as

X
p
Aijðn; f Þ ¼ aijðn; kÞei2pfk ð7Þ
k¼1

Using (5) and (6) we have:

Y ð f Þ ¼ Aðn; f ÞY ð f Þ þ W ð f Þ ð8Þ

Taking A(n, f), Y(f) to the L.H.S., we have

ðI  Aðn; f ÞÞY ð f Þ ¼ W ð f Þ ð9Þ

142 L. Shaw et al.

Here I is a M × M identity matrix

Let

^ ðn; f Þ ¼ ðI  Aðn; f ÞÞ
A ð10Þ

Using (9) and (10), we have

^ ðn; f ÞY ð f Þ ¼ W ð f Þ
A ð11Þ

The frequency domain MVAR equation in the form as in (11) can be used to
define most of the strictly causal connectivity estimators [1].
We will directly proceed to write down the formulae of time varying PDC,
g-PDC, OPDC and g-OPDC with the requisite explanation.

3.1 Time Varying PDC

The time varying version of PDC is given as:

 
Akl
^ ðn; f Þ
pklðn; f Þ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð12Þ
PM  2

^
m¼1 Amlðn; f Þ

Here πkl measures the amount of time varying information flow from yl to yk
through direct transfer path only relative to the total outflow leaving the structure at
which yk is measured [1].
Direct transfer path implies direct causality. The PDC measure does not take into
consideration any cascaded path, thus being different from DTF. But this classical
form of PDC is not scale invariant. It is affected by amplitude scaling which does
not affect the causality structure [4]. To overcome this problem the g-PDC was
developed.

3.2 Time Varying G-PDC

A scale invariant version of PDC can be given as [4]:

 1 
ck Akl ^ ðn; f Þ
_ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pkl   ð13Þ
PM 2 Aml^ ðn; f Þ2
m¼1 cm

Here ck 2 refers to the variance of the innovation processes wk(n). This is called
the generalized PDC or simply g-PDC. The physical interpretation of g-PDC is
same as that of PDC but the g-PDC is invariant to any amplitude scaling.
Generalised Orthogonal Partial Directed … 143

3.3 Time Varying Orthogonalized-PDC

The orthogonalized PDC is the result of a recent work [22]. The main concept behind
OPDC and g-OPDC is instead of performing orthogonalization process at amplitude
level, it is done at the level MVAR coefficients to mitigate the effect of volume
conduction (or mutual sources). As given in [22] the time varying OPDC is defined as:
  
jRealðAkl ^ ðn; f Þj imag Akl ^ ðn; f Þ 
aklðn; f Þ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð14Þ
PM  PM  2
^ ðn; f Þ2 ^ 
m¼1 Aml m¼1 Amlðn; f Þ

here k ≠l.

The physical interpretation is same as classical PDC except that the OPDC does
not take into consideration the effect of mutual sources.

3.4 Time Varying Generalized−Orthogonalized-PDC

The time varying g-OPDC is the scale invariant version of OPDC [16]. It is defined
as:
 
1 jRealðAklðn; ^ f Þj jimag Akl ^ ðn; f Þ j
_
aklðn; fÞ ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð15Þ
ck2 PM 1 jAmlðn; ^ f Þj2
PM 1
^
jAmlðn; f Þj2
m¼1 cm2 m¼1 cm2

here k ≠l.

Interpretation of this formula is same as OPDC except that this one is scale
invariant.

4 Statistical Assessment and Its Significance

for connectivity Measures

The PDC measures calculated in this paper have a nonlinear relation to the time
series data from which they are derived [23] hence the probability distributions of
their estimators are not well defined, making the statistical testing of significance
very difficult. In this paper we have used the surrogate data method [24].
144 L. Shaw et al.

4.1 Surrogate Data Analysis

Surrogate data is the random data generated, keeping the mean, variance and
autocorrelation function as same as the original data and this techniques of time
series can be used to test for nonlinear dynamics [25]. The data points in all the
channels are randomly permutated to remove any causal ordering. Then a time
varying MVAR model is fit to this shuffled data and the connectivity measures are
calculated from it. This process is undertaken several times to create an empirical
distribution for connectivity measures. The estimators calculated from the surrogate
data set serve the null hypothesis which assumes that there is no causal relationship
between the channels of the data set. Using this new distribution we can assess the
significance of causal measures calculated from actual data. The method has been
validated in [2] and found to be effective.
Hence, the basic problem in our analysis is to check the null hypothesis H0:

H0 : pkl ðn; f Þ ¼ 0 ð16Þ

This hypothesis if rejected implies the existence of direct information flow from
yl to yk. And this happens when akl(k) is non-zero at least for one k in [1, p].

5 Results and Discussion

To study intra and inter hemispheric interaction, eight electrodes have been selected
out of 64 [24]. The electrodes are F3, F4, Fz, C3, C4, P3, P4 and Pz. Here these
eight electrodes represent the midline hemisphere of the brain. The result of g-PDC
and g-OPDC for one of the meditators are given below. Figure 2 is the output for
g-PDC estimator which shows the amount of information flow out of an electrode
to another electrode in the direction of the arrow shown in the figure. The mag-
nitude of information flow for each corresponding time frequency point is shown by
the color in the plot. Red stands for high value of information flow while blue
stands for negligible information flow. The x axis of each subplot is samples and the
y axis is the frequency axis. For each time frequency pair we find the amount of
information flow. From the figure it is clearly visible that g-PDC is not a symmetric
measure which clearly implies

g-PDCij 6¼ g-PDCji ð17Þ

The figures are testimony to the directional nature of g-PDC and other PDC
based measures (Fig. 3).
According to the g-PDC figure, we see some information flow from F4 to almost
all electrodes except Fz. Similarly we find significant information flow from F4 to
almost all other electrodes. But the g-OPDC measure is not as colored as the g-PDC
Generalised Orthogonal Partial Directed … 145

F4 <--- F4 F4 <--- Fz F4 <--- F3 F4 <--- C4 F4 <--- C3 F4 <--- P4 F4 <--- Pz F4 <--- P3 0.5

60 50 50 50 50 50 50 50
40 40 40 40 40 40 40
40 30 30 30 30 30 30 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10
200 400 600 800 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 0.45
Fz <--- F4 Fz <--- Fz Fz <--- F3 Fz <--- C4 Fz <--- C3 Fz <--- P4 Fz <--- Pz Fz <--- P3
50 60 50 50 50 50 50 50
40 40 40 40 40 40 40
30 40 30 30 30 30 30 30
20 20 20 20 20 20 20 0.4
10 20 10 10 10 10 10 10
200 400 600 200 400 600 800 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600
F3 <--- F4 F3 <--- Fz F3 <--- F3 F3 <--- C4 F3 <--- C3 F3 <--- P4 F3 <--- Pz F3 <--- P3
50 50 60 50 50 50 50 50
40 40 40 40 40 40 40 0.35
30 30 40 30 30 30 30 30
20 20 20 20 20 20 20
10 10 20 10 10 10 10 10
200 400 600 200 400 600 200 400 600 800 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600
C4 <--- F4 C4 <--- Fz C4 <--- F3 C4 <--- C4 C4 <--- C3 C4 <--- P4 C4 <--- Pz C4 <--- P3
0.3
Frequency (Hz)

50 50 50 60 50 50 50 50
40 40 40 40 40 40 40
30 30 30 40 30 30 30 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10
200 400 600 200 400 600 200 400 600 200 400 600 800 200 400 600 200 400 600 200 400 600 200 400 600 0.25
C3 <--- F4 C3 <--- Fz C3 <--- F3 C3 <--- C4 C3 <--- C3 C3 <--- P4 C3 <--- Pz C3 <--- P3
50 50 50 50 60 50 50 50
40 40 40 40 40 40 40
30 30 30 30 40 30 30 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10 0.2
200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 800 200 400 600 200 400 600 200 400 600
P4 <--- F4 P4 <--- Fz P4 <--- F3 P4 <--- C4 P4 <--- C3 P4 <--- P4 P4 <--- Pz P4 <--- P3
50 50 50 50 50 60 50 50
40 40 40 40 40 40 40
30 30 30 30 30 40 30 30 0.15
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10
200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 800 200 400 600 200 400 600
Pz <--- F4 Pz <--- Fz Pz <--- F3 Pz <--- C4 Pz <--- C3 Pz <--- P4 Pz <--- Pz Pz <--- P3
50 50 50 50 50 50 60 50 0.1
40 40 40 40 40 40 40
30 30 30 30 30 30 40 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10
200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 800 200 400 600
P3 <--- F4 P3 <--- Fz P3 <--- F3 P3 <--- C4 P3 <--- C3 P3 <--- P4 P3 <--- Pz P3 <--- P3 0.05
50 50 50 50 50 50 50 60
40 40 40 40 40 40 40 40
30 30 30 30 30 30 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10 0
200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 800
samples(*10)

Fig. 2 Time varying connectivity analysis of the selected 8 scalp EEG electrodes: g-PDC measure

F4 <--- F4 F4 <--- Fz F4 <--- F3 F4 <--- C4 F4 <--- C3 F4 <--- P4 F4 <--- Pz F4 <--- P3 0.05

60 50 50 50 50 50 50 50
40 40 40 40 40 40 40
40 30 30 30 30 30 30 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10
200 400 600 800 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 0.045
Fz <--- F4 Fz <--- Fz Fz <--- F3 Fz <--- C4 Fz <--- C3 Fz <--- P4 Fz <--- Pz Fz <--- P3
50 60 50 50 50 50 50 50
40 40 40 40 40 40 40
30 40 30 30 30 30 30 30
20 20 20 20 20 20 20 0.04
10 20 10 10 10 10 10 10
200 400 600 200 400 600 800 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600
F3 <--- F4 F3 <--- Fz F3 <--- F3 F3 <--- C4 F3 <--- C3 F3 <--- P4 F3 <--- Pz F3 <--- P3
50 50 60 50 50 50 50 50
40 40 40 40 40 40 40 0.035
30 30 40 30 30 30 30 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10
200 400 600 200 400 600 200 400 600 800 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600
C4 <--- F4 C4 <--- Fz C4 <--- F3 C4 <--- C4 C4 <--- C3 C4 <--- P4 C4 <--- Pz C4 <--- P3
Frequency (Hz)

0.03
50 50 50 60 50 50 50 50
40 40 40 40 40 40 40
30 30 30 40 30 30 30 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10
200 400 600 200 400 600 200 400 600 200 400 600 800 200 400 600 200 400 600 200 400 600 200 400 600 0.025
C3 <--- F4 C3 <--- Fz C3 <--- F3 C3 <--- C4 C3 <--- C3 C3 <--- P4 C3 <--- Pz C3 <--- P3
50 50 50 50 60 50 50 50
40 40 40 40 40 40 40
30 30 30 30 40 30 30 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10 0.02
200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 800 200 400 600 200 400 600 200 400 600
P4 <--- F4 P4 <--- Fz P4 <--- F3 P4 <--- C4 P4 <--- C3 P4 <--- P4 P4 <--- Pz P4 <--- P3
50 50 50 50 50 60 50 50
40 40 40 40 40 40 40
30 30 30 30 30 40 30 30 0.015
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10
200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 800 200 400 600 200 400 600
Pz <--- F4 Pz <--- Fz Pz <--- F3 Pz <--- C4 Pz <--- C3 Pz <--- P4 Pz <--- Pz Pz <--- P3
50 50 50 50 50 50 50 0.01
40 40 40 60 40
40 40 40
30 30 30 30 30 30 40 30
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10
200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 800 200 400 600
P3 <--- F4 P3 <--- Fz P3 <--- F3 P3 <--- C4 P3 <--- C3 P3 <--- P4 P3 <--- Pz P3 <--- P3 0.005
50 50 50 50 50 50 50 60
40 40 40 40 40 40 40
30 30 30 30 30 30 30 40
20 20 20 20 20 20 20 20
10 10 10 10 10 10 10 0
200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 200 400 600 800
samples(*10)

Fig. 3 Time varying connectivity analysis of the selected 8 scalp EEG electrodes: g-OPDC
measure

because the g-OPDC does not take into consideration the effect of volume con-
duction and for this reason g-OPDC has lighter shades as compared to g-PDC plots.
All the interpretations written above for g-PDC also hold good for g-OPDC. Also
the magnitude of g-OPDC is significantly lower compared to g-PDC, which is
evident from the color bars present in both the figures.
146 L. Shaw et al.

The figure also shows the ability of the Kalman filtering algorithm to capture the
dynamics of the non-stationary EEG signal. The time varying versions of g-PDC
and g-OPDC based on Kalman filtering algorithm show good time frequency res-
olution. The volume conduction artifact is generally present in the low frequency
regions and unlike g-PDC we see that the g-OPDC has negligible values in the low
frequency range (*3 Hz). In this way, g-OPDC seems to be successful in elimi-
nating the effect of volume conduction.
Here we have used g-OPDC to study brain connectivity during meditation. The
g-OPDC measure takes time series scaling and information leakage into account
which gives the most desired presentation of the neural information flow. The
MVAR model based coherence connectivity estimator offers a common framework
for brain connectivity analysis. Here we discuss some general issues associated with
connectivity comparisonbetween g-PDC and g-OPDC.

6 Conclusion

This paper measures the brain connectivity during meditation. Not many studies in
literature have investigated the brain connectivity during meditation using Granger
causality measures using PDC and its other variants on EEG data. Our study is a
comparison based on visual inspection of g-PDC and g-OPDC measure applied to
the EEG signal captured during meditation.
Some studies have reported increase in alpha band coherence during meditation
while some reported reduced functional connectivity between cortical sources [26].
A Diffusion Tensor Imaging (DTI) based study has reported enhanced brain con-
nectivity in long term meditation practitioners.
While there is a plethora of conclusions drawn from varied number of studies,
the effect of mediation on human brain is still not well defined.
Though there has been a considerable amount of statistical study of meditative
EEG waves, the study of effective brain connectivity using EEG during meditation
is still an open research topic [27].

References

1. Laskovski A (ed) (2011) Multivariate frequency domain analysis of causal interactions in

physiological time series. Biomedical Engineering trends in electronics, communication and
software, Intech, pp 403–428
2. Kaminiski M, Ding M, Truccolo WA, Bressler SL (2001) Evaluating causal relationship in
neural systems: granger causality, DTF and statistical assessment of significance. Biol Cybern
85:145–157 (Springer verlag)
3. Baccala LA, Sameshima K (2001) Partial directed coherence: a new concept in neural
structure determination. Biol Cybern 84:463–474 (Springer verlag)
Generalised Orthogonal Partial Directed … 147

4. Baccala LA, Sameshima K (2007) Generalized partial directed coherence. In: 15th
International IEEE conference on digital signal processing, pp 163–166
5. Rankine L, Stevenson N, Mesbah M, Boashash B (2007) A non-stationary model of new born
EEG. IEEE Trans Bio Med Eng 54(1):19–28
6. Ting CM, Salleh SH, Zainuddin ZM, Bahar A (2011) Spectral estimation of non stationarity of
EEG using particle filtering with application to event related desynchronization. IEEE Trans
Bio Eng 58(2):321–331
7. Hesse W, Moller E, Arnold M, Schack B (2003) The use of time-variant EEG Granger
causality for inspecting directed interdependencies of neural assemblies. J Neurosci Methods
123:27–44
8. Astolfi L, Cincotti F, Mattia D, Fallani F, Tocci A, Colosimo A, Salinari S, Marciani MG,
Hesse W, Witte H, Ursino M, Zavaglia M, Babiloni F (2008) Tracking the time varying cortical
connectivity patterns by adaptive MV estimation. IEEE Trans Biomed Eng 55(3):902–913
9. Sommerlade L, Henschel K, Wohlmuth J, Jachan M, Amtage F, Hellwig B, Lückin C, Timmer
J, Schelte B (2009) Time variant estimation of directed influences during parkinsonian tremor.
J Physiol 103(6):348–352
10. Faes L, Nollo G (2010) Extended causal modelling to assess PDC in multiple time series with
significant instantaneous interaction. Biol Cybern Biol Cybern 103(5):387–400
11. Nolte G, Bai O, Wheaton L, Mari Z, Vorbach S, Hallett M (2004) Identifying true brain
interaction from EEG data using the imaginary part of coherency. Clin Neurophysiol
115:2292–2307
12. Gomez G (2010) Brain connectivity analysis with EEG. Doctoral dissertation, Department of
Signal Processing, Tampere University of Technology
13. Brookes MV, Woolrich MJ, Luckhoo M, Price H, Hale D, Stephenson JR, Barnes MC, Smith
GR, Morris SM, Peter G (2011) Investigating the physiological basis of resting state networks
using MEG. Proc Natl Acad Sci USA 108(40):16783–16788
14. Palva S, Kulashekhar S, Hmalinen, Palva JM (2010) Neural Synchrony reveals working
memory networks and predicts individual memory capacity. Proc Natl Acad Sci USA
107:7580–7585
15. Hipp JF, Hawellek DJ, Corbetta M, Siegel M, Engel AK (2012) Large scale cortical
correlation structure of spontaneous oscillatory activity. Nat Neurosci 15:884–890
16. Omidvarnia A, Azemi G, Boashash B, O’Toole JM, Colditz P, Vanhatalo S (2014) Measuring
time varying information flow in scalp EEG signals: orthogonalized partial directed
Coherence. IEEE Trans Biomed Eng 61(3):680–693
17. Murugappan M, Nagarajan R, Yaccob S Discrete wavelet transform based selection of salient
EEG frequency band for assessing human emotion. http://cdn.intechopen.com/pdfs.wm/
19508.pdf
18. Kumar P, Arumuganathan R, Sivakumar K, Vimal C (2008) A wavelet based statistical
method for denoising of ocular artifacts: artifact in EEG signals. Int J Comput Sci Netw Secur
8(9):87–92
19. Hytti H, Takalo R, Ihalainen H (2006) Tutorial on multivariate autoregressive modeling. J Clin
Monit Comput 20(2):101–108
20. Koehler AB, Murphree ES (1988) A comparison of akaike and Schwarz criteria for selecting
model order. Appl Stat 37(2):187–195
21. Arnold M, Miltner W, Witte H, Bauer R, Braun C (1998) Adaptive AR modeling of non-
stationary time series by means of kalman filtering. IEEE Trans Biomed Eng 45(5):553–562
22. Schneider T, Neumaier A (2001) Algorithm 808: ARFIT-a Matlab package for the estimation
of parameters and eigenmodes of multivariate autoregressive models. ACM-Trans Math Softw
27:58–65
23. Omidvarnia A, Azemi G, Boashash B, Toole J, Colditz P, Vanhatalo S (2012) Orthogonalized
partial directed coherence for functional connectivity analysis of newborn. Neural Inf Proc
EEG 7664:683–691
24. Theiler J, Eubank S, Longtin A, Galdrikian B, Farmer J (1992) Testing of non-linearity in time
series: the method of surrogate data. Physica D 58:77–94
148 L. Shaw et al.

25. Theiler J, Eubank S, Longtin A, Galdrikian B, Farmer D (1992) Testing for nonlinearity in
time series: the method of surrogate data. Physica D 58(92):77–94
26. Hebert R, Lehmann D, Tan G, Travis F, Alexander A (2005) Enhanced EEG alpha time-
domain phase synchrony during transcendental meditation: implications for cortical
integration theory. J Signal Process 85(11):2213–2232
27. Tang Y, Rothbart M, Posner MI (2012) Neural correlates of establishing, maintaining, and
switching brain states. Trends Cogn sci 16(6):330–337
An Approach for Identification Using
Knuckle and Fingerprint Biometrics
Employing Wavelet Based Image Fusion
and SIFT Feature Detection

Aritra Dey, Akash Pal, Aroma Mukherjee

and Karabi Ganguly Bhattacharjee

Abstract Identification is an essential part of our lives. Identification of authentic

candidate is essential in E-commerce, in keeping track of criminals, in airport and
railway surveillance and many more aspects of the modern world. In this paper an
approach has been proposed combining the knuckle and fingerprint biometrics for
the purpose of identification of a person using SIFT (Scale Invariant Feature
Transform).

Keywords SIFT
DWT
Knuckle-fingerprint

1 Introduction

In the field of identification, biometric identification techniques such as Fingerprint

recognition, Facial recognition, Voice pattern, Handwritten Signature, Retina rec-
ognition, Iris recognition are well established. However, none of the biometrics is
giving cent percent accuracy. Multi biometric systems [1] remove some of the
drawbacks of the uni-modal biometric systems by acquiring multiple sources of
information together in a group, which has richer detail. Utilization of these

A. Dey (&)
Department of Electrical Engineering, JIS College of Engineering, Kalyani, India
e-mail: [email protected]
A. Pal
Department of Computer Science and Engineering, JIS College of Engineering,
Kalyani, India
e-mail: [email protected]
A. Mukherjee
K.G. Bhattacharjee
Department of Biomedical Engineering, Kalyani, India
e-mail: [email protected]
K.G. Bhattacharjee
e-mail: [email protected]

© Springer India 2015 149

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_14
150 A. Dey et al.

biometric systems depends on more than one physiological or behavioral

characteristic for enrollment and verification/identification [2]. Multi-resolution
approach in wavelet is well suited to manage the different image resolutions. Many
research works have studied on multi-resolution representation of signals and have
established that multi-resolution information for a number of image processing
applications including the image fusion. Wavelet coefficients coming from different
images can be appropriately combined to obtain new coefficients, so that the
information in source images is collected appropriately. The discrete wavelet
transform (DWT) allows the image decomposition in different kinds of coefficients
preserving the image information. A wavelet-based image fusion method is
therefore required to identify the most important information in the input images
and to transfer it into the fused image [3].
Zhang et al. proposed a new kind of biometric identifier, called Finger-Knuckle-
Print (FKP) an alternate for personal identity authentication. In those works [4–8],
they used different techniques for Feature Detection and matching. In general, most
of the previous SIFT FKP based authentication systems claim that they have
achieved greater than 98 % accuracy. In all the previous works only knuckle has
been used. This work focuses in exploring new possibilities of identification by
combing the knuckle and finger print.

2 Wavelet Decomposition for Images

This process basically takes two images of knuckle and finger print [9] after wavelet
decomposition is performed on both images followed by a fusion of decomposition
of two images which produces fused image of low resolution. It has the capability
to provide good localization for both frequencies and space domains. The wavelet
based image fusion would be applied to two dimensional multispectral knuckle and
fingerprint at each level [10].

3 SIFT Descriptor

The scale invariant feature transform, called SIFT [11, 12] descriptor, has been
proposed by and proved to be invariant to image rotation, scaling, translation, partly
illumination changes. Following are the major stages of computation used to
generate the set of image features.
An Approach for Identification Using Knuckle … 151

3.1 Scale Space Extrema Detection

The first stage of computation is to create a scale space of images. This is done by
constructing a set of progressively Gaussian blurred images with increasing values
of sigma. Then the difference between pairs of Gaussian is taken to obtain a
Difference of Gaussian (DOG) which is similar to the function Laplacian of
Gaussian (LOG) to obtain potential locations for finding features. The image is then
sub-sampled (i.e. 1/4th resolution of lower octave) to obtain the next octave and the
same process is repeated to obtain DOG pyramid (Fig. 1).

Fig. 1 Operations within same octaves (same scale)

152 A. Dey et al.

3.2 Keypoint Localization

Accurately locates the feature keypoints by comparing a pixel (X) with 26 pixels in
current scale and adjacent scales (Green Circles). The pixel (X) is selected if it is
larger/smaller than all 26 pixels. There are still a lot of points; some of them are not
good enough. The locations of keypoints may be not accurate. Eliminating edge
points, keypoints are selected from the extrema based on measures of their stability.

3.3 Orientation Assignment

This step assigns orientation to each keypoint, the keypoint descriptor can be
represented relative to this orientation and therefore achieve invariance to image
rotation. It computes magnitude and orientation on the Gaussian smoothed images.
An orientation histogram is formed from the gradient orientations of sample points
within a region around the keypoint. Peaks in the orientation histogram correspond
to dominant directions of local gradients. The highest peak in the histogram is
detected, and then any other local peak that is within 80 % of the highest peak is
used to also create a keypoint with that orientation. One or more orientations are
assigned to each keypoint location based on local image gradient directions (Fig. 2).

Fig. 2 Operations between

different octaves (different
scale)
An Approach for Identification Using Knuckle … 153

Fig. 3 The keypoint descriptor

3.4 Keypoint Descriptor

This step describes the keypoint as a high dimensional vector. The local image
gradients are measured at the selected scale in the region around each keypoint. It
computes relative orientation and magnitude in a 16 × 16 neighborhood around
each keypoint. It forms weighted histogram (8 bin) for 4 × 4 regions. Finally it
concatenates 16 histograms in one long vector of 128 dimensions.
These are transformed into a representation that allows for significant levels of
local shape distortion and change in illumination. This approach has been named
the Scale Invariant Feature Transform (SIFT), as it transforms image data into scale
invariant coordinates relative to local features. An important aspect of this approach
is that it generates large number of features that densely cover the image over the
full range of scales and locations (Fig. 3).

4 Proposed Method

1. Two images (i.e. knuckle and fingerprint) of same size are read and 1-level
wavelet decomposition is performed for both images.
2. Fuse synthesized images.
3. Perform SIFT on the fused image.
4. Save extracted corners saved as a feature point for tracking and recognizing
objects in the database for matching.
154 A. Dey et al.

5 Flowchart

Knuckle Fingerprint

1-DWT 1-DWT

Synthesised
Synthesised
Image of
Image of
fingerprint
knuckle

Image Fusion

Fused image

SIFT
(Scale Invariant Feature Transform)

Save extracted point in database for feature

matching

6 Steps of Matching

1. Matching function reads two images, finds their SIFT features, and displays
lines connecting the matched Keypoints. A match is accepted only if its distance
is less than distance Ratio times the distance to the second closest match. It
returns the number of matches displayed.
An Approach for Identification Using Knuckle … 155

Fig. 4 Knuckle

Fig. 5 Fingerprint

Fig. 6 Synthesized image

156 A. Dey et al.

Fig. 7 Extracted SIFT points

2. Matching function finds SIFT (Scale Invariant Fourier Transform) Keypoints for
each image.
3. Assume some distance ratio for example suppose distance ratio = 0.4 it means
that it only keep matches in which the ratio is less than distance Ratio = 0.5.
Now for each descriptor in the first image, it selects its match to second image.
4. Then it creates a new image showing the two images side by side with lines
joining the accepted matches (Figs. 4, 5, 6 and 7).

7 Matching

See (Figs. 8, 9, 10 and 11).

Fig. 8 Normal matching

An Approach for Identification Using Knuckle … 157

Fig. 9 Image rotated by 90°

Fig. 10 Image rotated by

180°

Fig. 11 Image rotated by

270°
158 A. Dey et al.

8 Result

Number of keypoints found in the fused image is 193

Image transformation Match found

0° 193
90° 191
180° 188
270° 186

9 Conclusion

The total number of keypoints extracted can be stored. These points can be saved in
the database for identification and for future image processing operations like
tracking or recognition of objects. Future work would involve in establishing a
database of knuckle-fingerprint including plotting the ROC Receiver Operating
Characteristic (ROC) which would project the comparison of true vs. false positive
rate, at various threshold settings. Accuracy can also be measured in terms of Area
under the ROC Curve (AUC).

References

1. Jain AK, Ross A (2004) Multibiometric systems. Commun ACM 47(1):34–40

2. Kisku RD, Sing KJ, Tistarelli M, Gupta P (2009) Multisensor biometric evidence fusion for
person authentication using wavelet decomposition and monotonic-decreasing graph. In: 2009
seventh international conference on advances in pattern recognition, 2009 IEEE
3. Dey N, Das S, Rakshit P (2011) A novel approach of obtaining features using wavelet based
image fusion and Harris corner detection. Int J Mod Eng Res 1(2):396–399
4. Zhang L, Zhang L, Zhang D (2009) Finger-knuckle-print: a new biometric identifier. In: 2009
16th IEEE international conference on image processing (ICIP), Cairo, 7–10 Nov 2009, 1981–
1984
5. Zhang L, Zhang L, Zhang D (2009) Finger-knuckle-print verification based on band-limited
phase-only correlation (Lecture Notes). Computer analysis of images and patterns, pp 141–148
6. Zhang L, Zhang L, Zhang D, Zhu HL (2010) Online finger-knuckle-print verification for
personal authentication. Pattern Recogn 43(7):2560–2571
7. Zhang L, Zhang L, Zhang D (2010) Monogeniccode: a novel fast feature coding algorithm
with applications to finger-knuckle-print recognition. In: 2010 international workshop on
emerging techniques and challenges for hand-based biometrics (ETCHB), Istanbul, 22 Aug
2010, pp 1–4
8. Zhang L, Zhang L, Zhang D, Zhu H (2011) Ensemble of local and global information for
finger-knuckle-print recognition. Pattern Recogn 44(9):1990–1998
9. Shameem Sulthana ES, Kanmani S (2014) Implementation and evaluation of SIFT descriptors
based finger-knuckle-print authentication system. Indian J Sci Technol 7(3):374–382
An Approach for Identification Using Knuckle … 159

10. Karanwal S, Kumar D, Maurya R (2010) Fusion of fingerprint and face by using DWT and
SIFT. Int J Comput Appl 2(5):0975–8887
11. Lowe DG (2004) Distinctive image features from scale invariant keypoints. Int J Comput Vis
60(2):91–110
12. Fergus R, Perona P, Zisserman A (2003) Object class recognition by unsupervised scale
invariant learning. In: IEEE conference on computer vision and pattern recognition, Madison,
Wisconsin, pp 264–271
Development of a Multidrug Transporter
Deleted Yeast-Based Highly Sensitive
Fluorescent Biosensor to Determine
the (Anti)Androgenic Endocrine
Disruptors from Environment

Shamba Chatterjee and Sayanta Pal Chowdhury

Abstract A competent and consistent androgen receptor transactivation assay has

been developed using pleiotropic drug resistance transporters Pdr5, Snq2 and Yor1
deleted yeast strain, Saccharomyces cerevisiae, intended to express the human
androgen receptor and androgen response element (probasin promoter) driving the
expression of green fluorescent protein to determine endocrine disruptors from pulp
and paper mill effluents (PPME). Stimulation of cells by known androgens, corre-
lated with androgenic activities as measured by other reported bioassay systems. This
yeast-based assay system when applied to evaluate anti-androgenic activities, the
known anti-androgens effectively inhibited fluorescence reporter gene induction by
dihydrotestosterone. The specificity of the assay was experienced by incubating the
recombinant yeast cells with supraphysiological concentrations of non-androgenic
steroidal compounds and none of them yielded considerable response. Further, the
assay was used to analyze the extracted PPME from different mills confirmed strong
androgenic activities. In conclusion, these results support the earlier report by us that
PPME are rich in androgenic compounds and the employed detection system
provides novel high throughput fluorescence based biosensor system for successful
well sensitive (picogram level) screening of (anti)androgenic chemicals from various
environmental sources.

Keywords Endocrine disruptors
Androgen
Saccharomyces cerevisiae

Fluorescence based biosensor

1 Introduction

Environmental levels of an array of endocrine-disrupting chemicals have been

escalating and thus affecting human health by interacting with hormone receptors
followed by the disruption in normal endocrine activity [1–5]. Such compounds are

S. Chatterjee (&)
S.P. Chowdhury

Department of Biotechnology, Haldia Institute of Technology, Haldia, India
e-mail: [email protected]

© Springer India 2015 161

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_15
162 S. Chatterjee and S.P. Chowdhury

suspected to raise developmental, reproductive and tumorigenic effects to wildlife

and humans even at very low concentrations. Amongst all endocrine disruptors,
environmental estrogens are the most introspected arena [6]. Whereas, reports on
successful, robust and highly sensitive androgen assay with short time period are
available in a scanty mode. As we know, androgens are basically essential for the
masculinization of male genitalia in utero, the augmentation of secondary sex
characteristics in males, and the continuation of male sexual function in adults.
After entering in the target cell, it binds to androgen receptor (AR) driving a ligand-
dependent transcriptional activity. Finally, AR enters into the nucleus and binds
successively to the regulatory region of the target gene as a homodimer that in turn
modulates transcriptional activity [7, 8]. Earlier report shows a crucial involvement
of the xeno-androgens through androgen receptor in development of abnormal sex
related disorders that can be eventually inclined towards testicular cancer [9].
Several assays have been devised for the detection of environmental endocrine
disruptors (EDCs), using both mammalian and yeast cells. But, yeast-based assay
systems achieved its popularity following its low cost, easy handling approach,
complete absence of crosstalk (lack of known endogenous receptors), and use of
media that are devoid of steroids [10–18].
The major aim of this introspection was initially, to develop a yeast yEGFP
reporter-based, rapid, efficient, highly sensitive short timed and economic assay
system, appropriate for screening several samples concurrently for their androgenic
or anti-androgenic activities. To fulfill this notion, the pleiotropic drug resistance
transporters deleted yeast strain (well versed to facilitate efflux of organic com-
pounds including steroids and other chemotherapeutics), was transformed with
human AR (hAR) and androgen response element (ARE) and minimal yeast-based
promoter (CYC1) driving yEGFP reporter gene in the corresponding yeast-based
expression vectors. The recombinant Saccharomyces cerevisiae strain was then
employed to analyze the androgenicity of different steroids and synthetic chemicals.
The assay was also utilized to test the pulp and paper mill effluents for their
in vitro androgenicity. Earlier, one Swedish group has published a report on the
analysis of PPME using recombinant yeast-based reporter system [19]. Later on, we
have already reported a recombinant yeast-based rapid and sensitive β-galactosidase
biosensor assay [20]. Here, the developed yEGFP reporter based assay is much
more sensitive than the earlier framed one and can thus be utilized to screen (anti)
androgenic chemicals from various environmental sources.

2 Materials and Methods

2.1 Chemicals

D-glucose and yeast nitrogen base without amino acids and without ammonium
sulphate were acquired from Himedia (Mumbai, India). Trizol reagent, L-leucin,
tryptophan and uracil were from Sigma (St. Louis, MO, USA). Ammonium
Development of a Multidrug Transporter … 163

sulphate, chloroform, isoamyl alcohol, isopropanol, ethanol absolute and dimethyl

sulfoxide (DMSO) were obtained from Merck (Merck, Mumbai, India). DNase I,
Ribonuclease inhibitor and restriction endonucleases were all procured from Pro-
mega (Madison, WI, USA). Testosterone (T), Dihydrotestosterone (DHT), estrogen
(E), progesterone (Prog), all trans retinoic acid (RA), dexamethasone (Dex),
hydroxy flutamide (HF), cyproterone acetate (CPA), spironolactone, p,p′-DDE and
vinclozolin were kindly provided by Prof. Ilpo Huhtaniemi, Imperial College,
London, UK.

2.2 PPME Sample Collection and Preparation

PPME effluents were collected from the outlets of five different pulp and paper
industries of northern India. These samples (2 L) were extracted instantaneously
after collection by employing solid phase extraction process. They were first filtered
through 0.1 µm glass fibre filters (Type GMF5, Rankem, Mumbai, India), followed
by an acidification process with concentrated sulphuric acid to pH 2.0, and finally
separated into two 1 L samples. Then, 1 L of the sample was extracted utilizing
reverse phase C18 solid phase extraction columns (RP-C18 SPE, Rankem, Mumbai,
India) and dissolved in 1,000 µL of DMSO (a concentration factor of 1,000). Next, a
1:100 dilution in medium for the above extracted samples yielded in the highest test
concentration of 1 mL eq./well.

2.3 Plasmids and Yeast Strains

The pRS425-Leu2-ARS yeast expression vector (kindly provided by Dr. A.

Bachhawat, IMTECH, Chandigarh, India) was used to generate hAR construct.
Yeast compatible hAR construct, pRS425-Leu2-ARS-hAR was made as per the
procedure mentioned earlier by Chatterjee et al. [20]. The high copy Escherichia
coli/yeast shuttle vector pYEX-BX (Clontech, Palo Alto, USA) was employed to
construct the green-fluorescent protein reporter plasmid, pARE-CYC1-GFP. Incor-
poration of the CYC1 minimal promoter and yeast-enhanced yEGFP gene of Ae-
quorea victoria in plasmid pERE-CYC1-GFP was elaborated earlier [21]. Next, the
estrogen receptor-responsive element cassette (ERE) was primarily reinstated by
using the XbaI and BamHI restriction sites and end-filled with Klenow polymerase.
Finally, a 0.3 kb fragment of probasin promoter was utilized as androgen response
element (ARE) which was digested with BglII and HindIII from pGL3-Probasin-
Luciferase expression vector, endfilled with Klenow polymerase and both the blunt
ended products were ligated using T4 DNA Ligase to form pARE-CYC1-GFP
(ARE-GFP). The recombinant plasmids were mapped by restriction analysis and
confirmed by sequencing.
164 S. Chatterjee and S.P. Chowdhury

Table 1 Haploid yeast strains

Haploid yeast Genotype Source
strains
FY1679-28C MATa, ura 3-52, trp 1Δ63, leu 2Δ1, his3Δ200, GAL2+ Kolaczkowski
et al. (1998)
FYAK26/8-10B1 MATa, ura 3-52, trp 1Δ63, leu 2Δ1, his3Δ200, Kolaczkowski
GAL2+, pdr5-Δ1::hisG, snq2::hisG, yor1-1::hisG et al. (1998)
SCY1 FYAK26/8-10B1 This study
[pARE-CYC1-GFP]
SCY2 FYAK26/8-10B1 This study
[pARE-CYC1-GFP]
[pRS425-Leu2-ARS-hAR]

The yeast strain Saccharomyces cerevisiae FYAK26/8-10B1 (MATa, ura 3-52,

trp 1Δ63, leu 2Δ1, his3Δ200, GAL2+, pdr5-Δ1::hisG, snq2::hisG, yor1-1::hisG)
was a kind gift from Dr. M. Ghislain (Table 1).

2.4 Transformation and Culture of Recombinant Yeast Cells

Plasmids harboring hAR and ARE-lacZ were co-transformed into Saccaromyces

cerevisiae, FYAK26/8-10B1 using the lithium acetate protocol [20]. Transformants
were grown in minimal YNB media consisting of 0.67 % yeast nitrogen base
(YNB) with (NH4)2SO4 and 0.5 % d-glucose adjusted to pH 6.4 and supplemented
with the suitable amino acids without tryptophan, leucine and uracil. Glycerol
stocks of the recombinant strains were prepared and stored at −80 °C.

2.5 Isolation of RNA and RT-PCR

Total RNA was isolated according to the method described by Ausubel et al. [22].
The RNA pellet thus obtained was resuspended in 50 µL of DEPC treated water
and stored at −80 °C for future use.
RT-PCR was carried out with 1 µg of total RNA as template by using the MMLV
reverse transcriptase and oligo-dT primers (Promega, Madison, WI, USA). Oligo-
nucleotide primers were designed from areas conserved in the published sequences
of hAR cDNA sequences as follows: sense, 5′-ACCATGTTTTGCCCATTGAC-3′;
antisense, 5′-GCTGTACATCCGGGACTTGT-3′. The PCR was performed using
28 PCR cycles with Taq polymerase (94 °C for 30 s, 50 °C for 75 s and 72 °C for 90 s
and finally at 72 °C for 10 min for one cycle). The amplified DNA samples were run
on a 1.5 % agarose gel and bands were visualized with ethidium bromide.
Development of a Multidrug Transporter … 165

2.6 Western Blot

Western blot analysis was carried out according to the protocol depicted by
Laemmli [23]. In brief, the proteins thus obtained from yeast biosensors with/
without pRS425-Leu2-ARS-hAR expression vectors were loaded on a 10 % SDS
polyacrylamide gel. Finally, proteins were transferred to a nitrocellulose membrane
and it was incubated overnight with a polyclonal antibody for hAR (1:200) in the
presence of blocking buffer. This was further incubated with alkaline phosphatase
labeled secondary antibody (1:1,000). Color development was achieved in BCIP/
NBT solution. The extracted protein samples from LNCaP cells were utilized as
positive control.

2.7 Confocal Microscopy

Transformed yeast strains were cultured with exposure to testosterone for 16.5 h
and 1 ml of the culture was washed with MilliQ followed by formaldehyde fixation.
Next, cells were harvested and washed twice with PBS + BSA (1 mg/ml) +
0.1 % (v/v) Triton X-100. Then, the pellet was resuspended in 50 µl PBS/BSA
buffer. Finally, cells were washed twice with the same buffer without Triton X-100
and were examined under Olympus Fluoview FV-1,000 (Olympus, Japan) with a
60× oil-immersion objective.

2.8 Biosensor Assay Conditions and Fluorescence

Monitoring

For yeast cells with receptor/reporter construct, quantitative analysis of growth

phenotypes and fluorescence development have been carried out with logarithmic
growing cells (diluted to a start OD600 of 0.4) equivalent to 3.25 × 106 cells/ml
(Pharmacia Ultrospec 2000 Spectrophotometer). Recombinant yeast cells were
treated with different concentrations of standards (testosterone, dihydrotestosterone),
non-androgenic steroids (estradiol, progesterone, retinoic acid, dexamethasone), test
chemicals (hydroxyflutamide, cyproterone acetate, vinclozolin, p,p′-DDE, PPME) in
the presence or absence of dihydrotestosterone depending on experiments. Test
compounds were dissolved in DMSO and added to the test cultures of co-transformed
yeast strains in a total volume of 200 µl followed by the incubation at 30 °C. The
growth was obtained by OD600 measurements in 15 min intervals in 96-well plates
using a microplate reader (Tecan, Spectrofluor Plus).
For fluorescence measurements, the excitation wavelength was accustomed to
485 nm and emission was detected at 535 nm in 15 min intervals for 16.5 h. Dose–
response curves were achieved as follows:
166 S. Chatterjee and S.P. Chowdhury

FLUO/OD = (FLUOci − FLUOBci)/(ODci − ODBci), where the fluorescence at

the test concentration (ci) minus the corresponding values for the blank control
(Bci) is divided by growth determined as OD 600 (corrected for blanks).

2.9 Statistical Analysis

The statistical analysis of the results obtained was carried out using the Student’s
T-test. The acceptance level was set at p < 0.05.

3 Results and Discussion

3.1 Establishment of Recombinant Yeast Strain and Level

of Expression of AR

This paper portrays the development of a recombinant yeast-based highly sensitive

(picogram level) fluorescent bioassay for screening different environmental pollu-
tants from PPME. To achieve this purpose, Saccaromyces cerevisiae FYAK26/
8-10B1 was co-transformed with pRS425-Leu2-ARS-hAR and ARE-GFP. The
recombinant FYAK26/8-10B1 strain carrying yEGFP reporter plasmid under the
control of CYC1 promoter, the androgen-dependent interaction between hAR and
ARE can be sensed by the expression of reporter gene. Expression of the trans-
formed hAR cDNA was confirmed by transcriptomic analysis by isolation of total
RNA and RT-PCR amplification (Fig. 1).
Figure 2 demonstrates the Western blot of positive yeast cells that contain a single
copy of the ARE-GFP and pRS425-Leu2-ARS-hAR receptor expression constructs.
An immunoreactive protein band with apparent molecular weight of 110 kDa was
detected in transformant tested indicating the exact translation of the AR protein and
was at par with the band pattern from LNCaP cells, utilized as positive control. The
band was absent in untransformed yeast cells as negative control.
Successful expression of yEGFP in recombinant FYAK26/8-10B1 strain has
been confirmed further by confocal microscopy (Fig. 3).

NT AR LNCaP

Fig. 1 RT-PCR analysis of androgen receptor mRNA expression in yeast cells. The total RNA
isolated from the yeast cells and LNCaP cell line, reverse transcribed and cDNA obtained was
subjected to PCR. NT Nontransformed yeast cells; AR yeast cells transformed with androgen
receptor and LNCaP LNCaP cell line expression as positive control
Development of a Multidrug Transporter … 167

AR LNCaP NT

Fig. 2 Western blot analysis for the expression of androgen receptor in transformed yeast cells. In
the upper panel, the first lane represents AR protein in transformed cells (AR), the second lane is
from LNCaP cells used as positive control (LNCaP) and third lane is nontransformed yeast cell
extract used as negative control (NT)

Fig. 3 Confocal microscopy analysis of yEGFP expression in co-transformed yeast cells exposed
to testosterone

3.2 Dose-Response Curves Obtained with the YEGFP-Based

Yeast Androgen Bioassay

Androgen-response capability analysis of the transformed FYAK26/8-10B1 strain

was performed with increasing concentrations of androgens (Fig. 4). The yEGFP
activity was induced by testosterone (T) and dihydrotestosterone (DHT) in a dose
dependent manner with the saturation of yEGFP induction at around 60 pg/ml in
both the situations. The developed assay unveiled a half maximal effect (EC50) in
the transformed yeast strain at 55 pg/ml for T and 43 pg/ml for DHT, which is well
consistent with the relative AR binding affinity and agonist activity of these
androgens in yeast-based assays [20, 24–26] and mammalian cells [27, 28].

3.3 Dose-Dependent Ligand Specificity in the Transformed

Yeast Cells

Specificity with steroidal and non-steroidal hormones was analyzed by the induction
of yEGFP activity in the recombinant yeast strain. Recombinant FYAK26/8-10B1
strain was incubated with increasing concentrations of T, DHT, 17β-estradiol (E),
progesterone (Prog), all-trans-retinoic acid (RA) and dexamethasone (Dex) followed
by the measurement of their yEGFP activity (Fig. 5). T and DHT notably induced
168 S. Chatterjee and S.P. Chowdhury

Fig. 4 Dose-dependent yEGFP induction by increasing concentrations of T and DHT in

co-transformed yeast cells. The yEGFP activity in the absence of ligand was set as zero. The values
represent the mean ± S.E.M. of four independent experiments each performed in quadruplicates

Fig. 5 Determination of
ligand specificity in
recombinant yeast strains in
response to increasing
concentrations of androgenic
and non-androgenic steroids.
The values represent the
mean ± S.E.M. of four
independent experiments each
performed in quadruplicates

yEGFP activity at a concentration of 50 pg/ml, while the other hormonal compounds

failed to induce noteworthy enzymatic activity in the same range.
Only 17β-estradiol yielded a minor agonistic response. However, this compound
is already known to exert androgenic effects [24, 26]. Dexamethasone was also
depicted as an androgen receptor agonist earlier, showing androgenic activity in few
mammalian assays, but gave a negative result in our yEGFP-based yeast androgen
bioassay.
Development of a Multidrug Transporter … 169

3.4 Anti-androgenic Activity in Transformed Yeast Strain

The specificity of the new yeast androgen bioassay was further demonstrated by the
ability of anti-androgens to suppress the induction of yEGFP. Figure 6 shows the
anti-androgenic activity of the known antagonist flutamide and the other com-
pounds like cyproterone acetate, spironolactone, p,p′-DDE, and vinclozolin. The
antagonistic properties were examined by co-exposure with a concentration of DHT
that induced a submaximal response (50 nM). None of these five compounds were
able to show an agonistic response, but Fig. 6 clearly shows that all three were able
to inhibit the response induced by DHT. Their IC50 values were estimated
approximately as 5.8 µM for flutamide, 6.78 µM for cyproterone acetate (CPA),
9.93 µM for p,p′-DDE and 33.79 µM for vinclozolin (p < 0.05) using our yeast-
based yEGFP assay.

3.5 Detection of Androgenic Activity of PPME

Recombinant FYAK26/8-10B1 cells were incubated on 96-well plates applied with

1:100 dilution of solid phase extracted water sample of PPME at 30 °C and fluo-
rescence was measured in a time-dependent manner (16–24 h) (Fig. 7). Our results

Fig. 6 Anti-androgenic activities of some androgen receptor antagonists in recombinant yeast

cells. The cells were treated either with increasing concentrations of test compounds alone or in the
presence of 50 nM DHT. The yEGFP activities were compared with 50 nM DHT only, which was
set as 100. The values represent the mean ± S.E.M. of four independent experiments each
performed in quadruplicates
170 S. Chatterjee and S.P. Chowdhury

Fig. 7 Detection of androgenic effects of solid phase extracted effluents from five different PPME
samples (I–V) with respect to yEGFP induction by the developed recombinant yeast cells
employing 96-well plate format. The yEGFP activities were evaluated with that obtained with
control (vehicle treated). The values represent the mean ± S.E.M. of four independent experiments
each performed in quadruplicates

showed detectable levels of androgen receptor mediated yEGFP activity at

levels significantly greater than the background (p < 0.05). The data suggests
that the solid phase extracted water samples of all the five PPME samples
comprise potential androgen mimicking compounds and they are in the range of
0.001–1.5 µg/L eq. of T.

4 Conclusions

Earlier transactivation assays with AR and androgen-responsive reporter genes have

been carried out utilizing DU-145, COS, CHO, CV-1, MCF-7 and HepG2 cell lines
[5, 27, 29–32]. The major problem with these assay systems is the expression of
other non-specific endogenous receptors [5]. To overcome this drawback several
attempts have been made with yeast-based assays devoid of endogenous receptors
for screening of estrogenic compounds [21, 33–35], but not much has been unrav-
elled for screening androgenic or anti-androgenic compounds except few reports on
screening different synthetic chemicals like pesticides, fungicides using β-gal or GFP
as reporter genes [20, 25, 26, 33, 34, 36–38]. Above all, the major problem asso-
ciated with these reports is its sensitivity level [10–18]. This is mainly due to the
pleiotropic drug resistance (PDR) network comprises the major determinants of
multiple drug resistance in S. cerevisiae. Thus, induction by a drug or toxin mediates
the up-regulation of gene expressions that encode transporters resulting in drug
extrusion, followed by multidrug resistance. So to enhance the sensitivity of our
yEGFP-based assay (picogram level), the hAR and yEGFP constructs have been
expressed in a yeast strain deleted in the multidrug ABC transporter genes.
Development of a Multidrug Transporter … 171

The present report thus unveils a better model in sensitivity than the other as
reported earlier [24, 25, 36–38]. To the best of our knowledge, this is the only
report, other than Svenson and Allard [19] and our self report (Chatterjee et al.
[20]), where androgenicity of chemicals in PPME has been reported employing a
recombinant yeast-based model with picogram level sensitivity. As the initial
application of yeast-based androgen bioassay, we analyzed some PPME samples in
the northern region of India. Our data revealed detectable levels of androgen
receptor mediated transcriptional activity at levels significantly greater than the
background recorded in the effluents (p < 0.05) indicating its androgenic activity.
Above all, the assay seems to be more robust and more specific for detecting
compounds with a pure androgenic mode of action.

Acknowledgments SC would like to thank CSIR and DAAD for providing research fellowship
in order to carry out this work. We thank Dr. Anand Bachhawat for the vectors, Dr. M. Ghislain for
the the FYAK strain and Prof. Ilpo Huhtaniemi for the kind gift of all steroidal and non-steroidal
test chemicals. We thank Prof. P. Roy, Prof. C.B. Majumder and Prof. M. Höfer for their valuable
technical suggestions whenever needed.

References

1. Pike MC, Spicer DV, Dahmoush L, Press MF (1993) Estrogens, progestogens, normal breast
cell proliferation, and breast cancer risk. Epidemiol Rev 15:17–35
2. Routledge EJ, Parker J, Odum J, Ashby J, Sumpter JP (1998) Some alkyl hydroxy benzoate
preservatives (parabens) are estrogenic. Toxicol Appl Pharmacol 153:12–19
3. Skakkebaek NE, Jørgensen N, Main KM, Rajpert-DeMeyts E, Leffers H, Andersson A, Juul
A, Carlsen E, Mortensen GK, Jensen TK, Toppari J (2006) Is human fecundity declining? Int J
Androl 29:2–11
4. Kelce WR, Wilson EM (2001) Environmental anti-androgens: developmental effects,
molecular mechanisms and clinical implications. J Mol Med 75:198–207
5. Roy P, Salminen H, Koskimies P, Simola J, Smeds A, Sáuco P, Huhtaniemi IT (2004)
Screening of some anti-androgenic endocrine disruptors using a recombinant cell-based
in vitro bioassay. J Steroid Biochem Mol Biol 88:157–166
6. Jobling S, Reynolds T, White R, Parker MG, Sumpter JP (1995) A variety of environmentally
persistent chemicals, including some phthalate plasticizers, are weakly estrogenic. Environ
Health Perspect 103:582–587
7. Keller ET, Ershler WB, Chang C (1996) The androgen receptor: a mediator of diverse
responses. Front Biosci 1:59–71
8. Blankvoort BM, de Groene EM, van Meeteren-Kreikamp AP, Witkamp RF, Rodenburg RJ,
Aarts JM (2001) Development of an androgen receptor gene assay (AR-LUX) utilizing a
human cell line with an endogenously regulated androgen receptor. Anal Biochem 298:93–102
9. Henley DV, Lipson N, Korach KS, Bloch CA (2007) Prepubertal gynecomastia linked to
lavender and tea tree oils. N Engl J Med 356:479–485
10. Bovee TF, Lommerse JP, Peijnenburg AA, Fernandes EA, Nielen MW (2008) A new highly
androgen specific yeast biosensor, enabling optimisation of (Q)SAR model approaches.
J Steroid Biochem Mol Biol 108:121–131
11. Bovee TF, Schoonen WG, Hamers AR, Bento MJ, Peijnenburg AA (2008) Screening of
synthetic and plant-derived compounds for (anti)estrogenic and (anti)androgenic activities.
Anal Bioanal Chem 390:1111–1119
172 S. Chatterjee and S.P. Chowdhury

12. Rijk JC, Bovee TF, Wang S, Van Poucke C, Van Peteghem C, Nielen MW (2009) Detection
of anabolic steroids in dietary supplements: the added value of an androgen yeast bioassay in
parallel with a liquid chromatography-tandem mass spectrometry screening method. Anal
Chim Acta 637:305–314
13. Bovee TF, Thevis M, Hamers AR, Peijnenburg AA, Nielen MW, Schoonen WG (2010)
SERMs and SARMs: detection of their activities with yeast based bioassays. J Steroid
Biochem Mol Biol 118:85–92
14. Plotan M, Elliott CT, Scippo ML, Muller M, Antignac JP, Malone E, Bovee TF, Mitchell S,
Connolly L (2011) The application of reporter gene assays for the detection of endocrine
disruptors in sport supplements. Anal Chim Acta 700:34–40
15. Rijk JC, Ashwin H, van Kuijk SJ, Groot MJ, Heskamp HH, Bovee TF, Nielen MW (2011)
Bioassay based screening of steroid derivatives in animal feed and supplements. Anal Chim
Acta 700:183–188
16. Rijk JC, Bovee TF, Peijnenburg AA, Groot MJ, Rietjens IM, Nielen MW (2012) Bovine liver
slices: A multifunctional in vitro model to study the prohormone dehydroepiandrosterone
(DHEA). Toxicol In Vitro 26:1014–1021
17. Reitsma M, Bovee TF, Peijnenburg AA, Hendriksen PJ, Hoogenboom RL, Rijk JC (2013)
Endocrine-disrupting effects of thioxanthone photoinitiators. Toxicol Sci 132:64–74
18. de Rijke E, Essers ML, Rijk JC, Thevis M, Bovee TF, van Ginkel LA, Sterk SS (2013)
Selective androgen receptor modulators: in vitro and in vivo metabolism and analysis. Food
Addit Contam Part A Chem Anal Control Expo Risk Assess 30:1517–1526
19. Svenson A, Allard AS (2004) In vitro androgenicity in pulp and paper mill effluents. Environ
Toxicol 19:510–517
20. Chatterjee S, Majumder CB, Roy P (2007) Development of a yeast-based assay to detrmeine
the (anti)androgenic contaminants from pulp and paper mill effluents in India. Environ Toxicol
Pharmacol 24:114–121
21. Sievernich A, Wildt L, Lichtenberg-Frate H (2004) In vitro bioactivity of 17alpha-estradiol.
J Steroid Biochem Mol Biol 92:455–463
22. Ausubel FM, Brent R, Kingston RE, Moore DD, Seidman JG, Smith JA, Struhl K (1995)
Current protocols in molecular biology. Greene Publishing Associates and Wiley-Interscience,
New York
23. Laemmli UK (1970) Cleavage of structural proteins during the assembly of the head of
bacteriophage T4. Nature 227:680–685
24. Gaido KW, Leonard LS, Lovell S, Gould JC, Babai D, Portier CJ, McDonnell DP (1997)
Evaluation of chemicals with endocrine modulating activity in yeast-based steroid hormone
receptor gene transcription assay. Toxicol Appl Pharmacol 143:205–212
25. Lee HJ, Lee YS, Kwon HB, Lee K (2003) Novel yeast bioassay system for detection of
androgenic and antiandrogenic compounds. Toxicol In Vitro 17:237–244
26. Michelini A, Leskinen P, Virta M, Karp M, Roda A (2005) A new recombinant cell-based
bioluminescent assay for sensitive androgen-like compound detection. Biosens Bioelectrons
20:2261–2267
27. Kemppainen JA, Langley E, Wong CI, Bobseine K, Kelce WR, Wilson EM (1999)
Distinguishing androgen receptor agonists and antagonists: distinct mechanisms of activation
by medroxyprogesterone acetate and dihydrotestosterone. Mol Endocrinol 13:440–454
28. Terouanne B, Tahiri B, Georget V, Belon C, Poujol N, Avances C Jr, Orio F, Balaguer P,
Sultan C (2000) A stable prostatic bioluminescent cell line to investigate androgen and
antiandrogen effect. Mol Cell Endocrinol 160:39–49
29. Lobaccaro JM, Poujol N, Terouanne B, Georget V, Fabre S, Lumbroso S, Sultan C (1999)
Transcriptional interferences between normal or mutant androgen receptors and the activator
protein 1-dissection of the androgen receptor functional domains. Endocrinology 140:350–357
30. Vinggaard AM, Joergensen EC, Larsen JC (1999) Rapid and sensitive reporter gene assays for
detection of antiandrogenic and estrogenic effects of environmental chemicals. Toxicol Appl
Pharmacol 155:150–160
Development of a Multidrug Transporter … 173

31. Xu LC, Sun H, Chen JF, Bian Q, Qian J, Song L, Wang XR (2005) Evaluation of androgen
receptor transcriptional activities of bisphenol A, octylphenol and nonylphenol in vitro.
Toxicology 216:196–203
32. Sun H, Xu X-L, Xu L-C, Song L, Hong X, Chen J-F, Cui L-B, Wong X-R (2007)
Antiandrogenic activity of pyrethroid pesticides and there metabolite in reporter gene assay.
Chemosphere 66:474–479
33. Bovee TF, Helsdingen RJ, Koks PD, Kuiper HA, Hoogenboom RL, Keijer J (2004)
Development of a rapid yeast estrogen bioassay, based on the expression of green fluorescent
protein. Gene 325:187–200
34. Bovee TF, Helsdingen RJ, Rietjens IM, Keijer J, Hoogenboom RL (2004) Rapid yeast
estrogen bioassays stably expressing human estrogen receptors α and β, and green fluorescent
protein: a comparison of different compounds with both receptor types. J Steroid Biochem Mol
Biol 91:99–107
35. Wozei E, Hermanowicz SW, Holman H-YN (2006) Developing a biosensor for estrogens in
water samples: study of the real-time response of live cells of the estrogen sensitive yeast strain
RMY/ER-ERE using fluorescence microscopy. Biosens Bioelectron 21:1654–1658
36. Sohoni P, Sumpter JP (1998) Several environmental estrogens are also antiandrogens.
J Endocrinol 158:327–339
37. Beck V, Pfitscher A, Jungbauer A (2005) GFP-reporter for a high throughput assay to monitor
estrogenic compounds. J Biochem Biophys Methods 64:19–37
38. Bovee TFH, Helsdingen RJR, Hamers ARM, van Duursen MBM, Nielen MWF, Hoogenboom
RLAP (2007) A new highly specific and robust yeast androgen bioassay for the detection of
agonists and antagonists. Anal Bioanal Chem 389:1549–1558
Simulation of ICA-PI Controller of DC
Motor in Surgical Robots for Biomedical
Application

Milan Sasmal and Rajat Bhattacharjee

Abstract The medical device industry is considered a hub of innovation with

constantly improving designs and processes, the possibilities seem endless. Over
the past 20 years, dc motors have played its role in the development of medical
instruments like surgical robots, plate readers, liquid and specimen handling sys-
tems, chromatography, and In Vitro diagnostic machines. Regardless of what ever
the application may be, all devices in the medical industry should have high
Reliability and flexibility. This paper presents a new controller model using
Imperialist Competitive Algorithm (ICA) for efficient search and optimization of
Proportional Integral controller parameter in order to achieve a better close loop
control of the dc motor in bio-medical application. This control method was sim-
ulated using MATLAB/SIMULINK to control dc motor. From the simulation, it
was found that the proposed method gives better performance with improved
response in any user defined function as the settling time, overshoot and speed error
are reduced.

Keywords Surgical Robot DC motor Speed control

Controller
Optimi-


zation Imperialist competitive algorithm (ICA)

1 Introduction

Now a day’s high performance motor drives are very essential for medical appli-
cations especially in surgical robots, liquid and specimen handling system. A high
performance motor drive system must have good dynamic speed command tracking
and load regulating response. DC motors provide excellent control of speed for

M. Sasmal (&)
R. Bhattacharjee

Electrical Engineering Department, JIS College of Engineering, Kalyani, Nadia, India
e-mail: [email protected]
R. Bhattacharjee
e-mail: [email protected]

© Springer India 2015 175

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_16
176 M. Sasmal and R. Bhattacharjee

acceleration and deceleration. Separately excited DC motor has better voltage

variation, so it is used for speed and torque control applications.
DC drives, because of their simplicity, ease of application, reliability and
favorable cost have long been a backbone of industrial applications as well as
medical applications as compared to AC drive. Motion of the drive can be controlled
smoothly by adjusting proper parameters of the speed controller. Here the controller
used is basically a Proportional Integral type which removes the delay and provides
fast control. Modeling of separately excited DC motor is done by using MATLAB
7.0 version 2007. The controller value is optimized [1] by using an artificial intel-
ligence technique called Imperialist Competitive Algorithm (ICA) [2–7].
Furthermore, the motor is able to run smoothly at an extended speed range by the
help of reference speed signal. There is no need to hardware control, only by
changing the reference speed signal we can control the device. Thus we can get a
computer controlled automated system. Specifically, we aim to maximize the
smoothness and the speed with which the tasks are performed. Increased smooth-
ness reduces damage to human tissue and the robotic mechanism, and increased
speed reduces operation time.

2 Surgical Robot

In 1979, the Robot Institute of America, an industrial trade group, defined a robot
[8] as “a reprogrammable, multi- functional manipulator designed to move mate-
rials, parts, tools, or other specialized devices through various programmed motions
for the performance of a variety of tasks.” Such a definition leaves out tools with a
single task (e.g., stapler), anything that cannot move (e.g., image analysis algo-
rithms), and nonprogrammable mechanisms (e.g., purely manual laparoscopic
tools). As a result, robots are generally indicated for tasks requiring programmable
motions, particularly where those motions should be quick, strong, precise, accu-
rate, untiring, and/or via complex articulations. The greatest impact of medical
robots has been in surgeries, both radiosurgery and tissue manipulation in the
operating room, which are improved by precise and accurate motions of the nec-
essary tools. Through robot assistance, surgical outcomes can be improved, patient
trauma can be reduced, and hospital stays can be shortened, though the effects of
robot assistance.
Surgical robots cover the following multiple next-generation systems on the
future of the medical field [8]
• The expected benefit of robot assistance in orthopedics is accurate and precise
bone resection
• General Laparoscopy
• Percutaneous
• Steerable Catheters
• Radiosurgery
Simulation of ICA-PI Controller of DC Motor … 177

• Assistive and Rehabilitation Systems

• Prosthetics and Exoskeletons
• Emergency Response—Few medical robot systems are suitable for use outside
of the operating room, despite significant research funding on medical devices
for disaster response and battlefield medicine.

3 Circuit Model and Procedure

The block diagram of the model is shown in Fig. 1. DC motor is supplied from a
converter and the feedback of the motor is fed to a comparator after a filtration. The
error signal generated from comparator is fed to the controller [9] and the controller
output to the motor.
In the first circuit model, the dc motor is driven by a group of four mosfets as
shown in Fcircuit is modified as shown in Fig.2, two mosfets are for two quadrants
and rest two for two other quadrants i.e. four quadrant operation. Again these
mosfets are controlled by the gate signal generating from Pulse Width Modulation
block (PWM). To design a close loop control, Speed feedback is taken from the
measuring terminal of the motor through de-multiplexer, then is fed to the speed
controller (PI controller), again the output of PI controller [1] and armature current
feedback signal is fed to the current controller. The output of the current controller
is feed to the PWM as input. The PWM transfers the reference signal to the
corresponding pulses which will be fed to the two mosfet and the inverted pulses of
the PWM output is feed to other two mosfets. The mosfets will be turned on-off
according to the pulses.
For optimization the previous circuit is modified as shown in Fig. 3. To calculate
the instantaneous error, an error signal is generated by a comparison between the
motor speed and reference speed signal by the comparator. The error signal is fed to
the controller block and the output of which is fed to the MATLAB base workspace
for calculation of optimization. During the optimization the controller parameter is
again fed from base workspace to the controller of the model again in each step of
optimization. After a complete process of the ICA [6], the final accurate value of the
parameters are fed to the controller and system runs with this (Table 1).

Fig. 1 Block diagram of the model

178 M. Sasmal and R. Bhattacharjee

Fig. 2 Circuit diagram of speed controller before optimization

Fig. 3 Complete circuit diagram of speed controller for optimization

Simulation of ICA-PI Controller of DC Motor … 179

Table 1 DC motor parameter

Armature resistance 0.232 ohm
Armature inductance 0.000557 H
Field resistance 78.91 ohm
Inertia 0.2053 * e−5 kg.m2
Load torque 1 (constant)
Coulomb friction 5.282 N.m
Initial speed 0 rad/s

4 ICA to Calculate the Controller Parameter

The Imperialist Competitive algorithm is a technique of Artificial Intelligence for

optimization, inspired by the imperialistic competition. Like other evolutionary
ones, the algorithm starts with an initial population. Population individuals called
country are in two types: colonies and imperialists that all together form some
empires. Imperialistic competition among these empires forms the basis of the
evolutionary algorithm. During this competition, weak empires collapse and
powerful ones take possession of their colonies. Imperialistic competition hopefully
converges to a state in which there exist only one empire and its colonies are in the
same position and have the same cost as the imperialist. Applying this algorithm to
some of benchmark cost functions and in a expert system based on MATLAB
shows its ability to drive an computer operated surgical robot and many more things
which are used in medical application (Fig. 4).

5 Results and Discussion

In our experiment, our goal is to drive the system in a pre-defined motion which can
be used in a surgical robot. To achieve the user defined motion i.e. speed we use a
reference speed signal (speed signal 2) which is shown in Fig. 5, where we assume
that the drive has no initial speed and starts to accelerate up to 2 s and reaches a
speed of 100 revolution per minute, after that it maintain a constant speed and after
3 s it starts to decelerate and again reaches a speed of 100 revolution per minute in
opposite direction and again maintain a constant speed between 7 and 8 s.
The output results of the motor corresponding to the reference speed signal 2
before optimizing the value of PI controller, the output follows the reference speed
signal but the output consist with a large error including a settling time of 0.3 s
approximate and speed error of 4.5 % (Fig. 6).
After optimizing the value of PI controller, the output results of the motor shown
in Fig. 7 also follows the corresponding reference speed signal 2 where error is
minimized and settling time decreases to 0.035 s approximate with a very small
speed error.
180 M. Sasmal and R. Bhattacharjee

Fig. 4 Flowchart of ICA

Start

Initialize the empires

Move the colonies toward

their relevant imperialist

Is there a colony in an empire which

has lower cost than that of imperialist No
Yes

Exchange the positions of

that imperialist and colony

Compute the total cost of all empires

Pick the weakest colony from the weakest

empire and give it to the empire that has the
most likelihood to possess it

Is there an empire
No
with no colonies
Yes

Eliminate this empire

No Stop condition satified

Yes

Done

An another user defined motion is taken to drive the system, where reference
speed signal 3 is used and is shown in Fig. 8, where we assume that the drive has no
initial speed and has stepped accelerated motion from 0 to 100 revolution per
Simulation of ICA-PI Controller of DC Motor … 181

Fig. 5 Reference speed signal 2

Fig. 6 Speed-time characteristics of the motor for reference sig 2 before optimization

Fig. 7 Speed-time characteristics of the motor for reference sig 2 after optimization
182 M. Sasmal and R. Bhattacharjee

Fig. 8 Reference speed signal 3

minute and also has a constant speed for a while and then decelerate to 50 revo-
lution per minute in reverse direction.
The output results of the motor corresponding to the reference speed signal 3
before optimizing the value of PI controller, output shown in Fig. 9 follows the
reference speed signal 3 but output consist a overshoot of 3 % which is very large
for sophisticated medical instrument.
Without optimizing the value of PI controller the motor output response involves
a huge steady state error including long settling time, rise time with over shoot in
large magnitude, shown in Fig. 9.
Figure 10 is the output results of the motor corresponding to the reference speed
signal 3 after a complete optimizing the value of PI controller, where the overshoot
is completely minimized with a percent of 0.12.
The reference input speed time curve (Figs. 5 and 8) is taken from 0 to 10 s.
Between the two reference speed time curve first one is smooth controlling in both
direction and the second one (Fig. 8) is step motion.

Fig. 9 Speed-time characteristics of the motor for reference sig 3 before optimization
Simulation of ICA-PI Controller of DC Motor … 183

Fig. 10 Speed-time characteristics of the motor for reference sig 3 after optimization

Table 2 Performance criteria of the drive

Parameter values and their response Before optimization After optimization
Kp 9.971 6.174
Ki 3.854 0.021
Settling time (s) 0.28 0.035
Overshoot (%) 3 0.12
Speed error (%) 4.5 1.14

After optimization the over shoot has completely eliminated. Settling time and
rise time has reduced to great extent. Steady state error has been reduced which is
shown graphically (Table 2).

6 Conclusion

As the Surgical robot and medical appliances are very sophisticated tools so we
need very smooth and fast control with high reliability. This paper presents a new
design of controller for surgical robot and other medical appliances using ICA
based optimization to optimize the controller parameters. The comparison between
PI controller without and with ICA implemented has been made. From the result it
is shown that ICA method can be used to optimize PI controller is more efficient
and give better performance for motion system in biomedical appliances with
184 M. Sasmal and R. Bhattacharjee

reduced settling time, overshoot and speed error of the system. Further research is
intended to be focused in the application of the proposed method for designing the
controller to control multiple surgical devices in biomedical application.

References

1. Safura Hashim NL, Yahya A, Andromeda T, Abdul Kadir MR, Mahmud N, Samion S (2012)
Simulation of PSO-PI controller of dc motor in micro-EDM system for biomedical application.
ELSEVIER Procedia Eng 41:805–811
2. Haupt RL, Haupt SE (2004) Practical genetic algorithms, 2nd edn. Wiley, Hoboken
3. Melanie M (1999) An introduction to genetic algorithms. MIT Press, Massachusetts
4. Dorigo M, Blum C (2005) Ant colony optimization theory: a survey. Theoret Comput Sci
344:243–278
5. Johnston RL, Cartwright HM (2004) Applications of evolutionary computation in chemistry.
Springer, Berlin
6. Gargari EA, Lucas C (2007) Imperialist competitive algorithm: an algorithm for optimization
inspired by imperialistic competition: control and intelligent processing center of excellence
(CIPCE). School of Electrical and Computer Engineering, University of Tehran, North Kargar
Street, Tehran, Iran
7. Varol HA, Bingul ZA (2004) New PID tuning technique using ant algorithm, In: Proceeding of
the 2004 American control conference, Boston, Massachusetts, June 30–July 2 (2004)
8. Beasley RA (2012) Medical robots—current systems and research directions: Hindawi
Publishing Corporation. J Robot 2012(401613):14. doi:10.1155/2012/401613
9. Nasri M, Maghfoori M (2007) Pso—based optimum design of pid controller for a linear
brushless dc motor. World Acad Sci Eng Tech 26:211–215
Development of a Wireless Attendant
Calling System for Improved Patient Care

Debeshi Dutta, Biswajeet Champaty, Indranil Banerjee,

Kunal Pal and D.N. Tibarewala

Abstract The present proposal revolves around the fabrication of a finger move-
ment based wearable wireless attendant calling system. The system comprised of a
flex sensor and a hall-effect sensor coupled with Arduino UNO and worked syn-
chronously with patient hand movement. The concurrent activation of both the
sensors enables the conveyance of patient location (ward and bed numbers) to the
nurse station through Xbee protocol and a one-way SMS correspondence to a
preloaded mobile number through GSM protocol. The device is capable of handling
multiple patient requests at a time with minute time interval. A graphical user
interface in MATLAB program monitors the patient status at the nursing station.
The proposed device is expected to improve the quality of the patient care.

Keywords Wearable
SMS
Wireless
Attendant calling system

1 Introduction

There is an acute shortage of nursing staff across the globe. The condition is worse
in the developing and under-developed countries. As per the Nursing Council of
India (NCI), only 40 % of the registered nursing staffs are actively working. This
has resulted in the decrease in the nurse: patient ratio. The suggested ratio is 1:1, 1:3
and 1:6 in the critical care unit, intermediate care unit and general ward, respec-

D. Dutta
B. Champaty
I. Banerjee (&)
K. Pal (&)

Department of Biotechnology and Medical Engineering,
NIT-Rourkela, Rourkela, Odisha 769008, India
e-mail: [email protected]
K. Pal
e-mail: [email protected]
D.N. Tibarewala
School of Bioscience and Engineering, Jadavpur University,
Kolkata 700032, India

© Springer India 2015 185

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_17
186 D. Dutta et al.

tively. Since the duty of the nursing staff occurs in 3 shifts, the shortage of the
nursing staff looks grimmer. As per the recent report, in many public hospitals the
nurse: patient ratio hovers around 1:60 during the evening and the night shifts. This
puts a stress on the on-duty nurses. The stress and the work pressure have been
reported to be the major deterrent for the nurses to join as a nursing staff. The afore-
mentioned factors hamper the patient care to a great extent. Keeping this in mind, in
the current study we propose to develop a low cost attendant calling system which
can help reducing the work pressure for the nurses. Generally, the nurses have to be
regularly at round to get an insight about the need of the patients. We have devised
a prototype finger-movement based attendant calling system. The prototype
developed is a wearable device which can wirelessly transmit signal to the nursing
station and can inform the healthcare staffs about the need of the patient [1–3].

2 Materials

Xbee-S2 shield (Digi International, USA), flex sensor (Sparkfun, USA), hall-effect
sensor (Evelta Electronics, India), GSM-GPRS shield (Seedstudio, China), Arduino
UNO (Arduino, Italy) microcontroller boards, 16x2 LCD (Sunrom Technologies,
India) and Emic 2 Text-to-speech module (Parallax Inc., USA) are the major
components used in this study.

3 Methodology

The developed device consists of two units: (a) transmission unit and (b) receiver
unit. The construction of the complete device has been described below.

3.1 Transmission Unit

The transmission unit consists of three input interfaces, namely, flex sensor, hall-
effect sensor and a push button. The signals from the input interfaces are fed into an
Arduino UNO microcontroller for decision making and triggering of the commu-
nication protocol. Two types of communication protocol were used in this study:
one being the GSM protocol to send SMS to the nursing staff on activation and the
other being Xbee based wireless transmission of the alert signal to the nursing
station. The transmitter unit was a wearable device [3, 4]. The setup was assembled
on a wearable hand-glove. The circuit was powered with a 9 V rechargeable battery.
Development of a Wireless Attendant Calling System … 187

3.2 Receiver Unit

The receiver unit consists of an Xbee receiver connected to Arduino UNO mi-
crocontroller. The microcontroller was interfaced with a laptop. A GUI (graphical
user interface) based MATLAB program was made to monitor the status. The
microcontroller was also connected to a LCD panel and a text-to-speech module
[1, 3, 4]. The schematic diagram and the gist of functioning of the proposed device
has been shown in Fig. 1.

3.3 Testing of the Proposed Device

The device has been developed for application under hospital environment. Two
modules of the transmission units were developed to mimic a two bed situation in a
ward.

Fig. 1 Schematic diagram and functioning of the proposed device

188 D. Dutta et al.

4 Results and Discussions

An easy to use wearable attendant calling system was devised based on the
movement of the finger. The device consists of two sub-units, namely, transmission
unit and receiver unit. The transmission unit consists of two sensors (flex sensor and
Hall-effect (HE) sensor). Flex sensor is a resistive sensor, which changes its
resistance when it is bent. The sensor was attached over the index finger region of a
hand-glove such that when the index finger is flexed there is a change in resistance
[5, 6]. The change in the resistance was monitored in the microcontroller, which
generated control signals when the resistance increased beyond a threshold level.
But there are chances of accidental switching of the device when the patient is
carrying out other day-to-day activities. Hence, a magnet was attached towards the
arm-end of the glove [5, 7, 8].
The magnet was used to activate the HE sensor. The HE sensor was used as a
switching device. The program, monitoring the change in the resistance of the flex
sensor, was modulated to generate the control signals only when the outputs from
both the sensors were in a high state. The control signals were used to drive Xbee
and GSM shields. The Xbee shield wirelessly transmitted the control signals to the
nursing station. On the other hand, GSM shield sent an alert SMS to a specific
mobile. The alert SMS contained the location of the patient, i.e. ward number and
bed number. The schematic diagram of the developed wearable attendant calling
system is shown in Fig. 2.
The control signals transmitted by the wearable device are received by the
receiver unit (at the nursing station). The receiver unit consists of Xbee, which
receives the control signals. The signals were acquired in the Arduino UNO for
classification. The classified signals were used to drive the LCD panel and text-to-
speech module to display and announce the ward and bed numbers of the patient,
respectively (Fig. 3). The classified signals were also acquired in a computer.
A GUI based MATLAB program was made to display the ward and the bed
numbers [9].

Fig. 2 The developed

wearable attendant calling
system (transmitter unit)
Development of a Wireless Attendant Calling System … 189

Fig. 3 Receiver unit connected with the LCD and the text-to-speech modules

Fig. 4 Volunteers activating the prototype attendant calling units

190 D. Dutta et al.

Fig. 5 SMS sent by the wearable device to the specified mobile number

The developed prototype of the device was tested using two transmitter units.
Two volunteers were invited to participate in the study. The volunteers were trained
on the transmitter units for 10 min. Thereafter, the wearable transmission units were
put on the volunteers. The volunteers were advised to activate the transmission
units. The volunteers were able to easily activate the device. Figure 4 shows the
activation of the LCD, the text-to-speech and MATLAB based GUI display
modules at the nursing station. Figure 5 shows the activation of the GSM module
for sending SMS informing the ward and the bed number to the healthcare provider
[3, 7, 10].

5 Conclusion

The current study describes about the development of a user friendly wearable
device for attendant calling. There is a need for the devices which can assist the
patients to call the nurses when there is an emergency situation [1, 10]. The pro-
posed device will not only help the patients but also will be helpful for the nurses in
their day-to-day activities of the professional life [3].

References

1. George WR et al (1991) Communication system for deaf, deaf-blind, or non-vocal individuals

using instrumented glove. Google Patents (ed)
2. Jafari R et al (2005) Wireless sensor networks for health monitoring. In: Mobile and
ubiquitous systems: networking and services. The second annual international conference on
MobiQuitous, 2005, pp 479–481
3. Sahoo S et al (2014) Wireless transmission of alarm signals from baby incubators to neonatal
nursing station. In: First international conference on automation, control, energy and systems
(ACES), 2014, pp 1–5
4. Borges LM et al (2009) Smart-clothing wireless flex sensor belt network for foetal health
monitoring. In: 3rd international conference on pervasive computing technologies for
healthcare (PervasiveHealth 2009), 2009, pp 1–4
Development of a Wireless Attendant Calling System … 191

5. Simone L et al (2004) A low cost method to measure finger flexion in individuals with reduced
hand and finger range of motion. In: 26th Annual international conference of the IEEE
engineering in medicine and biology society IEMBS’04, 2004, pp 4791–4794
6. Saggio G (2012) Mechanical model of flex sensors used to sense finger movements. Sens
Actuators, A 185:53–58
7. Connolly J et al (2012) A new method to determine joint range of movement and stiffness in
Rheumatoid Arthritic Patients. In: Annual international conference of the IEEE engineering in
medicine and biology society (EMBC), 2012, pp 6386–6389
8. Grimaud JJG et al (1999) Tactile feedback mechanism for a data processing system. Google
Patents (ed)
9. Kim JH et al (2005) Hand gesture recognition system using fuzzy algorithm and RDBMS for
post PC. In: Fuzzy systems and knowledge discovery. Springer (ed), pp 170–175
10. Zimmerman TG et al (1987) A hand gesture interface device. In: ACM SIGCHI Bulletin
pp 189–192
A Review on Visual Brain Computer
Interface

Deepak Kapgate and Dhananjay Kalbande

Abstract The primary aim of brain computer interface (BCI) is to establish

communication link between computers and severely disable people those are
partially or totally paralyzed (locked in state) due to neurological disorders.
Recently much attention is focused on visual BCI (V-BCI) because of its low cost
and minimal user training time. Characteristics of visual stimuli plays important
role in V-BCI as it decides the strength of visual signals in brain and system
communication rate. Recently new concept called Hybrid V-BCI emerge based on
concept of combining advantages of one or more basic V-BCIs to further enhance
ITR and accuracy of system. In this paper we review properties of brain signals
involved in V-BCI, stimulation methods used, hybrid V-BCI and discuss current
challenges in V-BCI with their possible solutions.

Keywords Visual BCI Stimulation methods Visual signals
Hybrid V-BCIs



Information transfer rate Repetitive visual stimuli

1 Introduction

Brain computer interface (here after as BCI) also referred as brain machine interface
allow users to communicate with computer or external devices without any mus-
cular movement and provide interface to allow cerebral activity alone to convey
message and commands to computers [1]. Out of all neuroimaging methods
available for monitoring brain activity electroencephalography (EEG) is most
widely used and best neuroimaging technique due to its high portability, relative

D. Kapgate (&)
Nagpur University, Nagpur, India
e-mail: [email protected]
D. Kalbande
Department of C.T., S.P.I.T., Mumbai, India
e-mail: [email protected]

© Springer India 2015 193

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_18
194 D. Kapgate and D. Kalbande

low cost, high temporal resolution and few risks to users. EEG is direct cerebro-
electrical activity measurement technique, having approximate *0.05 s temporal
resolution and *10 mm spatial resolution [2].
There are four conventional non-invasive EEG based BCI paradigms based on
type of brain potential they are using for command generation as visual evoked
potentials (VEPs) based BCI, Slow cortical potentials (SCP) based BCI, P300
evoked potentials (P300) based BCI and sensorimotor rhythms (ERD/ERS) based
BCI [3]. Conventional EEG based BCI types e.g. VEPs based BCI uses only one
brain signal for command generation but recently this concept overcome by
researchers as potential BCI applications go far beyond clinical settings leads to
generation of new subtypes of BCIs as active BCI [4], Reactive BCI [5], Passive
BCI [6], Emotional BCI [7], Collaborative BCI [8], Visual BCI [9], Auditory BCI
[10]. There is no general consensus about whether the new BCI subtypes conform
to original BCI definition.
Active BCI is a BCI which derives its output as control commands from brain
activity which is directly consciously controlled by user, independent from external
events called active BCI [4]. Reactive BCI is based on external stimuli generated
brain activity to generate its output, which is indirectly modulated by the user for
controlling an application [5]. Passive BCI derives its output from arbitrary brain
activity without the purpose of voluntary control [6]. Emotional BCI is BCI which
generates its output as control commands from brain activity which can provide
significant insight into user’s emotional state [7]. Collaborative BCI is BCI which
generates its output by integrating brain activity information from multiple users
[8]. Visual BCI (here after referred as V-BCI) is based on external visual stimuli
generated brain signal modulations in visual cortex to generate its control com-
mands [9]. Auditory BCI is based on endogenous potentials linked to reaction to
external auditory stimuli to generate its control commands.
Recent study shows that many classical V-BCI paradigms demonstrate the
promising prospect of real life BCI application. First Farwell and Donchin [11]
demonstrated 6 × 6 matrix visual speller system based on P300 evoked potential in
1988. In first decade of new century number of research groups working on V-BCI
and V-BCI focused scientific publications increased tremendously [12]. This is due
to V-BCI systems provide high communication speed and classification accuracy
up to approximately 95 %, with minimal or none user training.
In this review is focused only on V-BCI, Sect. 2 explains different brain signals
involved in V-BCI, Sect. 3 discusses stimulation methods used in V-BCI. Section 4
discusses Hybrid V-BCI; Sect. 5 explains challenges and solutions of current V-
BCI systems.

2 Brian Signals in V-BCI

The brain signals commonly used in V-BCI are as follows –

A Review on Visual Brain Computer Interface 195

2.1 Brain Signals Modulated by Exogenous Stimuli

(1) Visual evoked potential (VEP)—The subtypes of VEPs include in V-BCI are
transient—VEPs (TVEPs), steady state visual evoked potentials (SSVEPs) and
Motion onset SSVEPs (M-SSVEPs). VEPs are generated by external visual stim-
ulus at visual cortex. Theses brain activity modulations are relatively easy to detect
as amplitude of VEPs depends on external stimuli [13]. Transient—VEPs occur
with lower visual stimulation frequency (<6 Hz) while SSVEPs occurs with higher
visual stimulation frequency (>6 Hz) [14]. SSVEP based V-BCI further classified as
(i) time modulated VEP (t-VEP based V-BCI) where the external flash visual
sequences of different targets are orthogonal in time [2]. (ii) Frequency modulated
VEP (f-VEP based V-BCI) where each target flashed at unique frequency [15].
Middendorf et al. [16] first developed frequency modulated VEP based BCI with
higher transfer rates. (iii) Pseudorandom code modulated VEP (c-VEP) based BCI
where pseudorandom sequence determine the duration of ON/OFF state of target
flash [17, 18]. Sutter [17] first demonstrated m-sequence c-VEP based BCI with
very high communication rate of 10–12 words/min. Recently Kimura et al. [18] use
FSK-modulated visual stimuli to achieve higher ITR.
(2) Flash TVEPs—It present a series of positive and negative peaks. The most
prominently peaks are negative N1 and N2 peaks at around 40 and 90 ms and
positive P1 and P2 peaks at around 60 and 120 ms respectively. Lee et al. [19] uses
flicker stimulus to generate Flash VEPs to generate N1, P1, N2, P2 responses.
Further Hong et al. [20] uses moving line stimulus to implement speller BCI with
negative (N2) and positive (P2) potentials. Yuan et al. [21] developed collaborative
BCI with visual Go/noGO stimulus to generate N2 signals in multiple subjects.
Steady state motion VEP based BCI rely on paradigm where human perception of
motion oscillates in two opposite direction [22]. Recently [23] demonstrated hybrid
V-BCI by combining P300 + SSVEP signals to generate BCI with higher accuracy
and improved information transfer rate (ITR).

2.2 Endogenously Modulated Brain Signals

(1) P300 evoked Potentials—Infrequent visual, auditory or somatosensory stimuli

evokes endogenic responses about 300 ms after attending oddball stimulus called
P300 potentials. According to [24] the accuracy and ITR of P300 based BCI is
largely based on classifier using by BCI. Researchers developed oddball paradigm
using P300 evoked potentials for large applications like visual speller, smart home
control, mental prosthesis control, entertainment and so on [2].
(2) N400 event related potentials (ERPs)—These ERPs are related to cognitive
responses to meaningful words in sentence. It is negative Peaks evoked about
400 ms post-stimulus onset. Principe [25] developed N400 ERP based BCI that use
users thoughts to move a ball on computer screen.
196 D. Kapgate and D. Kalbande

(3) N170 Event Related Potentials (ERPs)—These are evoked (negative peaks)
130–200 ms after stimulus presentation. N170 ERPs represent neural response to
faces (generally stimuli consist of images of faces). Zhang et al. [26] developed
hybrid BCI combining N170 and P300 ERPs sensitive to configure processing of
human faces.
Studies shown that hybrid V-BCI i.e. combining brain signals evoked due to
visual stimuli and pseudo random code modulated VEP (c-VEP) based BCI are
suitable for real life applications due to their higher communication speed (higher
ITR), no/minimal user training time and also user acceptability as compared to
other BCI types.

3 Stimulation Methods Used in V-BCI

In all V-BCI research there are three types of repetitive visual stimuli present as:

(A) Light Stimuli—LEDs, fluorescent lights and Xe-lights are used to generate light
stimuli which are modulated at specified frequency. V-BCI system using flickering
LED stimuli achieved a higher information transfer rate approximately about 68 bis/
min [27]. The important factor of light stimuli that affect visual signal modulation
are intensity of light stimulus, light luminance, background luminance, illumination
sequence and frequency of illumination. First SSVEP based BCI that uses fluo-
rescent light to render brain stimuli was presented in 1996 [28].
(B) Single graphic stimuli—Mostly computer screens with single graphic in form of
squares, rectangle and arrows [29] that appears from and disappear into background
used in single graphic stimuli. The performance of such BCIs is mainly depending
on color of stimulus and frequency of stimulus (stimulus rate); also these parameter
affects user safety and comfort fatigues and commercial acceptability of V-BCI.
(C) Patter reversal stimuli—Mostly used in transient—VEP based BCI research. It
consists of graphical patterns that are alterated in certain oscillations e.g. check-
erboards [30], line boxes, moving lines and so on.
Direct comparison of V-BCI performance based on different stimuli is difficult as
there are number of other factors that affect BCI performance. Unfortunately most of
articles fail to mention BCI performance (ITR) or did an offline analysis [31]. So in this
review paper we are doing comparison based on frequency of visual stimulus that are
broadly classifies into three frequently bands as Low (1–12 Hz), Medium (12–30 Hz)
and high (30–60 Hz) bands. Table 1 demonstrates the comparison as shown.

4 Hybrid Visual BCI

In recent BCI research term evolved called “Hybrid BCI” where two or more
conventional BCIs are combined in either simultaneous or sequential system
Table 1 Different stimulation methods used in visual BCI
Frequency band Device Visual stimuli type Response Color References Bit rate
(bits/min)
Low (1–12 Hz) LED Light stimuli SSVEP – [44] –
Green [45] –
LCD /CRT Single graphic stimuli SSVEP White/black [46] –
Green [29] –
Row-column paradigm P300 White/black [11] –
Single character paradigm [47] –
Pattern reversal stimuli [48] –
(checker board)
Pattern reversal stimuli SSVEP [49]
(checker board) [50] 10.3
A Review on Visual Brain Computer Interface

LCD Flashing stimuli FVEPs – [19] 23.06

Moving line N2, P2 Red, green, blue, [20] –
purple, yellow
CRT Visual oddball Protocol P300 – [51] 29.35
Meaningful words N400 – [25]
Medium (12–30 Hz) LED Light stimuli SSVEP Red [52]
Fluorescent – [28]
light
CRT Single graphic stimuli SSVEP White/black [53]
– [54] 7.5
LCD Pattern reversal stimuli SSVEP White/black [55]
High (30–60 Hz) LED Light stimuli SSVEP – [56]
Xe-light SSVEP – [57]
(continued)
197
Table 1 (continued)
198

Frequency band Device Visual stimuli type Response Color References Bit rate
(bits/min)
Low + Medium LED Flicker stimuli SSVEP White [15] –
(6–30 Hz) Red [58] 27–31.5
Green [59] 51.47
[27] 68
CRT/LCD Single graphic stimuli SSVEP White/black [60] 21–58
LCD Flickering stimuli C-VEP – [61]
TFT Hybrid SSVEP + P300 – [62]
LCD Pattern reversal stimuli SSVEP Red, green, yellow [63] –
Pattern reversal stimuli SSVEP White/black [64]
Pattern reversal stimuli mVEPs (motion- [65] 42.1
onset VEPs)
Medium + High LED Light stimuli SSVEP White [66]
(12–60 Hz) CRT Single graphic stimuli – [37]
Low + High CRT Single graphic stimuli SSVEP [67]
Low + Medium + High LED Light stimuli SSVEP Red [68]
(6–60 Hz)
D. Kapgate and D. Kalbande
A Review on Visual Brain Computer Interface 199

organization [32]. The main goal of hybrid BCI is to improve accuracy, ITR, and
error minimization by combining advantages of two conventional BCIs.
Here in this review we are discussing hybrid visual BCI where one or more brain
signals involved in hybrid BCI is evoked through external repetitive visual stimuli
(visual signal) i.e. out of all types of brain signals used in hybrid BCI at least one of
the brain signal is used in visual BCI systems. In this review we are further dividing
hybrid V-BCI in Partial Hybrid V-BCI and Complete Hybrid V-BCI.
(A) Partial Hybrid V-BCI are where brain signals to be processed consisting of
at least one of signal as visual signal and at least one is non-visual signal e.g.
SSVEP + ERD based BCI, P300 + ERD based BCI and so on.
(1) SSVEP + ERD based hybrid V-BCI—Both sequential and simultaneous
system organizations were possible in SSVEP/ERD based BCI. Pfurtscheller et al.
[33] used SSVEP + ERD hybrid V-BCI in sequential combination for orthosis
control application. Result shown that false positive rate reduced by more than
50 % in hybrid BCI as compared to independent conventional BCI.
(2) P300 + ERD based Hybrid V-BCI—Both system organizations as sequential
and simultaneous possible. In such BCI P300 potentials are suitable for discrete
control commands and motion imaginary based signals (ERD) are suitable for
continuous control commands. P300 + ERD based BCI found suitable for several
applications as wheelchair control, robotic control decision applications [34], vir-
tual environment and so on [32].
Another types of partial hybrid V-BCI proposed are NIRS + SSVEP based V-
BVI as brain switch to ON/OFF SSVEP BCI [35] for orthosis control applications
another type is ECG (heart rate) + SSVEP based sequential hybrid V-BVI where
changes of heart rate measured in RRT were used to ON/OFF SSVEP operated
prosthetic hand (Table 2).
(B) Complete Hybrid V-BCI are BCI where all of brain signals involved in
hybrid V-BVI are visual signals. Studies on SSVEP + P300 hybrid V-BCI revealed
that SSVEP signals are suitable for continuous control commands and P300 signals
suitable for discrete control commands. Yin et al. [23] showed simultaneous pro-
cessing in SSVEP + P300 hybrid V-BCI gives higher ITR and accuracy with
minimum user training. Table 3 gives brief overview of complete hybrid V-BCI
systems with their performance rates.

5 Challenges and Solutions

(A) System Communication Rate (ITR) improvement—Regarding to V-BCI the

system communication rate (ITR) depends broadly on two aspects as enhancement
of signal to noise ratio and separability of multiple classes.
(B) Enhancement in SNR—the ratio between related signal and nose is improved
by decreasing other nuisance level and increasing stimulation relates signal level.
Spatial filtering based signal processing methods are good in improving SNR
because they combines signals from different channels as signals recorded from
 200

Table 2 Partial hybrid visual BCI

Hybrid type System Device Frequency band (Hz) Visual stimuli type References Bit rate Accuracy
organization (bits/min) (%)
SSVEP + ERD Sequential LED 8 and 13 (Low + Medium) Light stimuli [33] – 80
Sequential LED 15, 17 and 19 (Medium) Light stimuli [69] – 98.1
Simultaneous LED 8 and 13 (Low + Medium) Light stimuli [70] 6.3 95.6
Simultaneous LED 8 and 13 (Low + Medium) Light stimuli [71] – 81.0
P300 + ERD Sequential LCD 4.5 Single graphic stimuli [72] – >80
Sequential LCD 7.6 Single graphic stimuli [73] 22.7 90.1
Simultaneous LCD/ 6.6 Single graphic stimuli [34] – 82
CRT
ECG (Heart Sequential LED 6.3 and 17.3 Light stimuli [74] – 80
rate) + SSVEP
D. Kapgate and D. Kalbande
Table 3 Complete hybrid visual BCI
Hybrid type System Device Frequency band (Hz) Visual stimuli type References Bit rate Accuracy
organization (bits/min) (%)
P300 + SSVEP Sequential LCD/CRT 18 Single graphic stimuli [75] 19.05 88.15
Simultaneous LED 8.18, 8.97, 9.98, 11.23, Light stimuli [23] 56.44 93.85
A Review on Visual Brain Computer Interface

12.85 and 14.99

N170 + P300 Sequential LCD/CRT – Facial images stimuli [26] 38.7 88.7
P300 + M-VEP Sequential LCD/CRT 4.9 Light stimuli [76] 26.7 96
(Motion onset VEP)
P300 + ErrP (error Simultaneous LCD/CRT 4.0 Single graphic stimuli [77] – 90
potential)
201
202 D. Kapgate and D. Kalbande

multiple channel are less affected by other nuisance signals. In [36] MCC pre-
processing methods is used to maximize ration between visual signals and back-
ground signals with little higher computation time, other also proposed as CCA
[37], ICA [38], PCA [39] etc. Also user efforts to maintain attention on stimuli
affect visual signal strength [40]. User attention distraction from stimuli can dete-
riorate SNR. One possible solution to this is make visual stimuli move along with
controlled elements.
(C) Focused research towards Independent BCI Mostly V-BCI are dependent in
nature, as they require gaze movement in stimuli direction which may not suitable
for people with severe disability in which user can not reliably control gaze [41].
Solution to this is to develop visual stimuli without users gaze.
(D) Non-linearity and Non stationarity of EEG signals EEG signals are non-
linear in nature which may deteriorate BCI performance (classification accuracy). A
solution is use of non-linear dynamic methods rather than linear methods for EEG
signal characterization. Diverse behavioral and mental states of human mind leads
to non-stationarity in EEG signals. Adaptive classification methods can address this
problem as they automatically update classifier during online session.
(E) Reduction in complexity of target detection to address challenges like
decrease in frequency resolution due to one frequency per target, target detection
time, user habituation, probable solution could be use of different relative phases of
stimuli rather than different frequencies, dual frequency stimulation to solve fre-
quency resolution problem; use of machine learning methods in single trail clas-
sification, adaptive methods and optimization of stimulus coding method to
minimize target detection time [42] and consideration of issues like attention blinks,
repetition blindness, target to target interval to overcome user habituation problem.
(F) Reduction of user fatigue Temporary inability of user to respond optical
stimuli due to use of system from longer period called mental fatigue. A lot of
solutions are proposed towards this as use of displays that do not produce negative
side effects, use of improved software’s and dry electrodes, optimization of physical
properties of stimulus e.g. Image based stimulus [43] but still this challenge need to
solve effectively [12, 42] in context of V-BCI.

6 Conclusion

BCI is attractive for disable people suffering from disorders like amyotrophic lateral
sclerosis, brain stem stroke or spinal cord injury as it enable them to perform many
daily life activities independently which improve their quality of life, making them
more independent, increasing productivity at the same time reducing the cost of
intensive care. This review provides a background for finding new methodologies
and paradigms to improve V-BCIs performance further. The future work in V-BCI
area will be focused on design and development of efficient Hybrid V-BCI also
integrating V-BCI with multimodal interfaces. Further work needed to make current
V-BCI technologies effective for real life applications. Collaborative efforts from
A Review on Visual Brain Computer Interface 203

engineers, researchers, neuroscience experts and users will be needed to achieve

this goal.

References

1. Wolpaw J, Birbaumer N, McFarland DJ (2002) Brain-computer inter-faces for communication

and control. Clin Neurophysiol 113:767–791
2. Nicolas-Alonso LF et al (2012) Brain computer interfaces, a review. Sensors 12:1211–1279.
doi:10.3390/s120201211
3. Pasqualotto E, Federici S, Belardinelli MO (2012) Toward functioning and usable brain-
computer interfaces (BCIs): a literature review. Disabil Rehabil Assist Technol 7:89–103
4. Yongwook C et al (2012) Toward brain-actuated humanoid robots: asynchronous direct
control using an EEG-based BCI. IEEE Trans Rob 28(5):1131–1144
5. Zander TO et al (2009) Detecting affective covert user states with passive brain-computer
interfaces. In: 3rd IEEE international conference on affective computing and intelligent
interaction and workshops, 2009
6. Zander TO, Kothe C (2011) Towards passive brain–computer interfaces: applying brain–
computer interface technology to human–machine systems in general. J Neural Eng 8
(2):025005
7. Molina GG et al (2013) Emotional brain–computer interfaces. Int J Auton Adapt Commun
Syst 6(1):9–25
8. Wang Y, Jung T-P (2011) A collaborative brain–computer interface for improving human
performance. PLoS ONE 6(5):e20422
9. Cheng M, Gao X, Gao S, Xu D (2002) Design and implementation of a brain–computer
interface with high transfer rates. IEEE Trans Biomed Eng 49(10):1181–1186
10. Hill NJ et al (2005) An auditory paradigm for brain–computer interfaces. In: Advances in
neural information processing systems. MIT Press, Cambridge, pp 569–576
11. Farwell LA, Donchin E (1988) Talking off the top of your head-toward a mental prosthesis
utilizing event-related brain potentials. Electroencephalogr Clin Neurophysiol 70(6):510–523
12. Rezaiey RF et al (2012) P300 brain computer interface: current challenges and emerging
trends. J Front Neuroeng 5:14
13. Yijun W et al (2006) A practical VEP-based brain-computer interface. IEEE Trans Neural Syst
Rehabil Eng 14:234–240
14. Xiaorong G et al (2003) A BCI-based environmental controller for the motion-disabled. IEEE
Trans Neural Syst Rehabil Eng 11:137–140
15. Wu ZH, Yao DH (2008) Frequency detection with stability coefficient for steady-state visual
evoked poten-tial (SSVEP)-based BCIs. J Neural Eng 5(1):36–43
16. Middendorf M, McMillan G, Calhoun G, Jones KS (2000) Brain–computer interfaces based
on the steady-state visual-evoked response. IEEE Trans Rehabil Eng 8(2):211–213
17. Sutter EE (1984) The visual evoked response as a communication channel. IEEE Trans
Biomed Eng 31(8):583
18. Kimura Y, Tanaka T, Higashi H, Morikawa N (2013) SSVEP-Based brain–computer
interfaces using FSK-modulated visual stimuli. IEEE Trans Biomed Eng 60(10):2831–2838
19. Lee PL, Hsieh JC, Wu CH, Shyu KK, Chen SS, Yeh TC, Wu YT (2006) The brain computer
interface using flash visual evoked potential and independent component analysis. Ann
Biomed Eng 34(10):1641–1654
20. Hong B, Guo F, Liu T, Gao X, Gao S (2009) N200-speller using motion-onset visual response.
Clin Neurophysiol 120(9):1658–1666
204 D. Kapgate and D. Kalbande

21. Yuan P et al (2013) A collaborative brain–computer interface for accelerating human decision
making. In: Stephanidis C, Antona M (eds) UAHCI/HCII, Part I, LNCS 8009. Springer,
Berlin, pp 672–681
22. Xie J et al (2012) Steady-state motion visual evoked potentials produced by oscillating
Newton’s rings: implications for brain-computer interfaces. J PLoS ONE 7(6):e39707
23. Yin E, Zhou Z, Jiang J, Chen F, Liu Y, Hu D (2014) A speedy hybrid BCI spelling approach
combining P300 and SSVEP. IEEE Trans Biomed Eng 61(2):473–483
24. Rivet B et al (2009) Algorithm to enhance evoked potentials: application to brain-computer
interface. IEEE Trans Biomed Eng 56:2035–2043
25. Principe JC (2013) The cortical mouse: a piece of forgotten history in noninvasive brain–
computer interfaces. IEEE Pulse 4(4):26–29
26. Zhang Y, Zhao QB, Jin J, Wang XY, Cichocki A (2012) A novel BCI based on ERP
components sensitive to configural processing of human faces. J Neural Eng 9(2):026018
27. Gao X, Xu D, Cheng M, Gao S (2003) A BCI-based environmental controller for the motion-
disabled. IEEE Trans Neural Syst Rehabil Eng 11(2):137–140
28. Calhoun GL, McMillan GR EEG-based control for human-computer interaction. In:
Proceedings of the 3rd annual symposium on human interaction with complex systems
(HICS’96), pp 4–9. Dayton, Ohio, USA, Aug 1996
29. Beverina F, Palmas G, Silvoni S, Piccione F, Giove S (2003) User adaptive BCIs: SSVEP and
P300 based interfaces. Psychol J 1:331–354
30. Kluge T, Hartmann M (2007) Phase coherent detection of steady-state evoked potentials:
experimental results and application to brain-computer interfaces. In: Proceedings of the 3rd
international IEEE EMBS conference on neural engineering, pp 425–429, May 2007
31. Zhu D et al (2010) A survey of stimulation methods used in SSVEP-based BCIs. J Comput
Intell Neurosci, Article ID 702357
32. Pfurtscheller G et al (2010) The hybrid BCI. Front Neurosci 4:30
33. Pfurtscheller G, Solis-Escalante T, Ortner R, Linortner P, Muller-Putz GR (2010) Self-paced
operation of an SSVEP-based orthosis with and without an imagery-based “brain switch”: a
feasibility study towards a hybrid BCI. IEEE Trans Neural Syst Rehabil Eng 18(4):409–414
34. Riechmann H, Hachmeister N, Ritter H, Finke A (2011) Asynchronous, parallel on-line
classification of P300 and ERD for an efficient hybrid BCI. In: Proceedings of the 5th
international IEEE/EMBS conference on neural engineering (NER’11), pp 412–415, May
2011
35. Coyle SM et al (2007) Brain–computer interface using a simplified functional near-infrared
spectroscopy system. J Neural Eng 4:219–226
36. Garcia-Molina G, Zhu DH, Abtahi S (2010) Phase detection in a visual-evoked-potential
based brain computer interface. In: Proceedings of 18th European signal processing
conference, pp 949–953
37. Lin Z, Zhang C, Wu W, Gao X (2007) Frequency recognition based on canonical correlation
analysis for SSVEP-based BCIs. IEEE Trans Biomed Eng 54(6):1172–1176
38. Kun L et al Single trial independent component analysis for P300 BCI system. In: Proceedings
of the 31th annual international conference of the IEEE engineering in medicine and biology
society (EMBCS’09), pp 4035–4038. Minneapolis, MN, USA, Sept 2009
39. Pouryazdian S, Erfanian A (2009) Detection of steady-state visual evoked potentials for brain-
computer interfaces using PCA and high-order statistics. Proc World Cong Med Phys Biomed
Eng 25:480–483
40. Muller MM, Malinowski P, Gruber T, Hillyard SA (2003) Sustained division of the attentional
spotlight. Nature 424(6946):309–312
41. Allison BZ et al (2008) Towards an independent brain-computer interface using steady state
visual evoked potentials. Clin Neurophysiol 119(2):399–408
42. Gao S et al (2014) Visual and auditory brain–computer interfaces. IEEE Trans Biomed Eng 61
(5):1436–1447
43. Rakotomamonjy A, Guigue V (2008) BCI competition III: dataset II-ensemble of SVMs for
BCI P300 speller. IEEE Trans Biomed Eng 55:1147–1154
A Review on Visual Brain Computer Interface 205

44. Maggi L, Parini S, Piccini L, Panfili G, Andreoni G A four command BCI system based on the
SSVEP protocol. In: Proceedings of the 28th annual international conference of the IEEE
engineering in medicine and biology society (EMBC’06), pp 1264–1267. New York, NY,
USA, Aug 2006
45. Piccini L, Parini S, Maggi L, Andreoni G A wearable home BCI system: preliminary results
with SSVEP protocol. In: Proceedings of the 27th annual international conference of the IEEE
engineering in medicine and biology society (EMBC’05), vol 7, pp 5384–5387. Shanghai,
China, Sept 2005
46. Wang Y, Gao X, Hong B, Jia C, Gao S (2008) Brain-computer interfaces based on visual
evoked potentials: feasibility of practical system designs. IEEE Eng Med Biol Mag 27(5):64–
71
47. Guger C et al (2009) How many people are able to control a P300-based brain-computer
interface (BCI)? Neurosci Lett 462:94–98
48. Townsend G et al (2010) A novel P300-based brain-computer interface stimulus presentation
paradigm : moving beyond rows and columns. Clin Neurophysiol 121:1109–1120
49. Trejo LJ, Rosipal R, Matthews B (2006) Brain-computer interfaces for 1-D and 2-D cursor
control: designs using volitional control of the EEG spectrum or steady-state visual evoked
potentials. IEEE Trans Neural Syst Rehabil Eng 14(2):225–229
50. Lalor EC, Kelly SP, Finucane C et al (2005) Steady-state VEP-based brain-computer interface
control in an immersive 3D gaming environment. EURASIP J Appl Sig Process 2005
(19):3156–3164
51. Lenhardt A, Kaper M, Ritter HJ (2008) An adaptive P300-based online brain–computer
interface. IEEE Trans Neural Syst Rehabil Eng 16(2):121–130
52. Leow RS, Ibrahim F, Moghavvemi M Development of a steady state visual evoked potential
(SSVEP)-based brain computer interface (BCI) system. In: Proceedings of the international
conference on intelligent and advanced systems (ICIAS’07), pp 321–324. Kuala Lumpur,
Malaysia, Nov 2007
53. Cecotti H, Graeser A Convolutional neural network with embedded fourier transform for EEG
classification. In: Proceedings of the 19th international conference on pattern recognition
(ICPR’08), pp. 1–4. Tampa, Fla, USA, Dec 2008
54. Kelly SP, Lalor E, Reilly RB, Foxe JJ Independent brain computer interface control using
visual spatial attention-dependent modulations of parieto-occipital alpha. In: Proceedings of
the 2nd international IEEE EMBS conference on neural engineering, pp 667–670. Arlington,
Va, USA, March 2005
55. Kelly SP, Lalor E, Finucane C, Reilly RB (2004) A comparison of covert and overt attention
as a control option in a steady-state visual eyoked potential-based brain computer interface. In:
Proceedings of the 26th annual international conference of the IEEE engineering in medicine
and biology society (EMBC’04), vol 2, pp 4725–4728. San Francisco, Calif, USA, Sept 2004
56. Garcia Molina G (2008) High frequency SSVEPs for BCI applications. In: Brain-computer
interfaces for HCI and games
57. Huang M, Wu P, Liu Y, Bi L, Chen H (2008) Application and contrast in brain-computer
interface between Hilbert-Huang transform and wavelet transform. In: Proceedings of the 9th
international conference for young computer scientists (ICYCS’08), pp 1706–1710, Nov 2008
58. Muller GR et al (2008) Control of an electrical prosthesis with an SSVEP-based BCI. IEEE
Trans Biomed Eng 55(1):361–364
59. Parini S, Maggi L, Turconi AC, Andreoni G (2009) A robust and self-paced BCI system based
on a four class SSVEP paradigm: algorithms and protocols for a high- transfer-rate direct brain
communication. Comput Intell Neurosci 2009:11 pp, Article ID 864564
60. Zhang Y, Xu P, Liu T, Hu J, Zhang R, Yao D (2012) Multiple frequencies sequential coding
for SSVEP-based brain–computer interface. PLoS One 7(3):e29519
61. Vasquez PM, Bakardjian H, Vallverdu M, Cichocki A (2008) Fast multi-command SSVEP
brain machine interface without training. In: Proceedings of the 18th international conference
on artificial neural networks (ICANN’08), pp 300–307, Sept 2008
206 D. Kapgate and D. Kalbande

62. Xu M, Qi H, Wan B, Yin T, Liu Z, Ming D (2013) A hybrid BCI speller paradigm combining
P300 potential and the SSVEP blocking feature. J Neural Eng 10:026001
63. Cheng M, Gao X, Gao S, Xu D Multiple color stimulus induced steady state visual evoked
potentials. In: Proceedings of the 23rd annual international conference of the IEEE engineering
in medicine and biology society (EMBC’01), vol 2, pp 1012–1014. Istanbul, Turkey, Oct 2001
64. Martinez P, Bakardjian H, Cichocki A (2008) Multi command real-time brain machine
interface using SSVEP: feasibility study for occipital and forehead sensor locations. In:
Advances in cognitive neurodynamics, pp 783–786
65. Liu T, Goldberg L, Gao S, Hong B (2010) An online brain–computer interface using non-
flashing visual evoked potentials. J Neural Eng 7:036003
66. Wang Y, Wang R, Gao X, Gao S (2005) Brain-computer interface based on the high-
frequency steady-state visual evoked potential. In: Proceedings of the 1st international
conference on neural interface and control, pp 37–39, May 2005
67. Sami S, Nielsen KD (2004) Communication speed enhancement for visual based brain
computer interfaces. In: Proceedings of the 9th annual conference of the international FES
Society
68. Ruen SL, Ibrahim F, Moghavvemi M (2007) Assessment of steady-state visual evoked
potential for brain computer communication. In: Proceedings of the 3rd Kuala Lumpur
international conference on biomedical engineering, pp 352–354
69. Savic A, Kisic U, Popovic M (2012) Toward a hybrid BCI for grasp rehabilitation. In:
Proceedings of the 5th European conference of the international federation for medical and
biological engineering, pp 806–809
70. Brunner C, Allison BZ, Altstätter C, Neuper C (2011) A comparison of three brain-computer
interfaces based on event related de-synchronization, steady state visual evoked potentials, or a
hybrid approach using both signals. J Neural Eng 8(2), Article ID 025010
71. Allison BZ, Brunner C, Kaiser V, Müller-Putz GR, Neuper C, Pfurtscheller G (2010) Toward
a hybrid brain-computer interface based on imagined movement and visual attention. J Neural
Eng 7(2), Article ID 026007
72. Yuanqing L et al (2010) An EEG-based BCI system for 2-D cursor control by combining Mu/
Beta rhythm and P300 potential. IEEE Trans Biomed Eng 57(10):2495–2505
73. Yu T (2013) A hybrid brain-computer interface-based mail client. J Comput Math Methods
Med 2013, Article ID 750934
74. Scherer R, Müller-Putz GR, Pfurtscheller G (2007) Self initiation of EEG-based brain-
computer communication using the heart rate response. J Neural Eng 4(4):L23–L29
75. Panicker RC, Puthusserypady S, Sun Y (2011) An asynchronous P300 BCI with SSVEP-
based control state detection. IEEE Trans Biomed Eng 58(6):1781–1788
76. Jin J et al (2012) A combined brain computer interface based on P300 potentials and motion
onset visual evoked potentials. J Neurosci Methods 205:265–276
77. Dal Seno B et al (2010) Online detection of P300 and error potentials in a BCI speller. Comput
Intell Neurosci, Article ID 307254
Design of Lead-Lag Based Internal Model
Controller for Binary Distillation Column

Rakesh Kumar Mishra and Tarun Kumar Dan

Abstract Lead-Lag based Internal Model Control method is proposed based on

Internal Model Control (IMC) strategy. In this paper, we have designed the Lead-
Lag based Internal Model Control for binary distillation column for SISO process
(considering only bottom product). The transfer function has been taken from
Wood and Berry model. We have find the composition control and disturbance
rejection using Lead-Lag based IMC and comparing with the response of simple
Internal Model Controller.

Keywords SISO

Lead-lag
Internal model control
Wood and berry

Distillation column

1 Introduction

For purification of final products distillation process is used in chemical and

petroleum industries. Distillation columns are used to enhance mass transfer or for
transferring heat energy. Binary distillation column consists of a vertical column,
where trays are used to increase the efficiency of separation [1]. Condenser and
reboiler is the main part of distillation column. Here condenser is use to condense
the vapour and a reboiler is used to provide heat for the necessary vaporization.
A reflux drum is used to hold the condensed vapour to recycle the liquid reflux to
back from top of the column [2]. Figure 1 has shown the distillation column.
Internal Model Control and Model Predictive Control have successful controller
design technique for industrial system. For SISO system, IMC design procedure is

R.K. Mishra (&)
T.K. Dan

Department of Electronics and Communication Engineering,
National Institute of Technology, Rourkela, India
e-mail: [email protected]
T.K. Dan
e-mail: [email protected]

© Springer India 2015 207

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_19
208 R.K. Mishra and T.K. Dan

Fig. 1 Distillation column

similar to open loop controller design and there are number of advantages such as
easier to tune, good set point tracking, compensation for disturbance and model
uncertainty [3, 4].

2 Internal Model Control Algorithm

The Internal Model control algorithm is given below the Fig. 2 indicate the block
diagram of internal model control [5, 6].
Where Q(s) is the primary controller (IMC) transfer function, Gp(s) is the pro-
cess transfer function, Gm(s) is the process model transfer function, r(s) is set point,
e(s) is error, c(s) is manipulated variable, d(s) is disturbance, ym(s) is model output
and y(s) is controlled variable (process output).
Internal Model Control perform faster in dynamics stage compare to statics one.
Hence

1
QðsÞ ¼ ð1Þ
GP ðsÞ

But Eq. (1) valid for stable system without delay, so we can design IMC for time
delay system using lag-lead network.
The controller design procedure has been generalized to the following step [3].
1. First we have identified the process model into good stuff and bad stuff by using
all pass formulation or using simple factorization.
Design of Lead-Lag Based Internal Model Controller for Binary Distillation Column 209

Fig. 2 Internal model control d(s)

structure
r(s) + e(s) c(s) y(s)
Q(s) Gp(s)
−

d’(s)
ym(s) +
Gm(s)
−

2. Invert the (good stuff) invertible portion of the process model and to make
proper add the filter.
3. After adding the filter we add lag-lead network which is a form of αs + 1/βs + 1.

f ðsÞ
Q ðsÞ ¼ TðsÞ ð2Þ
Gm ðsÞ

Different strategy in which Internal Model Control work which we given below.

When Model Perfect, No Disturbances [Gp(s) = Gm(s) and d(s) = 0]:

yðsÞ ¼ Gp ðsÞ QðsÞ rðsÞ ð3Þ

When model is Perfect and Disturbance Effect the process:

yðsÞ ¼ ½Gp ðsÞ QðsÞ rðsÞ

ð4Þ
þ f1  Gm ðsÞQðsÞgdðsÞ

With only Disturbance and Disturbance Rejection:

yðsÞ ¼ ½1  Gm ðsÞ QðsÞdðsÞ

ð5Þ
dðsÞ ¼ Gd ðsÞ LðsÞ

L(s) is load disturbance.

For IMC design we have taken filter

1
f ðsÞ ¼ ð6Þ
ð1 þ ksÞn

For disturbance rejection we have use the Lag-Lead network which is in form of
210 R.K. Mishra and T.K. Dan

ðas þ 1Þ
TðsÞ ¼ ð7Þ
ðbs þ 1Þ

where α and β are time constant and used as a tuning parameter for lead-lag based
Internal Model Controller.

3 Distillation Column

Distillation column has two output top XD and bottom XB composition and four
inputs such as two disturbance feed flow (F) and feed composition (XF) and two
manipulated variable reflux (L) and bottom product flow (B/S). The process model
is described by first order transfer function with dead time (8) having gain constant
and time constant for process channel [7, 8] (Fig. 3).

Km ess
G¼ ð8Þ
Ts þ 1

Km represents the process gain, T is the time constant and τ is the dead time. The
process model is a nonlinear and represented by using several parameters given
below in Table 1 [9, 10].

Fig. 3 Distillation process

block diagram

Table 1 Model parameters

Channel (bottom product) Km (process gain) T (time in min.) τ (dead time)
S−xB −19.4 14.4 3
XF−xB 4.9 13.2 3.4
Design of Lead-Lag Based Internal Model Controller for Binary Distillation Column 211

4 Internal Model Control for Distillation Column

We have taking wood and berry 2 × 2 process for distillation column and con-
sidering only bottom product (XB). We have found several response using Lead-
Lag based IMC controller. The several responses for different IMC strategies are
given below.

4.1 When Process is Perfect and No Distrubance

When process is Perfect and no disturbance, it mines process model is equal to

actual process and there is no mismatch. Hence manipulated and control variable
response given below.
Figure 4 shows the rate of stem flow (s), which indicates that by using lead-lag
based internal model controller response, is accurate, good set point tracking and
less settling time as compare to generalize internal model controller response. The
lead-lag based internal model controller response approach to critical damped but
generalizes internal model controller approach to over damp.
Figure 5 shows that controlled variable response XB (bottom product) of dis-
tillation column. It indicated that with lead-lag based internal model controller
output response is more accurate good set point tracking and less settling time as
compare to generalize internal model controller.

Fig. 4 Manipulated variable manipulated variable(S)response with step input

2
response
0

-2

-4
steam flow (S)

s1 for lamda=1 with lag lead

s1 for lamda=1 with lag lead
-6 s1 for lamda=1 without lag lead

-8

-10

-12

-14
0 1 2 3 4 5 6 7 8 9 10
time*(60) (sec)
212 R.K. Mishra and T.K. Dan

Fig. 5 Controlled variable outputResponse (XB)to input unit step

1
response
0.9

0.8
XB1 for lamda=1 with lag lead
0.7 XB2 for lamda=1 with lag lead
XB3 for lamda=1without lag lead

output variable XB
0.6

0.5

0.4

0.3

0.2

0.1

0
0 2 4 6 8 10 12
time sec*60 (sec)

4.2 When Process is Perfect and with Distrubance

When process is perfect with disturbance there is no mismatch occurs in an output

response and using α, β and λ tuning parameter we remove disturbance easily.
Figure 6 shows the difference between the responses of lead lag based internal model
controller and generalize internal model controller. The graph shows that using lead
lag based IMC for distillation column is less affected by the disturbance, here I have
taken k ¼ 1 min, a ¼ 0:9 min, b ¼ 0:01 min and a ¼ 0:4 min, b ¼ 0:1 min as a
tuning parameter.
Figure 6 shows the difference between the responses of lead lag based internal
model controller and generalize internal model controller.

4.3 When Process Only with Distrubance

Figure 7 shows the controlled variable response considering with only disturbance.
We find controlled variable response using lead-lag based Internal Model Controller
and with the help of this graph disturbance totally reject up to 60 min but using
generalize Internal Model Controller response shows the disturbance reject up to
80–90 min. Hence lead-lag based Internal Model Control gave accurate and dis-
turbance free response compare to generalize Internal Model Controller, here I have
taken k ¼ 1 min, a ¼ 0:9 min, b ¼ 0:01 min and a ¼ 0:4 min, b ¼ 0:1 min as a
tuning parameter.
Design of Lead-Lag Based Internal Model Controller for Binary Distillation Column 213

Response output(XB) with distrabance

2.5
XB1 for lamda=1 with lag leadwith diffrent alpha & beta
XB2 for lamda=1 with lag lead alpha & beta
XB3 for lamda=1without lag lead
2
Output Response XB

1.5

0.5

0
0 10 20 30 40 50 60 70 80
Time*60 (sec)

Fig. 6 Controlled variable response when model is perfect and with disturbance

Response output(XB) with only distrabance

0.7
XB1 for lamda=1 with lag lead with diffrent alpha & beta
XB2 for lamda=1 with lag lead alpha & beta
0.6 XB3 for lamda=1without lag lead

0.5
Output Response XB

0.4

0.3

0.2

0.1

0
0 10 20 30 40 50 60 70 80 90 100
Time*60 (sec)

Fig. 7 Controlled variable response considering only disturbance

214 R.K. Mishra and T.K. Dan

5 Conclusion

This paper shows a solution for composition control and disturbance rejection for a
binary distillation column. We have used Lead-Lag based Internal Model Controller
for composition control and disturbance rejection and comparing the response with
generalize Internal Model Controller response. In this paper we have taking single
input single output (SISO) process by considering only bottom product (XB). The
lead-lag based internal model control there are three tuning parameter such as α, β
and λ which is used for composition control and disturbance rejection. Further this
form is used for MIMO system (2 × 2, 3 × 3, 4 × 4) for good set point tracking and
early disturbance rejection.

References

1. Mikles J, Fikar M (2000) Process modeling, identification and control, vol I. STU Press,
Bratislava, p 22
2. Minh VT, Mansor W, Muhamed W (2010) Model predictive control of a condensate
distillation column. IJSC 1:4–12
3. Wayne Bequette B (2003) Process control modeling design and simulation. PHI Publication,
New Delhi
4. Muhammad D, Ahmad Z, Aziz N (2010) Implementation of internal model control (imc) in
continuous distillation column. In: Proceedings of the 5th international symposium on design,
operation and control of chemical processes, PSE ASIA 2010 Organizers
5. Cirtoaje V, Francu S, Gutu A (2002) Valente noi ale reglarii cu model intern. Buletinul
Universit tii Petrol-Gaze Ploiesti (Seria Tehnica) LIV(2):1–6. ISSN 1221-9371
6. Morari M, Zafiriou E (1989) Robust process control. Prentice Hall, Englewood Cliffs
7. Seborge DE, Edgar TF, Mellichamp DA (2004) Process dynamics and control. Wiley,
Singapore
8. Luyben WL (1987) Derivation of Transfer Functions for Highly NonlinearDistillation
Columns. Ind Eng Chem Res 26:2490–2495
9. Alina-Simona B, Nicolae P (2011) Using an internal model control method for distillation
column. In: Proceedings of the 2011 IEEE international conference on mechatornics and
automation, August 2011
10. Băieşu Alina-Simona (2011) Modeling a nonlinear binary distillation column. CEAI 13(1):49–53
Clinical Approach Towards
Electromyography (EMG) Signal
Capturing Phenomenon Introducing
Instrumental Activity

Bipasha Chakrabarti, Shilpi Pal Bhowmik, Swarup Maity

and Biswarup Neogi

Abstract The research on electromyography (EMG) signals analysis is allied with

clinical/biomedical applications, evolvable hardware chip (EHW) development, and
modern human computer interaction era. EMG signals acquired from muscles
require advanced methods for detection, decomposition, processing, and classifi-
cation. The purpose of this paper is to illustrate the capturing phenomenon of EMG
Signal introducing instrumental activity. It further point up some ideas about the
block diagram of the EMG Signal recording instrument and the procedural
approach towards the EMG recording techniques, provide efficient and effective
ways of understanding the signal and its nature. The clinical real-time activity of
EMG recording for biceps brachii muscle is presented with flow diagram. This
paper provides researchers concrete and valuable information of EMG signal and its
analysis procedures.

Keywords Electromyography Motor unit action potential (MUAP)
Nervous



system Surface EMG Invasive EMG EMG recorder

1 Introduction

Biomedical signal means a collective electrical signal acquired from any part of our
body that represents a physical variable of interest. This signal is normally a
function of time and is describable in terms of its amplitude, frequency and phase.

B. Chakrabarti (&)
Gargi Memorial Institute of Technology, Kolkata, India
e-mail: [email protected]
S.P. Bhowmik
S. Maity
EIE Department, JIS College of Engineering, Kalyani, India
B. Neogi
ECE Department, JIS College of Engineering, Kalyani, India
e-mail: [email protected]

© Springer India 2015 215

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_20
216 B. Chakrabarti et al.

The EMG signal is a biomedical signal that measures electrical currents generated
in muscles during its contraction representing neuromuscular activities. The ner-
vous system always controls the muscle activity (contraction/relaxation). Hence, the
EMG signal is a complicated signal, which is controlled by the nervous system and
is dependent on the anatomical and physiological properties of muscles. EMG
signal acquires noise while traveling through different tissues. Moreover, the EMG
detector, particularly if it is at the surface of the skin, collects signals from different
motor units at a time which may generate interaction of different signals. Detection
of EMG signals with powerful and advance methodologies is becoming a very
important requirement in biomedical engineering. The main reason for the interest
in EMG signal analysis is in clinical diagnosis and biomedical applications. Human
motor system manipulates posture, strength and gestures in humans, includes
central motor system and a large number of motor units (MU) [1]. Every single MU
incorporates a motor neuron in spinal cord, multiple branches of its axon and
muscle fibers for innervation. Human movements are possible for the relationship
between central nervous system and skeletal muscle.

2 Short Informatics Study on EMG Sensor

Inherently, two types of EMG capturing techniques are broadly used: surface EMG
(SEMG) and intramuscular EMG (Invasive EMG). A needle should be inserted into
muscle tissue at the time of experiment of intramuscular EMG. Though intra-
muscular EMG is very precise in sensing muscle activity, is normally considered to
be impractical for human computer interaction applications. Invasive EMG allows
effective information about the state of muscle and its nerve [2]. At the time of rest,
normal muscle acquires some normal electrical signal when needle is inserted into
muscles but abnormal impulsive activity indicates some muscle damage. Primarily
surface EMG is noisier than invasive EMG as motor unit action potentials (MUAP)
must proceed through body tissues such as fat and skin before a sensor on the
surface can capture them.
The raw EMG Signal consists of a series of spikes whose amplitude depends on
the amount of force delivered by the muscle, the stronger the contraction of the
muscle, the larger the amplitude of the EMG Signal. The frequencies of the spikes
are the firing rates of the motor neurons. Since the amplitude of the EMG Signal is
directly related to the force exerted by the muscle, it is used to determine the force
signal sent to the exoskeleton. Fundamentally, two types of electrodes are utilized:
Dry electrodes and Gelled electrodes. Dry electrodes are directly communicated
with skin and are used where geometry and size of electrode does not allow gel. An
electrolytic gel is required as chemical interface between skin and metallic part of
electrodes in gelled electrodes. This electrode may be disposable or reusable.
Disposable electrodes are very light in nature and with proper application it
Clinical Approach Towards Electromyography (EMG) Signal … 217

Fig. 1 a EMG NCV disc electrode, b EMG sensory ring electrode, c EMG GND disc electrode,
d needle electrode

minimizes the instability of electrode displacement even during rapid movement

[3]. Basically EMG sensors are essential to capture various EMG signals from
different body muscles for different movements. EMG sensor can be worn around
the arm or as wrist-watch. The different types of electrode are shown. Figure 1a
shows that EMG NCV disc electrode, which is basically an electrode used for
surface EMG signal measurement. Figure 1b shows EMG sensory ring electrode.
Figure 1c gives the idea about EMG GND disc electrode. Figure 1d shows idea
about needle electrode, inserted into the muscle to capture raw EMG signal.

3 Block Diagram of EMG Recorder Instrument

The schematic diagram represents the overall EMG recorder instrument and anyone
can assure about their muscle condition by this recorder. Power supply feeds power
to spike protector (surge protector). Surge protector is being used to protect elec-
trical devices from voltage spikes, either obstructing or shorting ground of excess
voltages beyond safe threshold. Spike protector links up two devices, personal
computer besides acquisition box. Acquition of data is major goal of acquisition
box. Data acquisition system adapts analog waveform (real world physical condi-
tions) into digital numeric values.
Data acquisition incorporates sensors, signal conditioning circuitry and ADC.
Physical phenomenon is being altered to equivalent electrical signals by sensor
transducer. Amplification, filtering, attenuation isolation, linearization are individ-
ual parts of signal conditioning circuit. Acquisition box colligates four parts of
EMG recording machine. Human muscle electrical signal is being fetched by two
electrodes, plug into two channel amplifier arm, attached to CPU. Stimulus and foot
switch are being employed to gather electrical signals. Headphones are being used
to Bera operation. Googles is being applied for blink operation (Fig. 2).
218 B. Chakrabarti et al.

Fig. 2 Schematic block

diagram of EMG recording
instrument

4 Procedural Approach Towards EMG Recording

Technique

The flow chart describes about the entire procedure of recording EMG data. The
convenient muscle is being properly selected for descent EMG data. Reference as
well as active muscles is being decently distinguished. Now, the threshold value, rise
time, delay time, sweep, amplifier value are being reform in EMG window to get
appropriate result. Needle/Electrode is being injected to proposed muscle (Fig. 3).
Now if the signal is not precise, each step is being scrutinized, such as muscle is
being delegated accurately and ground, active muscle, reference muscle should be
differentiated properly. If the signal is precise, true signal is being gathered through
amplifier arm and then feed to personal computer and is being saved as patient
information for future work.

5 Clinical Activity on EMG Captured and Reporting

The capability of muscle to produce tension is termed as muscle strength [5].

Muscles are innervated by nerves. Nerve simulation is the requisite condition to
produce muscular force that generates mechanical work. Many motor-units (MU)
incorporate a muscle and they consist of respective muscle fibers that are excited by
branches of muscle fibers of a single motor nerve cell. Resting potential is similar
across of both membrane and nerve fiber [6]. Depending on degree of constraint
and posture desired by muscle, one MU may contain 3–2,000 muscle fibers. A low
strength muscle have few muscle fibers in one MU, but the larger muscle having
high strength that controls larger movement may bear 100–1,000 fibers per MU [7].
At the time of contraction, an action potential establishes muscle action that pro-
ceeds to axon and is transmitting across the motor end-plates.
Clinical Approach Towards Electromyography (EMG) Signal … 219

Fig. 3 Basic flow chart of EMG recording technique

220 B. Chakrabarti et al.

Fig. 4 Block diagram

representation of motor
control mechanism and
different factors associated to
generate muscle force

All the muscle fibers in one MU are not contracted at same time, only those
fibers receive impulse that are needed for the implementation of specific function.
So, at a particular time some fibers take rest and others are inspired to contract and
all programs are monitored by central nervous system. As CNS handles all pro-
grams that are held by different fibers, one can implement a thought control arm that
is controlled by artificial CNS.
The Fig. 4 explains the elbow flexion and in this experiment only biceps brachii
is the active muscle. The term ‘muscle’ denotes entire muscle-tendon unit and it
signifies a group that covers from a distal muscle-tendon junction to a proximal one.
Muscle’s basic property establishes the actual capacity of muscle to produce force
or shorten at given velocity. This produced force also depends on muscle archi-
tecture and joint conception.
Biceps brachii muscle is a two headed muscle, lies on upper arm in the middle of
shoulder and elbow. Both heads of biceps are joined to form a single muscle belly
and is attached to upper fore arm. When it crosses shoulder and elbow, its principle
function is to flex the fore arm at the elbow and supinate the forearm [8]. It is tri-
effective, means it works across three junctions and its main purpose is to supinate
the forearm and flex the elbow (Fig. 5).
Clinical Approach Towards Electromyography (EMG) Signal … 221

Fig. 5 Diagram of elbow flexion, extension with null value

Fig. 6 EMG data collection procedure with EMG signal collecting instrument

The procedure of EMG data collection is shown in Fig. 6 and another picture is
total EMG recording machine. Then collected EMG data is shown in Fig. 7 where
two condition has indicated. One is action condition and another is resting condi-
tion. In action condition, EMG signal amplitude is increased. The amplitude of
surface EMG signal is affected by several things. Such as thickness, conductivity
and subcutaneous layer of skin, muscle fibre diameter, interval between action
muscle fiber, innervations zone, tendons of active motor unit site and electrodes
filtering properties. This can be calculated in terms of average rectified value (ARV)
and root mean square value (RMS) [9].
(ARV) is defined as a time windowed mean of the absolute value of the signal.
ARV is one of the various processing methods used to construct derived signals
from raw EMG data that can be useful for further analysis.
222 B. Chakrabarti et al.

Fig. 7 Collected EMG data when biceps brachii is the active muscle

ARV can be calculated

1 X Ne
ARV ¼ jxn j ð1Þ
Ne n¼1

Here, xn is EMG signal value at time index n and Ne is the epoch length.
Root Mean Square EMG (RMS EMG) is defined as the time windowed RMS
value of the raw EMG. RMS is one of a number of methods used to produce
waveforms that are more easily analyzable than the noisy raw EMG.
RMS can be calculated
rffiffiffiffiffi Ne
1 X
RMS ¼ jxn j ð2Þ
Ne n¼1

6 Conclusion

EMG signal carries valuable information regarding the nervous system. So the aim
of this paper is to present the detailed information about the EMG Signal recording
techniques with the procedural approach of the instrument RMS EMG EP MARK-II
Kit. Without acquiring the EMG signal it unable analyze. In this concern the detailed
information about the instrumental activity is very necessary. This study clearly
Clinical Approach Towards Electromyography (EMG) Signal … 223

points up the method of EMG signal analysis techniques so that right methods can be
applied during any clinical diagnosis, biomedical research, hardware implementa-
tions and end user applications.

Acknowledgements We would like to express our gratitude and appreciation to All India Council
of Technical Education (AICTE) The Govt. of India, for encouraging financially through Research
Promotion Scheme. Additionally, we acknowledge JIS College of Engineering, Kalyani and Gargi
Memorial Institute of Technology, Kolkata for the support towards this research paper.

References

1. Rissanen S (2012) Feature extraction methods for surface electromyography and kinematic
measurements in quantifying motor symptoms of Parkinson’s Disease. Eastern Finland,
Kuopio, 24 Feb 2012
2. Akumalla SC (2012) Evaluating appropriateness of emg and flex sensors for classifying hand
gestures. University of North Texas, Oct 2012
3. Day SJ (1997) The Properties of Electromyogram and Force in Experimental and Computer
Simulations of Isometric Muscle Contractions. Data from an Acute Cat Preparation.
Dissertation, University of Calgary, Calgary
4. Finni T (2001) Muscle mechanics during human movement revealed by in vivo measurements
of tendon force and muscle length. Neuromuscular Research Center, Department of Biology of
Physical Activity, University of Jyväskylä, Jyväskylä
5. Acierno SP, Baratta RV, Solomonow M (1995). A practical guide to electromyography for
biomechanists. Bioengineering Laboratory/LSUMC Department of Orthopaedies, Louisiana
6. Englehart K, Hudgin B, Parker PA (2001) A wavelet-based continuous classification scheme
for multifunction myoelectric control. IEEE Trans Biomed Eng 48:302–311
7. Kuriki HU, de Azevedo FM, Ota Takahashi LS The relationship between electromyography
and muscle force, University de São Paulo, Brazil
8. Lippert Lynn S (2006) Clinical kinesiology and anatomy, 4th edn. F. A. Davis Company,
Philadelphia
9. Rissanen S (2012) Feature extraction methods for surface electromyography and kinematic
measurements in Quantifying motor symptoms of Parkinson’s Disease. Eastern Finland,
Kuopio, 24 Feb 2012
10. Henneberg K, Plonsey R (1993) Boundary element analysis of the directional sensitivity of the
concentric EMG electrode. IEEE Trans Biomed Eng 40:621
11. Light CM, Chappell PH (2000) Development of a lightweight and adaptable multiple-axis
hand prosthesis. Med Eng Phys 22:679–684
12. Vinet R, Lozach N, Beaudry N, Drouin G (1995) Design methodology for a multifunctional
hand prosthesis. J Rehabil Res Dev 32:316–324
Brain Machine Interface Automation
System: Simulation Approach

Prachi Kewate and Pranali Suryawanshi

Abstract Brain Machine Interface (BMI) till now is generally preferred only for
repairing damaged hearing, sight and movements with the help of neuroprosthetics
application. These applications consist of some external unit which gathers some
information in the form of signals from the brain and processes it so as to transfer
them to the implanted unit. In this way these applications had helped the people to
bring back their ability from various neuromuscular disabilities. Similarly, the BMI
can be very useful for automation system. It will help in reducing accidents which
had contributed to high mortality rate. A brain actuated automation system will also
help motor disabled person to move independently. Signals from brain will be
acquired with the help of dry electrodes and those signals will be processed in the
system processor. The signal after processing will be then applied to the system
depending on the instructions given by the person sitting on it.

Keywords Brain Machine Interface (BMI)
Neural network

1 Introduction

A Brain Machine Interface (BMI) is a correspondence framework that does not rely
on upon the brains typical yield pathways of peripheral nerves and muscle. It is
another approach to impart between a working human brain and the computerization
framework. These are implanted interfaces with the mind, which can possibly
transmit and get receive replies from the brain. This interface changes mental choices
and responses into control motions by investigating the bioelectrical signals.

P. Kewate (&)
P. Suryawanshi

Department of Computer Science & Engineering, G.H. Raisoni Institute
of Engineering and Technology for Women’s, Nagpur, India
e-mail: [email protected]
P. Suryawanshi
e-mail: [email protected]

© Springer India 2015 225

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_21
226 P. Kewate and P. Suryawanshi

It was once recognized as a science fiction to connect the machines straight-

forwardly with the brain, however now-a-days with huge improvement in the
neuroscience it is conceivable to do so. Also still the examination is going on. Brain
is a complex sensory system whose operation we will need to investigate it in point
of interest. By utilizing new innovation and hi-tech technology, incapacitated
individual will have the capacity to paint with the assistance of automated arm, to
fly a plane and to drive vehicle.
Miscellaneous physiological activities of human will generate feeble amount of
electricity; the brain is composed of neurons. Therefore, the brain signal is bio-
logical signal associated with brain, i.e. brain waves. Brain is generally ruled with
its certain laws related to brain’s consciousness. The frequency of brainwave will be
significantly different at different states of joy, uneasiness, fear, sad, trust, etc. BMI
gives new yield pathways to the mind by deciphering estimations of cerebrum
movement into inputs for an outside gadget. These yield pathways normally work
in one of two separate ways: Process Control and Goal selection. Example for
process control is moving cursor left or right and for Goal Selection is moving
cursor directly to the selected goal such as at the end of the page [1].
The fundamental thought of BMI is to decipher the pattern of brain activity into
corresponding brain commands. A normal BMI is made up of signal acquisition,
signal analysis and automation system [2]. The key concept of BMI is to change
over the EEG information of the client into a control command. The very important
part of BMI research is to adjust the mutual adaptation relationship between the
human brain and the BMI system, i.e. to find a suitable signal processing, making
nerve signals convert into the command or an operation signal that can be recog-
nized by the computer in real time, quickly and accurately by the means of BMI
system.
BMI gives immediate pathway between human and physical gadgets which is
possible with the assistance of biological signals. There are various biological
signals such as EEG (electro-encephalograms), MEG (Magneto-encephalograms),
BOLD (blood oxygen level dependent). But due to its low cost and feasible to use
EEG signals are mostly used for BMI systems.
Neural Network has played a vital role in the system. Neural Network takes a
different approach than a regular computer. Conventional computer solve the
problem in the algorithmic approach i.e. the rules or program for the particular task
are already set in the computer. Thus these rules are used while solving the par-
ticular problem. The computer cannot solve the problem unless knows the steps to
execute them. This restricts the problem solving capability of the computer. Thus it
would be very easy if computer would solve the problems which we exactly don’t
know. Whereas, neural network solve the problem in the same way as the brain
works. There are many tasks getting processed in the parallel way. They learn from
examples. They cannot be programmed to do a specific task.
The BMI which is presenting in the paper is an automation system working from
the signals acquired from the brain with the help of dry electrodes. The EEG signals
Brain Machine Interface Automation System: Simulation Approach 227

acquired from the electrodes with the help of signal acquisition and then those
signals are processed in the processor using Neural Network. After the signals are
processed they are then applied to the automation system. Thus according to the
particular command given by the brain the system will work.
In this paper the First part describes the introduction of the paper has been given.
The second part describes the objective of the paper. The third part will give us the
research methodology about the proposed system.
Lastly, the expected outcome about the proposed system concluding with the
conclusion.

2 Objective

The primary objective of the proposed system depends on the three tasks which are
as follows (Fig. 1):
1. Signal Acquisition: Dry electrodes placed on the cap will acquire the EEG
signal. The acquired signals are in the analog format and hence need to be
converted into digital format with the help of A/D convertor.
2. Signal Processing: The digitized signals are then analysed using FFT and only
the required content are then extracted and is sent to the classifier.
3. Automation System: The classified signal is then given to the automation system
which performs the particular command given.

Fig. 1 Schematic view of

BMI showing three main
tasks of signal acquisition,
signal processing and
automation system
228 P. Kewate and P. Suryawanshi

3 Literature Survey

Recently, innovative work of brain-controlled robots have accepted a lot of con-

sideration in view of their capability to bring again to individuals with wrecking
neuromuscular issue and enhance the personal satisfaction and self-reliance of these
users. It incorporates different provisions, for example, a cursor on the screen [1],
selecting letters from virtual console [3], searching web [4], and playing games [5].
Akce et al. [6] proposed an interface for navigating a mobile robot which moves
at a fixed speed in a planar workspace [6] with noisy binary inputs. It uses hybrid
strategy where measurements of brain activity as evidence to reduce the uncertainty
about the entire path all at once.
Citi et al. [7] proposed BMI Mouse using the P300 waves. The system uses two
approaches: first it completely dispenses with the problem of detecting P300s by
logically behaving as analogue device and second approach is the single trial
approach where the mouse performs an action after every trial.
A Portable BMI for expressing inner ideas by voice is designed and fabricated in
Yan et al. [2]. EEG signal is recorded by dry electrode and is processed succes-
sively by the ASIC.
The same idea of BMI as illustrated in these papers above is going to be utilized
as a part of the proposed work. BMI could be utilized to create such a robotization
framework where engine incapacitated individual will have the capacity to move
starting with one spot then onto the next. The individual who have engine handi-
caps and are incapacitated are not ready to move any of his appendages so in such a
condition this proposed framework will be a shelter to them. They need to just think
about certain course where they need to move and along these lines the framework
will take them to that specific heading. A brain controlled wheelchair may serve as
a requisition to them. The subtle element clarification of the proposed framework is
demonstrated further in the Research system area.

4 Research Methodology

Research methodology for the proposed paper is divided according the hardware
used and the simulation base. The following is according the hardware used in the
system and for simulation will be explained later.
The system will consist of three main blocks as shown in the diagram above
Fig. 2. These blocks are again subdivided into sub blocks which are as shown in the
proposed architecture Fig. 3.
1. Input Block: The input block will perform the signal acquisition task. Signals
will be acquired from the electrodes which are mounted on the cap. The cap will
be placed on the scalp of the person sitting on the system. This signals acquired
will be amplified with the help of op-amp. As brain is made up of numerous
neurons, it will collect the large amount of data. This data will filter using notch
Brain Machine Interface Automation System: Simulation Approach 229

Fig. 2 Basic block diagram

Fig. 3 Proposed architecture

filter and the required amount of data will be further sent to the ADC convertor.
Then these digitized signals will be sent to the logical circuit where it will
compare the real time data with the standard files stored in the system processor.
2. System Processor: The signals from the logical circuit are fed to the System
Processor via Communication Protocol. Co-learning of the signals will take
place in this block. After training these signals, signals will be extracted and will
be classified according to mining. Single trial analysis will occur so as to
consider only one signal at a time.
3. Output Block: Communication Protocol transfers the signal from system pro-
cessor to the logical circuit of the output block, which are then fed to the motor
control of the automation system via relay driver.
Now, the Methodology used for the simulation part of the proposed work is as
follows (Fig. 4).
The first module will be of training the acquired signal. The signals which are
acquired from the electrodes are stored in the processor as data set. The file is
browsed and is selected for the training purpose with the help of neural network.
The trained signals are then saved.
The saved file is browsed in the second module to plot them as graph as dis-
played in the module. The first graph is for the error rate and the second graph is for
the success rate (Fig. 5).
230 P. Kewate and P. Suryawanshi

Fig. 4 Training of signals

Fig. 5 Plotting the signals on

graph

5 Performance Evaluation

In this section we will compare the performance achieved by the proposed system
with the existing system. The parameters considered for the comparison are the
error rate and the success rate.
Error rate: the number of times the system not work efficiently.
Success rate: the number of times the system has success after giving a particular
command.
Brain Machine Interface Automation System: Simulation Approach 231

Brain EEG Signal

Existing Approach
2.5

Error Rate 2.0

1.5

1.0

0.5

0.0
0 100 200 300 400 500 600
Sample

Fig. 6 Error rate for existing and our approach

The Error rate for the existing and our approach is plotted in the graph Fig. 6. On
X axis we have taken the symbol length and on the Y axis percentage error rate.
Thus we observe that we are getting low error rate than the error rate observed in
the existing approach from the graph. Thus our approach has proved to be better
than the existing approach.
Similarly, the success rate for the proposed system is compared with the existing
approach. On X axis we have taken the symbol length and percentage of success
rate on the Y-axis. But as can be seen from the graph it hadn’t proved to be that
much efficient. The graph for the success rate for the existing and for our approach
is as shown in the Fig. 7.

Brain EEG Signal

Existing Approach
2.0
Success Rate

1.5

1.0

0.5

0.0
0 100 200 300 400 500 600
Sample

Fig. 7 Success rate for existing and our approach

232 P. Kewate and P. Suryawanshi

6 Conclusion

Thus our proposed system has proved to be efficient in reducing the error rate than
the existing system. In this way our system can be much helpful for the persons
having motor disorders and can improve the quality of life. Still, there is lot of
research to be made in this field. Thus BMI can be implemented on the automation
system.

References

1. Wolpaw JR, McFarland DJ, Neat GW, Forneris CA (2008) An EEG-based brain-computer
interface for cursor control. IEEE Intell Syst 23(3):72–79
2. Yan Y, Mu N, Duan D, Dong L, Tang X, Yan T (2013) A dry electrode based headband voice
brain-computer interface device. In: International conference on complex medical engineering,
May 2013
3. Donchin E, Spencer KM, Wijesinghe R (2000) The mental prosthesis: assessing the speed of
P300-based brain computer interface. IEEE Trans Neural Syst Rehabil Eng 8(2):174–179
4. Karim AA, Hinterberger T, Richter J (2006) Neural Internet: web surfing with brain potentials
for the completely paralyzed. Neurorehabil Neural Repair 20(2):508–515
5. Krepki R, Blankertz B, Curio G, Muller KR (2007) The Berlin brain computer interface
(BBCI): towards a new communication channel for online control in gaming applications.
J Multimedia Tools Appl 33(1):73–90
6. Akce A, Johnson M, Dantsker O, Bretl T (2013) A brain–machine interface to navigate a
mobile robot in a planar workspace: enabling humans to fly simulated aircraft with EEG. IEEE
Trans Neural Syst Rehabil Eng 21(2):306–318
7. Citi L, Poli R, Cinel C, Sepulveda F (2008) P300-based BCI mouse with genetically-optimized
analogue control. IEEE Trans Neural Syst Rehabil Eng 16(1):51–61
 Part III
DSP and Clinical Applications
Cognitive Activity Classification
from EEG Signals with an Interval Type-2
Fuzzy System

Shreyasi Datta, Anwesha Khasnobish, Amit Konar

and D.N. Tibarewala

Abstract The present work attempts to classify Electroencephalogram (EEG)

signals corresponding to three different cognitive activities using an Interval Type 2
Fuzzy System (IT2FS) classifier. This approach is used in order to account for the
fact that EEG signals from the same person for the same stimulus have variations in
different observations in the same as well as over different days of experiments.
Adaptive Autoregressive Parameters, Hjorth Parameters, Hurst Exponents and
Approximate Entropy features are extracted from the acquired EEG signals to
obtain the maximum possible discrimination between different classes of activities.
A maximum classification accuracy of 85.33 % on an average over all subjects and
classes is obtained using IT2FS classifier on a combined feature space. The study is
validated using Friedman’s statistical test with respect to four other classification
algorithms.

Keywords Cognitive activity recognition Electroencephalogram Interval type-2

fuzzy systems

S. Datta (&)
A. Konar

Department of Electronics and Telecommunication Engineering,
Jadavpur University, Kolkata, India
e-mail: [email protected]
A. Konar
e-mail: [email protected]
A. Khasnobish
D.N. Tibarewala
School of Bioscience and Engineering, Jadavpur University, Kolkata, India
e-mail: [email protected]
D.N. Tibarewala
e-mail: [email protected]

© Springer India 2015 235

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_22
236 S. Datta et al.

1 Introduction

Brain Computer Interface (BCI) technology [1] finds applications in rehabilitation,

assistance and treatment of neuro-motor, sensory-motor and cognitive disabilities.
An interesting avenue in BCI research is brain signal based systems with the ability
to recognize the cognitive state of human beings. Such systems are important in the
development of context aware pervasive computing systems [2, 3]. Human brain
activities convey direct information regarding their cognitive state and location.
There have been various research works that study the cognitive state of the brain
through Electroencephalogram (EEG), Magnetoencephalogram (MEG) or func-
tional Magnetic Resonance Imaging (f-MRI) [4–6]. EEG has been popular in BCI
applications because of its superior temporal resolution, easy acquisition, simple
processing, portability, cost-effectiveness and freedom from exposure of the subject
to intense magnetic fields. There are several instances of EEG based BCI research
in literature, that include decoding mental states and cognitive activities [7–9],
prosthetic and robotic control through motor imagination [10], computer cursor
control [11], emotion recognition [12], object shape recognition from visual and
tactile exploration [13] to mention a few.
In [14, 15] various tools and techniques used commonly in EEG classification
related to BCI research have been stated, that include Support Vector Machines,
Linear Discriminant Analysis, Nearest Neighbour Classifiers, Neural Networks,
Bayesian methods, Hidden Markov Models among others. However standard
classifiers often fail to perform successfully in EEG analysis because of the nature
of EEG signals. These signals are non-stationary, non-periodic [16] and have
variations among different observations of the same class.
Fuzzy logic [17] can represent uncertainty associated with any system. Fuzzy
logic based classifiers [18] have been successful in handling variations in signals.
Type 1 Fuzzy Systems (T1FS) can be used for classification by representing the
variations in the signal values at different instances using a single membership
function [17, 19]. EEG responses to the same stimulus taken at different times are
variable for a particular person; these responses also have variations in between
persons. Therefore there exists an inherent difficulty in classification of these signals
to identify the different states of the brain. A T1FS is unable to capture the vari-
ations in the memberships for different trials of the same experiment over a period
of days. General Type 2 Fuzzy Systems (GT2FS) [20, 21] can overcome this
difficulty by assigning a secondary membership function to model the uncertainty
associated with the primary membership function. However, developing the sec-
ondary membership is challenging and the process is also computation intensive.
Interval Type 2 Fuzzy Systems (IT2FS) [22, 23] employ uniform and constant
secondary membership values and provide a simple solution to the problem.
In the present work, EEG signals corresponding to three activities, namely, reading,
watching a video with audio and relaxing have been classified using the concept of
IT2FS. As non-linear features are important in EEG classification problems [24], EEG
signals have been represented through Adaptive Autoregressive Parameters and Hjorth
Cognitive Activity Classification from EEG Signals … 237

Parameters, as well as two non-linear features, namely Hurst Exponents and

Approximate Entropy, and a combination of all the features in this study. The per-
formance of the IT2FS classifier has been compared with T1FS classifier considering
Gaussian membership function, as well as standard pattern classifiers like Support
Vector Machine, Neural Network Classifier and Naïve Bayes Classifier.
The rest of the paper is structured as follows. In Sect. 2 the principles used in this
work have been explained. Section 3 illustrates the experiments and the results.
Finally in Sect. 4 the conclusions are drawn and future scopes of work are stated.

2 Principles and Methods

This section provides the descriptions of the tools and techniques used in the
present work.

2.1 Feature Extraction Techniques

A linear approach, namely, Adaptive Autoregressive Parameters, a quasi linear

approach, namely, Hjorth Parameters and two nonlinear signal features, namely,
Hurst Exponents and Approximate Entropy have been used to represent EEG
signals in this work.

2.1.1 Adaptive Autoregressive Parameters

Autoregressive Parameters and Adaptive Autoregressive Parameters (AAR) are

time-domain features for EEG analysis [25, 26]. The AR approach models a sto-
chastic time series assuming the data to be stationary. For estimating the non-
stationary EEG signals computationally expensive methods like windowing can be
used with the AR technique. Another approach is to use AAR model, where the AR
parameters for representing EEG signals are estimated in a time-varying manner, as
explained by (1) and (2), where the index k is an integer to denote discrete, equi-
distant time points, y(k) is the kth instance of the signal, p is the order of the AAR
model, y(k − i) with i = 1 to p are the p previous sample values, ai,k are the time-
varying AR model parameters, and x(k) is a zero-mean-Gaussian-noise process with
time varying variance r2x ðkÞ:

yðkÞ ¼ a1;k yðk  1Þ þ


þ ap;k yðk  pÞ þ xðkÞ ð1Þ

xðkÞ ¼ Nf0; r2x ðkÞg ð2Þ

There are various algorithms to estimate the parameters of AAR such as, least
mean square (LMS), Kalman filtering, recursive AR or recursive least square (RLS).
238 S. Datta et al.

2.1.2 Hjorth Parameters

Activity, Mobility and Complexity, collectively known as the Hjorth Parameters [27]
are another set of time domain features. For an input signal y(k), Activity A(y),
Mobility M(y) and Complexity C(y) are defined by (3), (4) and (5) respectively, where
var(y) and y0 denote the variance and first derivative of the signal y(k) respectively.

AðyÞ ¼ varðyÞ ð3Þ

sffiffiffiffiffiffiffiffiffiffi
A ð y0 Þ
MðyÞ ¼ ð4Þ
AðyÞ

M ð y0 Þ
CðyÞ ¼ ð5Þ
MðyÞ

2.1.3 Hurst Exponents

Non-linear models are efficient for representations of complicated physiological

processes exhibiting chaotic behavior [28]. EEG responses at different excitations
may have random variations and can be described to be chaotic. Hence, Hurst
Exponent (H) [24, 28], a non-linear parameter is used as an EEG feature. The Hurst
Exponent is obtained using the Rescaled Range Analysis (R/S Analysis) [28] by
statistical methods. It estimates the occurrence of long-range dependence and its
degree in a time series by the evaluating the probability of an event to be followed
by a similar event. It is described by (6), where T denotes the sample duration and
R/S denotes the corresponding rescaled range value.

H ¼ logðR=SÞ= logðTÞ ð6Þ

H = 0.5 indicates that the time-series is similar to an independent random pro-

cess, 0  H\0:5 indicates a present decreasing trend in the process implies a future
increasing trend and vice versa whereas 0:5\H  1 indicates that a present
increasing/decreasing trend in the process implies a future increasing/decreasing trend.

2.1.4 Approximate Entropy

Approximate Entropy (AE) [24, 28], is another non-linear feature that provides a
measure of regularity of a signal. A deterministic signal with high regularity has a
very small AE value while a random signal with a low regularity has a high AE
value. Two parameters, the embedding dimension m and tolerance of comparison
value r are necessary for computation of AE. For a time series of length N, AE is
obtained as (7) and (8), where Cm(r) is the correlation integral as defined in [24].
Cognitive Activity Classification from EEG Signals … 239

AEðm; rÞ ¼ Um ðrÞ  Umþ1 ðrÞ ð7Þ

X
Nðm1Þ
Um ðrÞ ¼ ðN  ðm  1ÞÞ1 ln½Cim ðrÞ ð8Þ
i¼1

AE can be observed as the likelihood in logarithmic terms that certain patterns in

the data will be followed by similar patterns.

2.2 Interval Type 2 Fuzzy System for Classification

In a normal or Type 1 Fuzzy System (T1FS) [18, 19] every variable has a membership
value in the close interval [0, 1] according to a predefined membership function.
Commonly used membership functions are triangular, trapezoidal, Gaussian or sig-
moidal [29]. Given a set of variables, the Gaussian membership function of a variable
x can be easily determined from their mean μ and the standard deviation σ according
to (9), and can be implemented in classification problems using Fuzzy logic.
!
ðx  lÞ2
mðxÞ ¼ exp  ð9Þ
2r2

For classification using T1FS, the membership values of the test sample for each
class is computed using the parameters of the membership function (for example the
mean and the standard deviation of the observations in case of Gaussian membership
function) for that particular class and the sample belongs to the class having the
maximum value of membership. However incase the observations have varying
memberships (say, in case of Gaussian membership, the mean and standard deviation
are variable over a larger number of observations) as with signals taken at different
days in case of EEG signals, T1FS fail to provide accurate classification results. This
uncertainty in the primary membership function m(x) is overcome by assigning a
secondary membership function m(x, ~ m(x)) in T2FS [20, 21]. The union of all the
primary memberships for similar set of observations forms a region called the
Footprint of Uncertainty (FOU) bounded by the secondary membership function and
the minimum and maximum values of the primary memberships for the observations,
termed as the Lower Membership Function (LMF) and the Upper Membership
Function (UMF) respectively. In Interval Type-2 Fuzzy Systems (IT2FS) [22, 23],
the secondary membership function is uniform and assumes a constant value of 1 for
all values of m(x) in between the LMF and UMF and zero otherwise.
For using IT2FS in signal classification, let us consider P sets of similar
observations of each of K classes where the signal is represented by a feature space
FS of M × N dimensions, for M observations and N features in each set. For each
feature Fi (1 ≤ i ≤ N) in FS, considering the minimum and maximum values of Fi
240 S. Datta et al.

over the observations over P sets, the primary memberships, and consequently the
LMF, UMF and FOU are constructed. Say, a feature vector f corresponding to an
unknown instance of the signal has to be classified.
Each component fi of f (1 ≤ i ≤ N) is projected on the corresponding FOU to find
the intersections with the LMF and UMF of that component to obtain LMFi and
UMFi. Prior to projection, if it is found that fi falls outside the range of Fi
extrapolation is used.
For a particular class k the fuzzy t-norm of all LMFi and UMFi for 1 ≤ i ≤ N are
computed to obtain the LMFT,k and UMFT,k respectively. Using these values the
strength Sk of the class k (1 ≤ k ≤ K) is determined by the computation of the
centroid given by (10).

UMFT;k þ LMFT;k
Sk ¼ ð10Þ
2

Computing the strengths of all the classes, the class having the maximum
strength is determined to be the class of the test sample. The entire process for class
k is illustrated in Fig. 1.

UMF1

t-Norm
m(F1)

UMFN
UMFT,k

f1
F1
Mean Sk

LMFT,k

LMF1
UMF
t-Norm
m(FN)

LMF FOU
LMFN

fN FN

Fig. 1 Procedure for calculation of centroid or strength of a class k in the IT2FS approach, all
symbols having the same meanings as explained in Sect. 2.2
Cognitive Activity Classification from EEG Signals … 241

3 Experiments and Results

The steps of EEG signal processing in the present work has been illustrated in
Fig. 2.

3.1 EEG Acquisition

In the present work, we are interested in finding excitations corresponding to three

activities, namely, reading, watching a video with audio and relaxing. Accordingly,
EEG from the pre-frontal, frontal, occipital, parietal and temporal regions are
continuously recorded through the 10 electrodes FP1, FP2; F3, F4; O1, O2; P3, P4
and T3, T4 respectively of a 21 channel Neurowin system according to the 10–20
international standard of electrode placement [30] as shown in Fig. 3a, at a sam-
pling rate of 250 Hz.
EEG data is acquired from 4 subjects, 2 male and 2 female in the age group of
25 ± 5 years, for a period of 3 consecutive days. The pattern of stimulus used for
data acquisition is shown in Fig. 3b, where ‘STIMULUS’ corresponds to the visual
stimulus for activity A1 (reading the English sentences displayed) or the Audio-
Video stimulus for activity A2 (watching a video clip with sound) or a black screen
with instructions to relax with closed eyes for activity A3 (relaxing with closed
eyes). In all cases the respective stimuli are displayed on a screen using a projector
in front of the comfortably seated subject. The sequence of stimulus is repeated
5 times for each activity in random order for each day of experiment.

Data Pre- Feature Recognized

Stimulus Classification
Acquisition processing Extraction Activity

Fig. 2 Steps of EEG signal processing

(a) (b) Audio Audio Beep

Command Command Sound

Relax and wait

Start STIMULUS Stop
for next stimulus

1 30 1 5
Duration (seconds )

Fig. 3 a Electrode placement showing the selected electrodes in green, A1 and A2 are the
reference electrodes. b An instance of presented stimulus
242 S. Datta et al.

3.2 EEG Preprocessing

3.2.1 Filtering

The normal EEG bandwidth ranges between 0.5–70 Hz and is made up of the delta
(0.5–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz) and gamma (above
30 Hz) bands [16]. It was experimentally found that the significant changes in the
EEG spectrum for decoding the three activities is limited to 4–30 Hz, hence we
have considered EEG signals in the theta, alpha and beta bands for our work. To
extract the EEG signals in the desired frequency range, and thereby eliminate the
other frequencies, the acquired EEG is filtered using an Elliptical Band pass filter of
order 6 with 1 dB passband ripple and 50 dB stopband ripple and bandwidth
4–30 Hz.

3.2.2 Common Average Referencing

Common average referencing has been performed on the acquired EEG signals for
spatial filtering to remove the effect of interference between the signals of adjacent
channels. For each EEG channel, all the channels equally weighted are subtracted to
eliminate the commonality of that channel with the rest and preserve its specific
temporal features. Let the signal at the primary and the 10 channels be xi(t) and xj(t),
for i, j = 1–10. Then with equal weights for x1(t) through x10(t), we obtain (11).

1 X10
xi ðtÞ xi ðtÞ  xj ðtÞ ð11Þ
10 j¼1

3.3 Feature Extraction and Classification

AAR parameters have been computed using Kalman filtering as the estimation
algorithm and the order of 6 has been selected after trials with different orders and
finding the best performance. The AAR parameters are adapted at a rate given by
the update coefficient which is heuristically selected as 0.0085. For each electrode
the dimension of AAR features is 6. In case of Hjorth Parameters, the dimension of
features for each electrode is 3 while each of Hurst Exponents and Approximate
Entropy yield one feature per electrode. In the computation of Approximate
Entropy the value of embedding dimension has been selected as 2 and that of the
tolerance has been selected as 0.2× (standard deviation of the concerned time series)
from literature [24]. Features are extracted from the time series obtained at each
electrode and these features from the 10 electrodes are concatenated to obtain the
Cognitive Activity Classification from EEG Signals … 243

respective feature spaces. When feature spaces are combined, each feature space is
normalized with respect to its maximum value.
For classification, data from the 3 days of experiments have been utilized. 5 s of
data is taken as a single instance. For each day, 5 (times) × 30 s/5 s (instan-
ces) × (250 × 5) (samples/instance) i.e. 30 total instances each of 1,250 samples are
obtained from each class which is subjected to feature extraction. The resulting data
is cross validated to obtain the testing and training instances. Classification is
carried out in a One versus One approach i.e. binary classification considering the
instances of two classes at a time. Classification has been carried out using the
IT2FS approach as explained in the previous section. The performance is compared
that of a T1FS classifier [17, 19], a Support Vector Machine (SVM) [14], a Neural
Network (NN) classifier [14, 31] and a Naïve Bayes (NB) classifier [14]. In T1FS
and IT2FS classifiers Gaussian membership functions have been used for the
simplicity in determining the memberships using the mean and the standard devi-
ations of the data and also it is appropriate because of the nature of EEG features. In
IT2FS each set of observation comprise of the observations taken in a particular
day. The Fuzzy t-norm has been implemented by taking the minimum of the input
values. For SVM an RBF kernel is used with the width of the Gaussian kernel taken
as 1. The NN classifier is implemented with back-propagation learning using gra-
dient descent search [29] with 3 hidden layers. The NB classifier is used with the
assumption that the features have a normal distribution whose mean and covariance
are learned during the process of training.

3.4 Performance Analysis

The classification results are produced from the respective confusion matrices in
terms of Classification Accuracy (CA), Sensitivity (ST) and Specificity (SP) given by
(12), (13) and (14) respectively [32]. Here TP, TN, FP and FN denote the number of
samples classified as true positive, true negative, false positive and false negative
respectively. The ideal values of CA, ST and SP should each be very close to 1.

TP þ TN
CA ¼ ð12Þ
TP þ TN þ FP þ FN
TP
ST ¼ ð13Þ
TP þ FN
TN
SP ¼ ð14Þ
TN þ FP
 244

Table 1 Classification performance of IT2FS

Class S Features
AAR + Hjorth Hurst + AE AAR + Hjorth + Hurst + AE
CA ST SP CA ST SP CA ST SP
A1 versus A2 1 0.80 0.87 0.73 0.72 0.70 0.73 0.78 0.77 0.80
2 0.85 0.90 0.80 0.88 0.83 0.93 0.88 0.90 0.86
3 0.74 0.77 0.72 0.68 0.73 0.63 0.78 0.83 0.73
4 0.83 0.83 0.83 0.80 0.87 0.73 0.82 0.83 0.80
A1 versus A3 1 0.89 0.88 0.90 0.87 0.85 0.90 0.92 0.90 0.93
2 0.86 0.83 0.90 0.89 0.85 0.93 0.89 0.94 0.83
3 0.78 0.87 0.70 0.80 0.75 0.85 0.88 0.90 0.87
4 0.88 0.83 0.93 0.78 0.77 0.80 0.83 0.80 0.85
A2 versus A3 1 0.77 0.85 0.70 0.72 0.80 0.63 0.92 0.93 0.90
2 0.85 0.83 0.87 0.83 0.80 0.87 0.86 0.83 0.90
3 0.70 0.67 0.73 0.68 0.76 0.60 0.77 0.80 0.73
4 0.85 0.81 0.88 0.78 0.80 0.77 0.91 0.95 0.87
S. Datta et al.
Cognitive Activity Classification from EEG Signals … 245

Table 2 Comparison of classifier performance

Parameter (mean over all subjects) Classifiers (using AAR + Hjorth + Hurst + AE
features)
IT2FS T1FS SVM-RBF NB NN
Classification accuracy (%) 85.33 71.65 83.50 81.27 75.45
Rank 1.25 4.75 1.75 3.75 3.5

3.5 Experimental Results and Discussions

The results of classification with IT2FS is illustrated in Table 1, where S denotes the
Subject ID, A1, A2 and A3 denote the classes of activities as mentioned in Sect. 3.1.
It is observed that using a combination of all features a maximum of 85.33 %
classification accuracy is obtained on an average over all subjects and OVO clas-
sifications. The minimum classification accuracy does not go below 68 % in any of
the cases.
For comparison between the different classification algorithms, Friedman Test
[33], has been carried out for N = 4 databases of the 4 subjects and for k = 5
algorithms taking the mean classifier accuracy for each classification algorithm
used, as shown in Table 2. The null hypothesis states that all algorithms are
equivalent and hence their ranks Rj should be equal. The Friedman statistic is given
by (15).
" #
12N X
k
kðk þ 1Þ2
v2F ¼ R 
2
ð15Þ
kðk þ 1Þ j¼1 j 4

It is distributed according to v2F with k − 1 degrees of freedom. Here k is the

number of algorithms used and N is the number of datasets which are 5 and 4 in our
case respectively. The null hypothesis is rejected if evaluated v2F > v24;0:95 = 9.49,
which indicates, for 4° of freedom, the null hypothesis is correct to an extent of only
5 %. In this case v2F is found to be 13.6 hence the null hypothesis is rejected and the
classifier performance is evaluated by its rank.

4 Conclusion and Future Directions

In this paper cognitive activities have been classified from EEG signals using
different features and classifiers, with an aim at the development of a cognitive state
aware BCI device. A combination of four signal features, namely, AAR parameters,
Hjorth Parameters, Hurst Exponents and Approximate Entropy yield the best results
of 85.33 % classification accuracy employing an IT2FS classifier which is able to
perform better due to its inherent capability to deal with the uncertainties and
246 S. Datta et al.

variations in EEG signals. The future scopes of work include real time imple-
mentation of the present scheme.

Acknowledgments This study has been supported by University Grants Commission, India,
University of Potential Excellence Program (UGC-UPE) (Phase II) in Cognitive Science, Jadavpur
University and Council of Scientific and Industrial Research (CSIR), India.

References

1. Vallabhaneni A, Wang T, He B (2005) Brain—computer interface. In: He B (ed) Neural

Engineering. Kluwer/Plenum, Springer, New York, pp. 85–121. doi:10.1007/0-306-48610-5_3
2. Riva G, Vatalaro F, Davide M, Alcañiz M (2005) Ambient intelligence
3. Henricksen K, Indulska J, Rakotonirainy A (2002) Modeling context information in pervasive
computing systems. In: Pervasive computing. Springer, New York, pp 167–180
4. Haynes JD, Rees G (2006) Decoding mental states from brain activity in humans. Nat Rev
Neurosci 7(7):523–534
5. Mitchell TM, Hutchinson R, Niculescu RS, Pereira F, Wang X, Just M, Newman S (2004)
Learning to decode cognitive states from brain images. Mach Learn 57(1–2):145–175
6. Freeman WJ, Ahlfors SP, Menon V (2009) Combining fMRI with EEG and MEG in order to
relate patterns of brain activity to cognition. Int J Psychophysiol 73(1):43–52
7. Vuckovic A, Radivojevic V, Chen AC, Popovic D (2002) Automatic recognition of alertness
and drowsiness from EEG by an artificial neural network. Med Eng Phys 24(5):349–360
8. Gevins A, Smith ME (1999) Detecting transient cognitive impairment with EEG pattern
recognition methods. Aviat Space Environ Med 70(10):1018–1024
9. Wilson GF, Fisher F (1995) Cognitive task classification based upon topographic EEG data.
Biol Psychol 40(1):239–250
10. Pfurtscheller G, Neuper C (2001) Motor imagery and direct brain-computer communication.
Proc IEEE 89(7):1123–1134
11. Fabiani GE, McFarland DJ, Wolpaw JR, Pfurtscheller G (2004) Conversion of EEG activity
into cursor movement by a brain-computer interface (BCI). IEEE Trans Neural Syst Rehabil
Eng 12(3):331–338
12. Schaaff K, Schultz T (2009) Towards an EEG-based emotion recognizer for humanoid robots.
In: The 18th IEEE international symposium on robot and human interactive communication,
RO-MAN, September 2009, pp 792–796
13. Khasnobish A, Konar A, Tibarewala DN, Bhattacharyya S, Janarthanan R (2013) Object shape
recognition from EEG signals during tactile and visual exploration. Pattern Recognit Mach
Intell 459–464. Springer Berlin, Heidelberg
14. Lotte F, Congedo M, Lécuyer A, Lamarche F, Arnaldi B (2007) A review of classification
algorithms for EEG-based brain–computer interfaces. J Neural Eng 4
15. Besserve M, Jerbi K, Laurent F, Baillet S, Martinerie J, Garnero L (2007) Classification
methods for ongoing EEG and MEG signals. Biol Res 40(4):415–437
16. Teplan M (2002) Fundamentals of EEG measurement. J Measur Sci Rev 2(2):1–11
17. Zadeh LA (1988) Fuzzy logic. Computer 21(4):83–93
18. Mitra S, Pal SK (2005) Fuzzy sets in pattern recognition and machine intelligence. Fuzzy Sets
Syst 156(3):381–386
19. Chakraborty A, Konar A, Chakraborty U K, Chatterjee A (2009) Emotion recognition from
facial expressions and its control using fuzzy logic. IEEE Trans Syst Man Cybern Syst Hum
39(4):726–743
20. Karnik NN, Mendel JM, Liang Q (1999) Type-2 fuzzy logic systems. IEEE Trans Fuzzy Syst
7(6):643–658
Cognitive Activity Classification from EEG Signals … 247

21. Herman P, Prasad G, McGinnity T (2006) Investigation of the type-2 fuzzy logic approach to
classification in an EEG-based brain-computer interface, In: 27th Annual international
conference of the IEEE engineering in medicine and biology society. pp 5354–5357
22. Liang Q, Mendel JM (2000) Interval type-2 fuzzy logic systems: theory and design. IEEE
Trans Fuzzy Syst 8(5):535–550
23. Konar A, Chakraborty A, Halder A, Mandal R, Janarthanan R (2012) Interval type-2 fuzzy
model for emotion recognition from facial expression. Perception and machine intelligence.
Springer, New York, pp 114–121
24. Balli T, Palaniappan R (2010) Classification of biological signals using linear and nonlinear
features. Physiol Meas 31(7):903
25. Pfurtscheller G, Neuper C, Schlogl A, Lugger K (1998) Separability of EEG signals recorded
during right and left motor imagery using adaptive autoregressive parameters. IEEE Trans
Rehabil Eng 6(3):316–325
26. Nai-Jenand H, Palaniappan R (2004) Classification of mental tasks using fixed and adaptive
autoregressive models of EEG signals. In: 26th Annual international conference of the IEEE
engineering in medicine and biology society, IEMBS’04, Sept 2004 vol 1. pp 507–510
27. Vidaurre C, Krämer N, Blankertz B, Schlögl A (2009) Time domain parameters as a feature
for EEG-based brain–computer interfaces. Neural Netw 22(9):1313–1319
28. Acharya UR, Faust O, Kannathal N, Chua T, Laxminarayan S (2005) Non-linear analysis of
EEG signals at various sleep stages. Comput Methods Programs Biomed 80(1):37–45
29. Konar A (2005) Computational intelligence principles, techniques and applications. Springer,
New York
30. Dornhege G (2007) Towards brain-computer interfacing. MIT Press, Cambridge
31. Mitchell TM (1997) Machine learning. McGraw Hill, Burr Ridge, p 45
32. Fielding AH, Bell JF (1997) A review of methods for the assessment of prediction errors in
conservation presence/absence models. Environ Conserv 24(1):38–49
33. Conover WJ, Iman RL (1981) Rank transformations as a bridge between parametric and
nonparametric statistics. Am Stat 35(3):124–129
Performance Analysis of Feature
Extractors for Object Recognition
from EEG Signals

Anwesha Khasnobish, Saugat Bhattacharyya, Amit Konar

and D.N. Tibarewala

Abstract Recognition of objects from EEG signals requires selection of appropriate

feature extraction and classification techniques with best efficiency in terms of
highest classification accuracy with lowest run time for its applications in real time.
The objective of this paper is to analyze the performance of various feature
extraction techniques and to choose that particular method which can be imple-
mented in real time system with best efficiency. The EEG signals are acquired from
subjects while they explored the objects visually and visuo-tactually. Thus acquired
EEG signals are preprocessed followed by feature extraction using adaptive auto-
regressive (AAR) parameters, ensemble empirical mode decomposition (EEMD),
approximate entropy (ApEn) and multi-fractal detrended fluctuation analysis
(MFDFA). The performance of these features are analyzed in terms of their
dimension, extraction time and also depending upon the classification results pro-
duced by three classifiers [Support Vector machine (SVM), Naïve Bayesian (NB),
and Adaboost (Ada)] independently according to classification accuracy, sensitivity
and classification times. The experimental results show that AAR parameter has an
optimum dimension of 36 (not too large like EEMD i.e. 7,680 or too small like ApEn
i.e. 6) and required minimum extraction as well as classification time of 0.59 and
0.008 s respectively. AAR also yielded highest maximum classification accuracy
and sensitivity of 80.95 and 92.31 % respectively with NB classifier. Thus AAR
parameters can be chosen for real time object recognition from EEG signal along
with Naïve Bayesian classifier.

A. Khasnobish (&)
D.N. Tibarewala

School of Bioscience & Engineering, Jadavpur University, Kolkata, India
e-mail: [email protected]
D.N. Tibarewala
e-mail: [email protected]
S. Bhattacharyya
A. Konar
Department of Electronics & Telecommunication Engineering,
Jadavpur University, Kolkata, India
e-mail: [email protected]
A. Konar
e-mail: [email protected]

© Springer India 2015 249

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_23
250 A. Khasnobish et al.

Keywords Object recognition Electroencephalography Adaptive autoregressive



parameter Ensemble empirical mode decomposition Approximate entropy



Multifractal detrended fluctuation analysis Naïve bayesian classifier Support



vector machine Adaboost

1 Introduction

Object recognition is a vital process in understanding our environment, which in

turn enhances exploration and maneuvering. Both vision and tactile sensation along
with cognitive processing is essential for proper object recognition. The visual
sensation is related to the occipital lobe of human brain, whereas its tactile coun-
terpart originates from the parietal lobe and the cognitive processing in frontal and
fronto-central regions. The synchronous activity of both these sensory systems
facilitates object classification in human brain [1, 2]. Thus analyzing visually and
tactually stimulated electroencephalography (EEG) signals from occipital and
parietal and fronto-central channels provide information on object perception.
Object recognition has been performed by many researchers mainly using computer
vision [3], tactile signals, and tactile images [4, 5]. But object recognition directly
from EEG signals [6] will help in enhancing the controllability of the brain com-
puter interface (BCI) [7] controlled robots.
Recognition of objects from EEG signals requires selection of appropriate fea-
ture extraction and classification techniques with best efficiency in terms of highest
classification accuracy with lowest run time for its applications in real time. The
objective of this paper is to analyze the performance of various existing feature
extraction techniques to choose that particular method which can be implemented in
real time system with best efficiency.
The EEG signals are non-stationary, non-Gaussian signals [8, 9]. Various linear
and non linear features extraction methods in time as well as frequency domain are
applied for EEG signals viz. power spectral density (PSD), Hjorth parameters,
wavelet transforms (WT), autoregressive parameters (AR), adaptive autoregressive
(AAR) parameters, Hurst coefficients, approximate entropy (ApEn), extreme energy
ratio criterion (EER), empirical mode decomposition (EMD), detrended fluctuation
analysis (DFA), and statistical parameters [10, 11]. Depending of the type of
stimuli, the performances these features vary.
In this work we presented the subjects with visual and visuo-tactile stimuli of ten
different objects and corresponding EEG signals are acquired. For visual stimulation
the subjects are presented with pictures of the objects and they examine the objects
only visually. For the visuo-tactile stimulation, the subjects explored the objects by
visual examination as well as palpated them at the same time. During the object
exploration the EEG signals are acquired from six electrode locations, namely FC5,
FC6, P7, P8, O1 and O2. The electrodes are chosen so as to acquire the brain signals
from frontal, parietal and occipital relating the cognitive processing, somato-sensory
Performance Analysis of Feature Extractors … 251

and visual sensations respectively. The acquired EEG signals are preprocessed and
features are extracted. In this work we used four feature extraction techniques,
comprising two linear methods viz. adaptive autoregressive (AAR) parameter and
ensemble empirical mode decomposition (EEMD), and two non linear techniques
viz. approximate entropy (ApEn) and multifractal detrended fluctuation analysis
(MFDFA). The performances of these extracted features are analyzed in terms of
their dimension, extraction time as well as classification accuracy, sensitivity and
runtimes on classification by three classifiers namely, Support Vector Machine
(SVM), Naïve Bayesian (NB) and Adaboost classifier [12–15]. It is observed that
AAR parameter yielded best results with all the three classifiers in both visual and
visuo-tactile stimulated signals. Moreover Adaboost classifier with SVM as base
classifier produced better results with all four features, performing the best with
AAR parameters, followed by NB and SVM classifiers. However, the run time of
NB classifier is the least, which is very essential for online classifications.
The rest of the paper is presented in five sections. The introduction is followed
by the methodology in Sect. 2. The selected EEG modalities are described in
Sect. 3. Experimental paradigm and the results are depicted in Sect. 4. The con-
cluding remarks are stated in Sect. 5.

2 Methodology

The acquired EEG signals are subjected to preprocessing followed by feature

extraction for classification of corresponding objects.

2.1 Preprocessing

The power spectrum of the visual and visuo-tactual stimulated EEG signal is found
to be in the 4–16 Hz (Fig. 1). The acquired EEG signal is thus filtered using an
elliptical band pass filter of order 8 and bandwidth 4–16 Hz. The filtering removes
the noises due to power line interference, head movement, and eye blinks as well as
it extracts the signal of the required bandwidth.

2.2 Feature Extraction

Appropriate feature extraction is a pre-requisite for effective classifications. The aim

of this paper is to compare the performances of various features based on classification
results to ultimately select a competent technique for real time classification of object
recognition from EEG signals. For this purpose we have considered two linear and
two non-linear features. Among the two linear features, i.e. adaptive auto-regressive
252 A. Khasnobish et al.

Fig. 1 a Electrode placement, b Different object shapes used in the experiments. 1 Cone, 2 Cube,
3 Cylinder, 4 Sphere, 5 Prism, 6 Hemisphere, 7 Square based pyramid, 8 Hexagonal based
cylinder, 9 Lock and 10 Mouse

(AAR) parameters and ensemble empirical mode decomposition (EEMD), the first
one is a time domain feature and the other is a frequency domain feature. Two non-
linear features viz. approximate entropy (ApEn) and multifractal detrended fluctua-
tion analysis (MFDFA) are also taken in account for the present work.

2.2.1 Adaptive Auto-regressive Parameter

Autoregressive parameters (AR) and adaptive autoregressive parameters (AAR) are

parametric approaches for EEG analysis [16, 17]. EEG signals non-stationary
random signals. AR has the capability to characterize the stochastic behavior of
EEG signals without requirement of any prior information of any relevant fre-
quencies. The AR uses various stationary processes for estimating the parameters.
Since the EEG signals are non-stationary, moving window method can be applied
for AR parameter estimation, but it is computationally extensive. To overcome
these inherent drawbacks of AR approaches, AAR methods are implemented for
EEG feature extraction since it yields time varying parameters with high temporal
resolution. AAR model describes the signal x(t) defined by Eq. (1).

xðtÞ ¼ p1 ðtÞ  xðt  1Þ þ


þ pk ðtÞ  yðt  kÞ þ sðtÞ ð1Þ

where, pk(t) are the parameters, s(t) is white noise with zero mean and k is the order
of the model.
There are various algorithms to estimate the parameters of AAR such as, least mean
square (LMS), Kalman filter, recursive AR, recursive least square (RLS) and as like.
In this work, AAR model with Kalman filter method for parameter estimation is
considered. The order and update coefficient (UC) are experimentally determined to
be 6 and 0.0085 respectively.
Performance Analysis of Feature Extractors … 253

2.2.2 Ensemble Empirical Mode Decomposition

Empirical mode decomposition (EMD) is method for adaptive analysis of non-

stationary and non-linear signals [18]. This method separates a signal in fast and
slow oscillations by local data-driven approach. Mode-mixing i.e. presence of very
divergent oscillations in one mode or presence of similar oscillations in different
modes, is a familiar problem encountered in EMD. This setback is overcome by
performing EMD over an ensemble of white Gaussian noise and the signal. This
progression is termed as ensemble empirical mode decomposition (EEMD) [18].
Addition of white noise takes the advantage of dyadic filter bank behavior of EMD
by populating the total time-frequency space and thus eliminates the mode-mixing
problem. In EMD the signal is decomposed in number of intrinsic mode functions
(IMFs), also called modes. A number ensemble trials are obtained by adding dif-
ferent apprehensions of white noise to the original signal x(t) and their corre-
sponding IMFs computed. The ensemble trials are represented by Eq. (2).

x j ½n ¼ x½n þ g j ½n ð2Þ

where j = 1, 2, …, J, and gj[n] is Gaussian noise of various apprehensions. IMFs are

obtained from completely decomposing each xj[n]. EEMD calculates the mean of
the obtained IMFs, termed as true IMF (IMFtrue).
We have computed five IMFs of the preprocessed EEG signals for feature
extraction.

2.2.3 Approximate Entropy

Approximate entropy (ApEn) is a statistical method proposed by Pincus [19] and [20]
for quantifying the unpredictability or irregularity of stochastic signals. For non-linear
time-series signals this technique measures the randomness or degree of complexity.
The steps in estimation of ApEn are as follows:
I. Let a time series signal containing N data points be represented as X = [x(1), x
(2),…, x(N)]. Specify the embedding dimension or window length (l) and the
tolerance (t).
II. l vectors X(1), X(2),…, X(N − l + 1), representing l successive values of x is
defined as

XðiÞ ¼ ½xðiÞ; xði þ 1Þ; . . .; xði þ l  1Þ ð3Þ

where i = 1, 2,…, N − l + 1.
III. Compute the distance between two vectors X(i) and X(j) by calculating the
maximum absolute difference between their corresponding scalar elements, as
given by Eq. (4).
254 A. Khasnobish et al.

d½XðiÞ; XðjÞ ¼ maxk¼1;...;l j xði þ k  1Þ  xðj þ k  1Þj ð4Þ

IV. For a given X(i) find the number of j = 1, 2, …, N − l + 1, j ≠ i, such that d[X(i),
X(j)] ≤ t and denote is as Ml(i). Then for i = 1, 2,…, N − l + 1, compute

M l ðiÞ
Ctl ðiÞ ¼ ð5Þ
Nlþ1

This measures the frequency of patterns alike the one given by window of
length l and tolerance t.
V. Compute the natural logarithm of each Ctl ðiÞ and find the mean over all i, as
given by Eq. (6).

1 X
Nlþ1
ult ðtÞ ¼ ln Ctl ðiÞ ð6Þ
N  l þ 1 i¼1

VI. Compute Cilþ1 ðiÞ and ulþ1

t ðiÞ by increasing the dimension to (l + 1) and
repeating the Steps II–V. Calculate approximate entropy (ApEn) according to
Eq. (7).

ApEnðl; t; NÞ ¼ ul ðtÞ  ulþ1 ðtÞ ð7Þ

For feature extraction of the preprocessed EEG signal experimentally we have

determined l = 2 and t = 0.02.

2.2.4 Multifractal Detrended Fluctuation Analysis

If a time-series signal repeats itself on the subintervals of the signal, then it posses
scale invariant structures. For EEG signals, the scale invariant structures of inter-spike
interval of firing of the neurons are capable of discriminating between the neural
activities of brain. Alterations in scale invariant structure of bio-signals indicate
adaptability of physiological processes. Detrended fluctuation analysis (DFA)
quantifies the self-resemblance of a signal [21, 22]. Unlike fluctuation analysis (FA),
DFA is not affected by non-stationarities of a signal which measures long range
correlations of non-stationary signals. Time series with complicated time behavior
necessitate different scaling exponents for different part of the series. In such case
multi-fractal analysis is performed with provides multitude scaling exponents for
complete description of complex scaling behavior. Regular partition function multi-
fractal formalism developed for multifractal characterization and normalized sta-
tionary measure is the basis of simplest MFA. Non-stationary time series signals are
affected by trends that cannot be normalized. Due to this characteristic of non-
stationary signals MFA produces erroneous results, which is overcome by imple-
menting multi-fractal detrended fluctuation analysis (DFA) [21, 22].
Performance Analysis of Feature Extractors … 255

Let xk is a time series of length N of compact support that xk = 0 for an insig-

nificant amount of values. The steps required to estimate MFDFA is as follows:
I. Compute the ‘Profile’ P(i)

X
i
PðiÞ ¼ ½xk  x; i ¼ 1; . . .; N ð8Þ
k¼1

II. Divide profile in a number of non-overlapping segments denoted by Nl = N/l,

of equal length l. The same process is repeated from end to start to the series to
consider the small parts that can remain at the end of the series. Thus we obtain
total 2Nl segments.
III. Compute the local trend for each of the obtained segments by least square fit of
the series and calculate the variance, as depicted by Eq. (9).

1X l
F 2 ðl; vÞ ¼ fP½ðv  1Þl þ i  yv ðiÞg2
l i¼1

where v is a segment such that v = 1,…, Nl, and

1X l
F2 ðl; vÞ ¼ fP½N  ðv  Nl Þl þ i  yv ðiÞg2 ð9Þ
l i¼1

for v = Nl + 1,…, 2Nl, where yv(i) is the fitting polynomial in the segment v.
IV. Calculate qth order fluctuation function by averaging over all segments.
( )1=q
1 X2Nl
q=2
Fq ðlÞ ¼ ½F 2 ðl; vÞ ð10Þ
2Nl v¼1

where q can take any value other than zero.

To determine the dependency of generalized q dependent fluctuations on time
scale l, repeat Steps II–IV.
V. Analyze the log-log plots of Fq(l) versus l for each value of q for quantifying
the scaling behavior of fluctuation function. For long-range power-law cor-
rected series, xi, Fq(l) increases as power-law for large values of l.

Fq ðlÞ  lhðqÞ ð11Þ

The generalized Hurst exponent h(q) may be q-dependent.

We have used MFDFA of 7th order fitting polynomial (i.e. yv(i)) and varied q in
the range −5 to 5 with 101 discreet intervals.
256 A. Khasnobish et al.

2.3 Classification

The extracted features are classified separately using three supervised classifiers
namely support vector machine (SVM), Naïve Bayesian (NB) and Adaboost (Ada)
[12–15]. These features are classified in corresponding object classes by imple-
menting these classifiers. The one-against-all (OAA) approach is implemented for
classification of ten objects.
SVM is a supervised, non-probabilistic binary classifier that separates the data
points by a hyperplane, by maximizing its distance from the support vectors. For
classification, it considers only the support vectors, thus the search space is reduced,
and pattern recognition is performed in less time. Due the formation of the
hyperplane with maximum margin, the efficiency is also high in case of SVM
classifiers. On the other hand Naïve Bayesian is a probabilistic classifier, which
makes conditional independence assumptions which further facilitates the reduction
of high complexity in case of general Bayesian classifiers. Adaboost (adaptive) is an
ensemble classifier, utilizing a number of weak learners to build a strong learner
iteratively and is suitable for both binary and multi-class classification problems. In
the present study, for Adaboost we have used SVM as the base classifier.

3 EEG Modalities

3.1 Frequency Band Selection

The bandwidth of EEG signals is in the range 0.5–70 Hz. As evident from the
power spectrum of the visually and visuo-tactually stimulated EEG signals’ activity
lie in the range of 4–16 Hz. Thus we have considered 4–16 Hz bandwidth, com-
prising of theta, alpha and central beta rhythms.

3.2 Electrode Selection

The acquired EEG signals are either visually or visuo-tactually stimulated. Here we
are concerned with decoding object recognition from EEG signals. Complete course
of object recognition is related to visual and tactile sensory information and also
cognitive processing. The brain regions related to vision, tactile sensation, and
cognition are occipital, parietal, and fronto-central regions. Thus we have acquired
the EEG signals from six electrodes, viz. O1, O2; P7, P8; and FC5, FC6, where
each pair are positioned on occipital, parietal and fronto-central regions (Fig. 1a).
Performance Analysis of Feature Extractors … 257

4 Experiments and Results

4.1 Experimental Paradigm

Ten rigid objects consisting of eight geometrical shapes and two non-geometrical
objects are considered for experiments (Fig. 1b). Thirteen subjects, (8 male and 5
female) of age group 22 ± 5 years, took part in the experiments. The subjects are all
right handed with normal or corrected to normal vision. The experimental procure is
described to them and they signed a consent form before starting the experiments.
EEG from six electrode channels (FC5, FC6, P7, P8, O1 and O2) is acquired using
Emotive EEG.
The subjects are presented with audio-visual cues. At the start of the experiment
there appears a blank screen for 30 s to relax the subject and detect the baseline EEG,
followed by a fixation cross, indicating the subject to get ready. For only visual
stimulation phase, a picture of a particular object appears on screen for 5 s. In case of
visuo-tactile stimulation, followed by the fixation cross the subjects are instructed to
explore the object by palpating for 5 s whose picture appears on the screen. At the end
of 5 s a beep sound is indicates the end of object examination and again a blank screen
appears for 10 s. In this manner every object is examined 10 times by each subject in
two experimental phase i.e. visual stimulation and visuo-tactual stimulation phases.

5 Results

Four features (i.e. AAR, EEMD, ApEn, MFDFA) are extracted from the prepro-
cessed EEG signals. Each feature space is tenfold cross validated to form corre-
sponding training and test instances which are classified independently using three
classifiers, i.e. SVM, NB and Adaboost. The efficacy of each feature is analyzed
depending upon the classification results based on three metrics viz. classifier run
time, classification accuracy and sensitivity. The feature dimension and extraction
time are also considered for their performance estimation.
Table 1 presents the feature dimensions over all six electrodes and also their
mean extraction times. It is evident that EEMD has highest dimension as well as
needs longest time for extraction, thus is inapplicable in real-time scenarios. As
noted from Table 1, AAR requires the minimum amount of computational time.

Table 1 Features’ dimension

and extraction time Features Feature dimension Mean extraction time
(s)
1. AAR 36 00.59
2. EEMD 7,680 43.24
3. ApEn 6 00.83
4. MFDFA 606 03.42
258 A. Khasnobish et al.

Table 2 Classification results

Features Classifier Run time (s) Classification Sensitivity (%)
accuracy (%)
AAR SVM 0.015 Min. 52.38 Min. 78.57
Max. 85.71 Max. 94.12
Av. 72.12 Av. 85.55
NB 0.008 Min. 52.38 Min. 83.33
Max. 80.95 Max. 92.31
Av. 70.06 Av. 85.93
Ada 0.926 Min. 66.67 Min. 82.35
Max. 95.24 Max. 94.74
Av. 80.95 Av. 86.53
EEMD SVM 0.055 Min. 57.14 Min. 82.35
Max. 80.95 Max. 88.89
Av. 65.98 Av. 83.36
NB 0.032 Min. 51.14 Min. 75.00
Max. 80.50 Max. 100
Av. 69.12 Av. 84.40
Ada 12.37 Min. 66.67 Min. 82.35
Max. 85.71 Max. 89.47
Av. 78.23 Av. 85.54
ApEn SVM 0.006 Min. 47.62 Min. 76.92
Max. 71.43 Max. 92.86
Av. 56.46 Av. 82.39
NB 0.004 Min. 66.67 Min. 73.73
Max. 80.95 Max. 88.89
Av. 64.62 Av. 82.69
Ada 0.881 Min. 71.43 Min. 81.25
Max. 85.71 Max. 88.89
Av. 75.5 Av. 85.06
MFDFA SVM 0.013 Min. 47.62 Min. 49.21
Max. 71.43 Max. 71.40
Av. 53.74 Av. 80.65
NB 0.012 Min. 47.62 Min. 61.90
Max. 76.19 Max. 76.19
Av. 59.86 Av. 83.26
Ada 1.172 Min. 76.21 Min. 80.23
Max. 81.17 Max. 89.33
Av. 80.95 Av. 85.00

The classification results are depicted in Table 2. The run time is the time in
seconds taken by the classifier to classify the corresponding features. The classi-
fication accuracy [23] is presented in three terms, i.e. minimum (Min.), maximum
Performance Analysis of Feature Extractors … 259

(Max.) and average (Av.) over all subjects and classes for both visual and visuo-
tactile stimulation phases. The sensitivity [23] is the measure of detecting the
positives correctly, as defined by Eq. (12).

True Positive
Sensitivity ¼ ð12Þ
True Positive þ False Negative

Since classification is performed on one-against-all basis, thus we are concerned

with detecting a particular class with respect to other classes taken altogether as
another class. So the rate of correctly detecting the true positives, i.e. our class of
concern for particular instance, is the main aim. The sensitivity is also presented as
Min., Max., and Av.
It is observed from the Table 2 that the AAR parameter requires the least clas-
sification time of 0.008 s with Naïve Bayesian (NB) classifier. The mean classifi-
cation rates over all three classifiers for AAR, EEMD, ApEn and MFDFA are 74.37,
71.11, 65.52 and 64.85 % respectively. Their corresponding mean sensitivities are
86, 84.43, 83.38, and 82.97 % respectively. In terms of both classification accuracy
and sensitivity AAR parameter performed best followed by EEMD, ApEn and
MFDFA. Among the three classifiers though Adaboost showed highest classification
accuracies and sensitivities with every feature, but its run time is also the longest,
making it unsuitable for online classifications. The next best classification accuracy
and sensitivity is observed for Naïve Bayesian classifier for each feature.

6 Conclusions

EEG signals are acquired from thirteen subjects while they explore ten rigid objects
visually and visuo-tactually. The acquired EEG signals are preprocessed followed
by feature extraction using four techniques i.e. adaptive autoregressive (AAR)
parameters, ensemble emipirical mode decomposition (EEMD), approximate
entropy (ApEn) and multi-fractal detrended fluctuation analysis (MFDFA). The
performance of these features are analyzed in terms of their dimension, extraction
time and also depending upon the classification results produced by three classifiers
(SVM, NB, and Adaboost) independently according to classification accuracy,
sensitivity and classification times. The experimental results show that AAR
parameter has an optimum dimension (not too large like EEMD or too small like
ApEn) and required minimum extraction as well as classification time. The clas-
sification accuracies and sensitivities are also found to be the highest for AAR
parameters. Though Adaboost performed with highest classification accuracy and
sensitivity, but due to is long execution time it cannot be applied in real time
recognitions. The NB classifier classified the features with classification accuracy
and sensitivity better than SVM with least execution time. Thus it can be concluded
that among all the features considered, AAR parameters can be chosen for real time
object recognition from EEG signal along with Naïve Bayesian classifier.
260 A. Khasnobish et al.

In future we will implement more feature and classifiers for analysis. We are also
working on implementation of feature selection techniques to improve the efficacy
of object recognition from visually and visuo-tactually stimulated EEG signals.

Acknowledgments This study has been supported by University Grants Commission, India,
University of Potential Excellence Program (UGC-UPE) (Phase II) in Cognitive Science, Jadavpur
University and Council of Scientific and Industrial Research (CSIR), India.

References

1. Schacter DL, Gilbert DL, Wegner DM (2009) Psychology, 2nd edn. Worth Publishers, New
York
2. Mishkin M, Ungerleider LG (1982) Contribution of striate inputs to the visuospatial functions
of parieto-preoccipital cortex in monkeys. Behav Brain Res 6(1):57–77
3. Kuo CC, Yau HT (2006) A new combinatorial approach to surface reconstruction with sharp
features. IEEE Trans Visual Comput Graphics 12(1):73–82
4. Pezzementi Z, Reyda C, Hager GD (2011) Object mapping, recognition and localization from
tactile geometry. In: Proceedings of IEEE international conference robotics and automation,
pp 5942–5948
5. Singh G et al (2012) Object shape recognition from tactile images using regional descriptors.
In: Fourth world congress on nature and biologically inspired computing (NaBIC) 2012,
pp 53–58
6. Khasnobish A, Konar A, Tibarewala DN, Bhattacharyya S, Janarthanan R (2013) Object shape
recognition from EEG signals during tactile and visual exploration. In: Accepted in international
conference on pattern recognition and machine intelligence (PReMI), 10–14 Dec 2013
7. Vallabhaneni A, Wang T, He B (2005) Brain–computer interface in neural engineering.
Springer, Heidelberg, pp 85–121
8. Teplan M (2002) Fundamentals of EEG measurement. Meas Sci Rev 2(2):1–11
9. Sanei S, Chambers JA (2007) Brain computer interfacing. EEG Signal Process, pp 239–265
10. Yom-Tov E, Inbar GF (2002) Feature selection for the classification of movements from single
movement-related potentials. IEEE Trans Neural Syst Rehabil Eng 10:170–178
11. Bhattacharyya S, Khasnobish A, Konar A, Tibarewala DN (2010) Performance analyisis of
LDA, QDA and KNN algorithms in left-right limb movement classification from EEG data.
In: Accepted for oral presentation in international conference on systems in medicine and
biology, IIT Kharagpur, 2010
12. Tae-Ki A, Moon-Hyun K (2010) A new diverse AdaBoost classifier. In: International
conference on artificial intelligence and computational intelligence (AICI) 2010, pp 359–363
13. Cunningham P (2009) Evaluation in machine learning: objectives and strategies for
evaluation. In: European conference on machine learning and principles and practice of
knowledge discovery in databases 2009, p 26
14. Daniel WW (2002) Biostatistics. Hypothesis testing, 7th edn. Wiley, New York, pp 204–229
15. Thulasidas M, Guan C, Wu J (2006) Robust classification of EEG signal for brain computer
interface. IEEE Trans Neural Syst Rehabil Eng 14(1):24–29
16. Vickenswaran J, Samraj A, Kiong LC (2007) Motor imagery signal classification using
adaptive recursive band pass filter and adaptive autoregressive models for brain machine
interface designs. J Biol Life Sci 3(2):116–123
17. Schloegl A et al (1997) Using adaptive autoregressive parameters for a brain-computer-
interface experiment. In: Proceedings of the 19th annual international conference of the IEEE
engineering in medicine and biology society 1997, vol 4, pp 1533–1535
Performance Analysis of Feature Extractors … 261

18. Torres ME et al (2011) A complete ensemble empirical mode decomposition with adaptive
noise. In: IEEE international conference on acoustics, speech and signal processing (ICASSP)
2011, pp 4144–4147
19. Pincus SM (1991) Approximate entropy as a measure of system complexity. In: Proc Natl
Acad Sci USA 88(6):2297–2301
20. Lei W et al (2007) Feature extraction of mental task in BCI based on the method of
approximate entropy. In: 29th annual international conference of the IEEE engineering in
medicine and biology society, EMBS 2007, pp 1941–1944
21. Kantelhardt JW et al (2002) Multifractal detrended fluctuation analysis of nonstationary time
series. J Physica A 316, 82:1–14
22. Ihlen EAF (2012) Introduction to multifractal wavelet and detrended fluctuation analyses.
Front Physiol: Fractal Physiology 3(141):1–18
23. Mahajan K, Rajput SM (2012) A comparative study of EEG and SVM for EEG classification.
J Eng Res Technol 1(6):1–6
Rectangular Patch Antenna Array Design
at 13 GHz Frequency Using HFSS 14.0

Vasujadevi Midasala, P. Siddaiah and S. Nagakishore Bhavanam

Abstract This paper presents a new element antenna array of rectangular topology
microstrip patches is introduced to operate at Ku band. The antenna has been
designed as an arrays of patches, where number of elements, spacing’s and feeding
currents has been optimized to fulfil the requirements of low side lobe level and
good cross polarization. The operating frequency range of antenna array is from 12
to 18 GHz. The antenna has been designed and simulated on FR4 Substrate with
dielectric constant of 4.4. This paper also presents that, the detail steps of designing
and simulating the rectangular patch antenna and rectangular patch antenna Array,
in Ku-band. The design is analysed by Finite Element Method (FEM) based HFSS
Simulator Software 14.0 by which return loss, Impedance, 3D polar plot, Direc-
tivity and Gain of the antenna are computed. The simulated results are shows that
the proposed antenna provides good performance in terms of return loss and
radiation pattern for dual frequency applications.

Keywords Microstrip antenna Rectangular patch antenna
HFSS 14.0
Return




loss Impedance 3D polar plot Directivity Gain

1 Introduction

In Modern Communication Systems the Antennas are the most important compo-
nents to create a communication link. Microstrip Patch antennas are widely used in
wireless communication systems because of, they are low profile, of light weight,

V. Midasala (&)
Department of ECE, JNTUA, Ananthapuranu, India
e-mail: [email protected]
P. Siddaiah
S.N. Bhavanam
University College of Engineering and Technology, Acharya Nagarjuna University,
Nagarjuna Nagar, Guntur, Andhra Pradesh, India
e-mail: [email protected]

© Springer India 2015 263

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_24
264 V. Midasala et al.

low cost, conformal design, low power handling capacity and easy to integrate and
fabricate. They can be designed in a variety of shapes in order to obtain enhanced
gain and bandwidth. Microstrip Patch Antenna implementation is a mile stone in
wireless communication systems.
The design of microstrip patch antenna, operating in Ku band is a very difficult
task. Ku band is primarily used in the satellite communication. Mostly for, fixed
and broadcast services and for specific applications for NASA. Ku-band is also
used for satellite from Remote Location (RL) back to a television network studio
for editing and broadcasting. Ku-band provides reliable high-speed connectivity
between the personal organizers and other wireless digital appliances.
The proposed model is one such antenna which is a Rectangular Microstrip fed
patch antenna. It can be operated at Ku Band. In addition to this, Rectangular Patch
Antenna Arrays designed. Its operation in Ku band the proposed antenna is also a
dual band antenna. Dual frequency antenna is mainly used in applications where
transmission and reception should be done by using same antenna. Many dual band
antennas have been improved to face the rising demands of a modern portable
wireless Communication Devices.

2 Design of Proposed Antenna

In this paper Rectangular Microstrip Patch Antenna Array Design at 13 GHz fre-
quency has been modelled and simulated at Ku-band. The patch (radiating part) is
the dominant figure of a microstrip antenna; the other components are the substrate
and ground, which are the two sides of the patch.
The above figures can represent that, the Rectangular Microstrip Patch Antenna
and Rectangular Microstrip Patch Antenna Array Design. Figure 1 can explains about
single element patch antenna design but from Fig. 2 we can found that, it consists an
array of Rectangular Microstrip Patch Antennas by this we can get better gain.

Method used to analyze antenna: FEM (Finite Element Method)

Excitation Technique Used: Probe Feeding
The spacing between antenna elements: λ/4

Fig. 1 Rectangular
microstrip patch antenna
design
Rectangular Patch Antenna Array Design at 13 GHz Frequency … 265

Fig. 2 Rectangular
microstrip patch antenna array
design

The design considerations are also shown below:

Design Considerations

Substrate material: FR4

Relative permittivity 2.2, loss tangent 0.0009
Position “−subX/2, −subY/2, 0 cm”
“−1.15cm, −0.95cm, 0 cm”
XSize subX 2.3 cm
YSize subY 1.9 cm
ZSize subH 62 mil
Patch:
Position “−patchX/2, −patchY/2, subH”
“−0.455 cm, −0.335 cm, 62 mil”
Axis Z
XSize patchX 0.91 cm
YSize patchY 0.67 cm

The Table 1 can give the corresponding values of Design Parameters.

Table 1 Design parameters

and corresponding values Design parameter Value
Operating frequency 13.28 GHz
Dielectric constant of the substrate 4.4
height of the substrate 0.157 cm
Length of substrate 3 cm
Width of substrate 3 cm
Height of coax pin 0.157 cm
Radius of coax pin 0.2 mm
266 V. Midasala et al.

Name Return Loss Patch_Antenna_ADKv1

m10.00 13.4246 -25.8444 Curve Info
m2 12.8032 -10.0095 dB(St(1,1))
m3 14.0630 -10.0046 Setup1 : Sweep1

-5.00

m2 m3
-10.00
dB(St(1,1))

-15.00

-20.00

-25.00 m1

-30.00
6.00 8.00 10.00 12.00 14.00 16.00 18.00 20.00
Freq [GHz]

Fig. 3 Return loss for single patch antenna

3 Simulation Results Using HFSS

The design is analyzed by Finite Element Method. The return loss, Impedance, 3D
polar plot, Directivity and peak gain is obtained by using HFSS 14.0. The results
are shows below:

Return Loss

The Fig. 3 can represents the Return Loss of Single patch antenna. At −10 db we
can get a Band Width −1.25 Ghz. And have a gain of 6.844 db.
The Fig. 4 can represents the Return Loss of patch antenna Array Design. At
−10 db we can get a Band Width −1.25 Ghz. And have a gain of 8.5548 db.

Impedance Diagrams

Figures 5 and 6 can shows about Impedance diagram of Single Patch antenna and
patch antenna Array Design.

3D Polar Plots

Figures 7 and 8 can show 3D Polar Plots of Single Patch antenna and patch antenna
Array Design.

Directivity
For Single Patch Antenna: 6.375
For Array Antenna Design: 6.874
Rectangular Patch Antenna Array Design at 13 GHz Frequency … 267

Name X Y XY Plot 1 HFSSDesign1 ANSOFT

m1 13.1667 -28.0277 Curve Info

dB(S(1,1))
Setup1 : Sweep

-5.00

-10.00

-15.00
dB(S(1,1))

-20.00

-25.00

m1

-30.00
5.00 7.50 10.00 12.50 15.00 17.50 20.00
Freq [GHz]

Fig. 4 Return loss for array design

Fig. 5 Impedance for single Input Impedance Patch_Antenna_ADKv1

patch antenna 110
100 90 80
70
Curve Info
1.00 S11
120 60 Setup1 : Sweep1
130 50
0.50 2.00
140 40

150 30

160 0.20 5.00 20

170 10

0.00 0.20 0.50 1.00 2.00 5.00

180 0.00 0

-170 -10

-160 -0.20 -5.00 -20

-150 -30

-140 -40
-0.50 -2.00
-130 -50
-120 -1.00 -60
-110 -70
-100 -90 -80

Peak Gain

For Single Patch Antenna: 6.436

For Array Antenna Design: 7.169
268 V. Midasala et al.

HFSSDesign1
Smith Chart 1
Curve Info
100 90 80
110 1.00 70 S(1,1)
120 60 Setup1 : Sweep
130 0.50 2.00 50
140 40
150 30

160 0.20 5.00 20

170 10

0.00
180 0.00 0.20 0.50 1.00 2.00 5.00 0

-170 -10

-160 -0.20 -5.00 -20

-150 -30
-140 -40
-0.50 -2.00
-130 -50
-120 -60
-110 -1.00 -70
-100 -90 -80

Fig. 6 Impedance for array antenna design

Fig. 7 3D polar plot for single patch antenna

Rectangular Patch Antenna Array Design at 13 GHz Frequency … 269

Fig. 8 3D polar plot for array antenna design

4 Conclusion

To improve the better gain of different antennas and also to nullify the side lobes we
can go for Phase array antennas. This paper explains about a Rectangular patch
antenna and also Rectangular Patch Antenna by Phased Arrays using HFSS 14.0.
For rectangular Microstrip Patch Antenna Design, this paper got a gain of 6.844 db,
for rectangular Microstrip Patch Antenna with Array Design got a gain of 8.5548
db. By using arrays this paper gets better gain. And also return loss, Impedance, 3D
polar plot, Directivity and Gain of the antenna are computed and corresponding
results are shown. The proposed antenna provides good performance in terms of
return loss and radiation pattern and gain.

Acknowledgments Extending our grateful thanks to the authorities of JNTUA, Acharya

Nagarjuna University and K L University for their support and encouragement to write this paper
and to use R&D Labs.
270 V. Midasala et al.

Author Biographies

Smt. Vasujadevi Midasala is presently pursuing her Ph.D degree

from JNTUA, Anantapuram in the area of Antennas and Commu-
nications. She received her degree in Electronics and communication
Engineering from S.V.V.S.N Engineering College, Acharya Nagarj-
una University, Guntur in 2008, M.Tech Degree from BSIT, JNT
University, Hyderabad in 2011. She is currently working as Assistant
Professor, Deptartment of ECE in K L University, Guntur, India.
She has 6 years of teaching experience. She has published ‘26’ papers
in national/international journals/conferences. His research interests
include Antennas, Signal Processing, Communications and VLSI
Design. He has guided ‘04’ M.Tech Projects. She is the member of
IEEE and ISTE. e-mail: [email protected]

Dr. P. Siddaiah obtained B.Tech degree in Electronics and

communication Engineering from JNTUA college of engineering in
1988. He received his M.Tech degree from SV University Tirupathi.
He did his Ph.D program in JNTU Hyderabad. He is the Chief
Investigator for several outstanding Projects sponsored by Defense
Organizations, AICTE, UGC and ISRO. He is currently working as
Professor and Principal, Department of ECE in University College of
Engineering and Technology, Acharya Nagarjuna University,
Guntur, India. He has taught a wide variety of courses for UG and
PG students and guided several projects. Several members success-
fully completed their Ph.D under his guidance. Several members
pursuing their Ph.D degree. He has published ‘76’ papers in National
and International Journals and Conferences. He is the life member of
FIETE, IE and MISTE.

Mr. S. NagaKishore Bhavanam is presently pursuing his Ph.D

degree from JNTUA, Anantapuram in the area of Signal Processing
and Communications. He obtained M. Tech Degree from Aurora’s
Technological and Research Institute, JNT University, Hyderabad in
2010. He obtained B.Tech degree in Electronics and communication
Engineering from S.V.V.S.N Engineering College, Acharya Nagarjuna
University, Guntur in 2008, He is currently working as Assistant
Professor, Department of ECE in University College of Engineering
and Technology, Acharya Nagarjuna University, Guntur, India. He
has 6 years of teaching experience. He has published ‘44’ papers in
national/international journals/conferences. His research interests
include Antennas, Signal Processing, Electro Magnetic Field Theory
and VLSI Design. He has guided around ‘06’ M.Tech Projects. He is
the student member of IEEE. e-mail: [email protected]
Automated Neural Network Based
Classification of HRV and ECG Signals
of Smokers: A Preliminary Study

Suraj Kumar Nayak, Ipsita Panda, Biswajeet Champaty,

Niraj Bagh, Kunal Pal and D.N. Tibarewala

Abstract Smoking of cigarettes has been reported to alter the cardiac electro-
physiology by modulating the autonomic nervous system. A preliminary investi-
gation of the heart rate variability (HRV) parameters suggested sympathetic
predominance in smokers. An in-depth analysis of the time domain and wavelet
processed ECG signals indicated that the automated neural networks (ANNs) were
able to classify the signals with an accuracy of ≥85 %. This suggested that smoking
not only modulates the functioning of the autonomic nervous system but is also
capable of modulating the cardiac conduction pathway.

Keywords Smokers

Heart rate variability
Autonomic nervous system

Automated neural network

1 Introduction

Electrocardiogram (ECG) is the electrical potential which is associated with the

functioning of the heart. The potential is initiated from the sino-atrial (SA) node.
The functioning of SA node is controlled by the Autonomic Nervous System
(ANS). ANS consists of two major subsystems, namely, parasympathetic and
sympathetic nervous systems. The subsystems of ANS control the pacing of the SA
node. ANS try to maintain the heart rate of a person by either increasing or
decreasing the activities of the parasympathetic and sympathetic nervous systems.
Due to this reason, there is a variation in the timing of the subsequent heart beats.

S.K. Nayak
I. Panda
B. Champaty
N. Bagh
K. Pal (&)

Department of Biotechnology and Medical Engineering,
NIT-Rourkela, Odisha 769008, India
e-mail: [email protected]
D.N. Tibarewala
School of Bioscience and Engineering,
Jadavpur University, Kolkata 700032, India

© Springer India 2015 271

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_25
272 S.K. Nayak et al.

An in-depth study of this variation is regarded as Heart Rate Variability (HRV). The
measure of the HRV parameters divulges information about the condition of the
ANS in a non-invasive manner [1].
The functioning of the ANS has been reported to be affected by training the
persons with regular exercises or by regular smoking. Regular exercise helps in
improving the health of the cardiovascular system by inducing changes of the
structural and the functional capability of the heart. The functional changes due to
the exercises are mainly attributed to the parasympathetic dominance [2]. On the
contrary, cigarette smoking has often been associated with cardiovascular diseases
(e.g. coronary heart disease, aortic aneurysm, sudden death and peripheral artery
disease). This can be explained by the reduced activity of the cardiac autonomic
function, which in turn, results in the cardiac vulnerability of the smokers [3]. Due
to the above reason, there is an increased risk of cardiac mortality in the smokers.
Keeping the above facts in mind, in this study, we have tried to understand the
cardiac activity of the smokers taking athletes as the control group. A conscious
effort was made for not selecting the sedentary group as the control because our
current study is based on the short-term HRV studies, where the differences in most
of the parameters of the cardiac activities of the smokers and the sedentary groups
might be statistically insignificant [4]. Further, the statistical parameters of the ECG
signals were calculated. The HRV and the ECG signal parameters were used for
classifying the smokers from the athletic group using ANN.

2 Methods

Forty volunteers were invited to participate in the study. Out of the forty volunteers,
twenty volunteers were the smokers who smoke at least ten cigarettes per day and
twenty volunteers were the sports personnel (member of various athletic teams of
NIT Rourkela) who practiced for at least 2 h/day. All the volunteers were the
students of NIT Rourkela and were within the age group of 21–27 years
(24.27 ± 1.82). The smokers were leading sedentary life. The invited athletes didn’t
smoke at all. All the volunteers were informed about the study and a written consent
form was obtained from the volunteers before the start of the study. The ECG of the
volunteers was recorded for 5, 90 min after dinner. The HRV parameters, time
domain parameters and wavelet processed parameters of the ECG signal were
calculated using the trial version of the software Biomedical Workbench (National
Instruments, USA) and tabulated in the statistica spreadsheet. The important
parameters were determined using t-test, classification and regression tree (CART),
boosted tree (BT) and random forest (RF) analysis. The important parameters were
used in various combinations as input for ANN based classification.
Automated Neural Network Based Classification … 273

3 Results and Discussions

Thirty five HRV features were obtained from the study. To determine the important
parameters, various statistical methods (t-test, CART, BT and RF) were employed.
The t-test is a linear importance predictor whereas CART, BT and RF are non-linear
classifiers for predicting important predictors.
A predictor importance of ≥95 % was considered acceptable. The analysis of the
HRV features using t-test suggested that VLF % (FFT), VLF Power (AR) and VLF
% (AR) were the important predictors. All the important parameters were higher in
smokers as compared to the control group (athletes). This can be explained by the
sympathetic dominance in the smokers [5].
The importance prediction using CART analysis showed that the important
predictor was SD1. SD1 has been reported to be a marker of short-term variability
of the heart rate [6]. Since the HRV parameters were calculated using 5 min of ECG
signal, the occurrence of SD1 as the major predictor is justified. The SD1 values
were higher in smokers as compared to the athletes. As per the reported literatures,
a higher value of SD1 is associated with the sympathetic dominance of the auto-
nomic control of the heart. The boosted tree classification showed that the VLF
Power (FFT) and the VLF % (AR) were the main important predictors. Both the
values were higher in smokers as compared to the athletes indicating a sympathetic
dominance in smokers [7]. Similar to the previous linear and non-linear classifi-
cation, Random Forest also suggested that the VLF Power (AR) was the major
important predictor indicating the sympathetic dominance in smokers (Table 1).
The above important parameters were used in various permutations and com-
binations as the input variables for automated neural network (ANN) based clas-
sification. Best classification efficiency was obtained using the features VLF power
(AR) and VLF % (FFT) with both MLP and RBF algorithms. The classification
efficiency using MLP algorithms was found to be 82.5 % (Table 2) whereas the
classification efficiency was found to be 95 % (Table 3) when RBF algorithm was
used. The architecture properties of the MLP and RBF networks have been tabu-
lated in Table 4.
Similar to the HRV parameters, the important time domain ECG signal
parameters were calculated using t-test, CART, BT and RF. t-test and BT suggested
that kurtosis and skewness were the important predictors, whereas CART and RF
suggested that only skewness was the important predictor during classification
(Table 5). Both the kurtosis and the skewness were found to be higher in the control
group. Since only two important predictors were obtained, both the predictors were
used as input for probable classification in ANN. The MLP algorithm showed a
classification efficiency of 77.5 % (Table 6) whereas the RBF algorithm has showed
a classification efficiency of 90 % (Table 7). The parameters of the network
architecture have been provided in Table 8.
The ECG signals were decomposed using db6 wavelet up to the level 8. The
reconstruction of the signal was achieved using d7 + d8 levels. The analysis of the
parameters using t-test didn’t show any important predictor. CART analysis showed
 274

Table 1 Important predictors from HRV features

Classifiers HRV features Mean A Mean SM SD A SD SM p Predictor importance
t-test VLF (%) (FFT) 7.995 14.125 10.47776 8.348433 0.047692 –
VLF Power (AR) 94.3575 252.451 123.4777 223.7172 0.008693 –
VLF (%) (AR) 6.768 13.2665 6.294338 6.537892 0.00555 –
CART SD1 (ms) 29.7365 36.828 22.0389 19.65057 – 1
BT VLF power (FFT) 177.4445 315.0765 459.3199 289.5487 – 0.96826
VLF (%) (AR) 6.768 13.2665 6.294338 6.537892 – 1
RF VLF power (AR) 94.3575 252.451 123.4777 223.7172 – 1
S.K. Nayak et al.
Automated Neural Network Based Classification … 275

Table 2 Confusion matrix of MLP network

Category-A Category-SM Total
Total 20 20 40
Correct 16 17 33
Incorrect 4 3 7
Correct (%) 80 85 82.5
Incorrect (%) 20 15 17.5

Table 3 Confusion matrix of RBF network

Category-A Category-SM Total
Total 20 20 40
Correct 18 20 38
Incorrect 2 0 2
Correct (%) 90 100 95
Incorrect (%) 10 0 5

Table 4 Parameters of the RBF and MLP networks

Networks Features used C.E. Algorithm E.F. Hidden act Output act
RBF VLF power and VLF % 95.00 RBFT CE Gaussian Softmax
2-25-2
MLP VLF power and VLF % 82.50 BFGS 40 CE Logistic Softmax
2-16-2
C.E. classification efficiency, E.F. error function

Table 5 Important predictors from time domain features

Classifiers Time domain Mean Mean SD A SD SM p Predictor
features A SM importance
t-test Kurtosis 5.074 3.819 1.879 1.437 0.022 –
Skewness 1.341 1.122 0.299 0.264 0.019 –
CART Skewness 1.341 1.122 0.299 0.264 – 1
BT Kurtosis 5.074 3.819 1.879 1.437 – 1
RF Skewness 1.341 1.122 0.299 0.264 – 1

Table 6 Confusion matrix of

MLP network Category-A Category-SM Total
Total 20 20 40
Correct 14 17 31
Incorrect 6 3 9
Correct (%) 70 85 77.5
Incorrect (%) 30 15 22.5
276 S.K. Nayak et al.

Table 7 Confusion matrix of RBF network

Category-A Category-SM Total
Total 20 20 40
Correct 17 19 36
Incorrect 3 1 4
Correct (%) 85 95 90
Incorrect (%) 15 5 10

Table 8 Parameters of the RBF and MLP networks

Networks Features used C.E Algorithm E.F Hidden act Output act
RBF 2-25-2 Kurtosis skewness 90 RBFT CE Gaussian Softmax
MLP 2-4-2 Kurtosis skewness 77.5 BFGS 5 CE tanh Softmax
C.E. classification efficiency, E.F. error function

Table 9 Important predictors from time domain wavelet features

Classifiers Wavelet Mean Mean SD A SD SM p Predictor
features A SM importance
CART AM 0.00 0.000 0.002 0.001 – 1
Skewness −0.291 3.184 12.140 6.127 – 1
BT RMS 0.124 0.111 0.043 0.040 – 0.95
Std. dev 0.124 0.111 0.043 0.040 – 0.95
Variance 0.017 0.014 0.011 0.009 – 0.95
Median 0.002 0.002 0.007 0.011 – 1
Mode 0.002 0.002 0.007 0.011 – 1
RF Kurtosis 3.743 3.329 0.814804 0.574 – 1

that the arithmetic mean (AM) and skewness were the important predictors. BT
analysis indicated that root mean square (RMS), standard deviation (Std. Dev.),
variance, median and mode were the important parameters. RF analysis indicated
that kurtosis was the only important predictor (Table 9). Like HRV and time
domain signal parameters, the important parameters of the wavelet processed sig-
nals were used in various combinations and permutations as input for ANN clas-
sification. The best classification efficiency using MLP algorithm was obtained
when AM and skewness were used as the input parameters. An efficiency of 72.5 %
was achieved (Table 10). On the contrary, an efficiency of 85 % was achieved when
RMS and standard deviation were used as the input parameters for the ANN
classification using RBF algorithm (Table 11). The architecture of the best networks
have been provided in Table 12.
As a matter of fact, a classification efficiency of ≥85 % has been considered as a
good efficiency while classifying ECG signals using automated neural networks [6].
During classification using HRV parameters a classification efficiency of 95 % was
Automated Neural Network Based Classification … 277

Table 10 Confusion matrix of MLP network

Category-A Category-SM Total
Total 20 20 40
Correct 14 15 29
Incorrect 6 5 11
Correct (%) 70 75 72.5
Incorrect (%) 30 25 27.5

Table 11 Confusion matrix of RBF network

Category-A Category-SM Total
Total 20 20 40
Correct 18 16 34
Incorrect 2 4 6
Correct (%) 90 80 85
Incorrect (%) 10 20 15

Table 12 Parameters of the RBF and MLP networks

Networks Features used C.E Algorithm E.F Hidden act Output act
RBF 2-16-2 RMS std. dev 85.00 RBFT CE Gaussian Softmax
MLP 2-30-2 AM skewness 72.50 BFGS 22 CE tanh Softmax
C.E. classification efficiency, E.F. error function

achieved using RBF algorithm. Similarly, classification efficiency of 90 and 85 %

were achieved when time domain and wavelet processed signals, respectively, were
used for classification using RBF algorithm. From the above results, it can be
observed that the classification efficiency of the HRV, time domain and wavelet
processed signals were ≥85 %, which is within the accepted limit of the recom-
mended classification efficiency for the ECG signals. In our study, we have seen
that the RBF algorithms showed better results compared to the MLP algorithms.
The observation may be explained by the architecture of the ANN networks. MLP
uses an inner product of the input and the weight vectors whereas RBF uses the
Euclidean distance between the input and the weight vectors [7]. The MLP algo-
rithm consists of multilayer hidden layers. Each layer is connected to the other
layers using a linear combination function. In other words, the connection between
each layer with the next layer is given by a linear mathematical equation. On the
contrary, RBF algorithms usually contain only one hidden layer and the inputs are
not weighted before connecting the hidden layers. Due to this reason, the devel-
opment of RBF model requires less time to obtain an optimum model parameter.
This can be explained by the fact that there is no need for repetition during the
training period [8]. Additionally, the RBF type ANNs have been reported to allow
278 S.K. Nayak et al.

universal approximations and learning without having any local minimum. This, in
turn, allows quick convergence of the model parameters to provide results. Because
of the above mentioned facts it has been reported that the RBF model provides
better accuracy in prediction than the MLP models.

4 Conclusion

In the current study, the effect of smoking on the cardiac activity of the smokers
was analyzed and compared with cardiac activities of athletes (control group). The
athletic group was taken as a control. This was done, because the athletic activities
have been reported to improve the cardiac activity. Hence, a pronounced difference
in the cardiac activity between the groups was expected. The HRV analysis of the
ECG signals suggested sympathetic predominance in the smokers as compared to
the athletes. Even though the sympathetic activity was found to be prominent from
the VLF % (FFT), VLF Power (AR), VLF % (AR), SD1 and VLF Power (FFT)
features, the differences in the heart rate of the smokers and athletes was not
statistically significant. This may be explained by the participation of a less number
of volunteers. Apart from the linear classifier (t-test), the important predictors from
the non-linear classifiers (CART, BT, RF) also suggested sympathetic dominance.
RBF ANN network was able to efficiently classify the smokers from the athletes
with a classification efficiency of 95 %. The ECG signals provide information about
the cardiac electrophysiology. In our study, we found that though the linear clas-
sifier was not able to classify the ECG signals (time domain and wavelet processed),
the non-linear classifiers were able to figure-out the important predictors. When the
important predictors were used in probable classification using ANN, a classifica-
tion efficiency of ≥85 % was achieved [8]. This indicated that the smoking of
cigarettes have a marked effect on the cardiac electrophysiology. The changes in the
cardiac electrophysiology can’t be detected using linear classifiers.

References

1. Champaty B et al (2013) Artificial intelligence based classification of menstrual phases in

amenorrheic young females from ECG signals. In: India conference (INDICON), 2013 annual
IEEE, pp 1–6
2. Saboul D et al (2013) The impact of breathing on HRV measurements: implications for the
longitudinal follow-up of athletes. Eur J Sport Sci 13:534–542
3. Harte CB et al (2013) Association between smoking and heart rate variability among
individuals with depression. Ann Behav Med 46:73–80
4. Pal K et al (2013) Heart rate variability and wavelet-based studies on ECG signals from smokers
and non-smokers. J Inst Eng (India) Ser B 94:275–283
5. Harte CB, Meston CM (2014) Effects of smoking cessation on heart rate variability among
long-term male smokers. Int J Behav Med 21:302–309
Automated Neural Network Based Classification … 279

6. Ciszewski P et al (2013) Lower preoperative fluctuation of heart rate variability is an

independent risk factor for postoperative atrial fibrillation in patients undergoing major
pulmonary resection. Interact Cardiovasc Thorac Surg 17:680–686
7. Saperova E, Dimitriev D (2014) Effects of smoking on heart rate variability in students (545.6).
FASEB J 28:545.6
8. Baxt WG (1990) Use of an artificial neural network for data analysis in clinical decision-
making: the diagnosis of acute coronary occlusion. Neural Comput 2:480–489
Reliable, Real-Time, Low Cost Cardiac
Health Monitoring System for Affordable
Patient Care

Meghamala Dutta, Sourav Dutta, Swati Sikdar, Deepneha Dutta,

Gayatri Sharma and Ashika Sharma

Abstract It is increasingly being made aware by various organizations including

the WHO (World Health Organization) that Cardiovascular Disease (CVD) is the
‘Silent Killer’ and is the perhaps the most important cause for death globally in the
present times. The prime reason being the fact that CVD often goes unnoticed at
the early stages and clinical intervention is soughted only during emergency. A
thorough understanding of the nature of the disease reveals the association of
physiological abnormalities and subdued physiological disorders that might have
raised due to stress, lifestyle related disorders, pathological conditions etc. Studies
have revealed that all these problems if addressed to at an early stage can be
controlled and the potential risks and hazards lessened to a considerable extent.
Early detection is possible in developed countries with expensive and advanced
machineries, which can keep the data recorded in a user-friendly device affordable
by the patient. But in a developing country like ours a low cost affordable device is
yet to be available to the rural population especially addressing the problem. The
aim of this study is to design an affordable and reliable Cardiac Health Monitoring
platform for identification, data storage and availability to the physician even after a
considerable time interval for diagnosis of the condition and post diagnosis con-
tinued therapy. The Cardiac Health Monitoring Systems (CHMS) was developed to
capture and pre-process physiological parameters (ECG, Heart Sound, Heart Rate in
Normal, Stressed and Physiologically abnormal conditions) in real-time and
transmit the same using wired as well as wireless communication technology. The
CHMS has been so designed is designed to capture the patient’s/user’s physio-
logical parameters in electrical form e.g. the ECG signal, apply preliminary adap-
tive filtering to shape the requisite signals and forward the processed data. This data

Presented at International Conference on Advancements of Medical Electronics ICAME 2015,

29–30th January 2015.

M. Dutta (&)
S. Sikdar
D. Dutta
G. Sharma
A. Sharma

Department of Biomedical Engineering, JIS College of Engineering, Kalyani, India
e-mail: [email protected]
S. Dutta (&)
IBM, New Delhi, India

© Springer India 2015 281

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_26
282 M. Dutta et al.

is acquired by Smart hand held mobile device, and the excess processing power of
the device is utilized for further processing as well as storage of medical records and
transmission of data to the hospital management system or for isolated diagnosis
and assessment. Critical readings crossing the set threshold (per WHO standards)
would set off an alarm [locally as well as over the mobile network] and thus alert
any imminent life-threatening situation.

Keywords ECG
Heart sound
CHMS
LPF
FFT
Smartphone

1 Introduction

It has been proved by scientific research and years of collected data that Heart rate,
ECG, Heart Sound signals can be considered as monitoring parameters related to
Cardiovascular Functioning. Timely monitoring and analysis can predict subdued
cardiovascular disorders [1]. The Gold standard for Heart Beat sets the normal
Heart Beat for adults at 60–100 beats/min and any deviation is considered to be
abnormal or irregular indicating subtle or pronounced malfunctioning of the heart,
the effects needless to say might be fatal [2]. Though till date the Electrocardiogram
(ECG) is the most widely adopted clinical tool to diagnose and assess the risk of
CVD, ECG is clinically effective when the disturbances of the Cardiovascular
system are reflected in the recording [1]. Often at an early stage of CVD, recordings
in an ECG are near normal and the physician fails to diagnose and treat CVD [1].
Hence other physiological signals like Heart Sounds and Heart Rate should be
seriously considered to be an important clinical marker for diagnosis and treatment
of CVD in addition to ECG [3]. Monitoring mechanisms, let alone the continuous
ones, are often absent in majorities of townships and rural areas in a developing
country, like ours and it becomes difficult and expensive for the patient to visit the
nearest hospital or diagnostic center for a sensation of increased heart beat
(Tachycardia), or feelings of breathlessness. This early warning mechanism by
which the body indicates its problems goes unnoticed or are neglected. There has
been increasing demand for an easy to use and low cost device suitable for the
patient as well as the physician, which would be able to forecast at an early stage
any future possibilities of CVD and related disease.
Studies have clearly demonstrated that cigarette smoking, physical inactivity,
and increased body mass index (BMI) are associated with increased risks of
premature cardiovascular disease (CVD) and death especially in the age group
35–60 years [4]. Though the urban population gets the benefit of state-of-the-art
hospitals with CVD monitoring devices the rural population especially in the
economically backward countries is ignorant about the devastating effects of CVD.
Hence a large population of men and women continue with subdued or untreated
CVD with fatal results. Monitoring of abnormal Heart Sound signals in real time is
an effective way to indicate the probable onset of CVD and CVD related deaths
Reliable, Real-Time, Low Cost Cardiac … 283

especially in this age group [3]. The pathogenesis of CVD is poorly understood.
The untimely occurring CVD leading to heart attacks and deaths can be largely
prevented by the timely diagnosis of CVD and its arrest with the available medi-
cines [5]. The widely used ECG studies is limited to detection of abnormalities
rising due to disturbances in the conduction of electrical impulse across the heart
musculature while Heart sound studies e.g. the Phonocardiograph or Echo car-
diograph deal with the mechanical defects of the heart. However, these diagnostic
tools have their own limitations, as they require skilled medical personnel for
interpretation of result, laboratory or hospital infrastructure for carrying out the tests
thus making the tests too expensive and are condition specific. Hence cases of mild
or subdued cardiac abnormalities often go unnoticed and results in under treatment.
The ECG signals along with the heart sound signals, being a direct expression of
the mechanical activity of the cardiovascular system, is potentially an unique source
of information for identifying significant events in the cardiac cycle and detecting
irregular heart activity [6]. A simple recording of ECG along with Heart Sound
signals from audible sounds at the heart apex, in real time can be a useful marker for
CVD, which would enable early diagnosis and treatment. Conversion of these
Biosignals into electrical form using a microphone can enable us to visualize,
interpret and analyze the sound information [7].
Considering the fact that these abnormalities are unpredictable and may occur at
anytime and anywhere especially with those with hypertension and high serum
cholesterol levels, a strong need has been felt for the development of a unique
device working on the Cardiac Health Monitoring Platform combining an unique
indigenously developed device and the mobile phone which will be non-invasive,
cost effective, simple to operate and can be used by anybody for prolonged home
monitoring application. Our approach was to couple existing technology with a
user-friendly device like the smart phone, which is within the affordable reach of
the large populace and at the same time has the extra capacity to process the signal
as well as store it for later reference and use. The prototype designed by us could be
used for recording and reading the both the ECG signals and Heart Sound signals in
real time and is aware of any abnormality with these parameters. If the intensity of
Heart Sound increased to very high or dropped to very low, a signal would be given
suggesting urgent attention required. Similarly for the ECG signals,which is of
single electrode as well as multi-electrode type, the device can filter out the noise
and accurately show the signal on the smart phone’s screen.

2 Previous Work

Earlier works attempted to develop a mobile Biosignal recording device for signals
originating from the Heart and adjoining areas. Prior research activities were mainly
focused on developing:
284 M. Dutta et al.

A. Electronic ECG Sensing Device [8]

B. Recording and data storage of the ECG signals in the mobile [7]
Most of the mobile ECG research used either a PDA or a cell phone only as a
data display or a transmission device, which sends the collected ECG data to a
remote PC for later review and diagnosis by a physician [7, 9]. But this prototype
has the capability to apply advanced signal processing techniques to filter out the
noises (which is very common while acquiring heart sound) and electrical inter-
ferences in the smart phone itself. Although many devices are being worked upon
on similar areas, no single device can be considered as the ultimate one; each has its
own deficiencies and limitation.
Our work was mainly focused on developing an indigenously made low cost,
reliable portable ECG system capable of transmitting signals via wired as well as
wireless mode to the users’ regular smart phone. The device has the capability of
acquiring signals through various input sources—single electrode, multiple elec-
trode and through microphone. The unused and unutilized storage space in the
mobile phone would collect, record, display, and transmit the signals originating
from the heart data in real time, but is also analyze the acquired Signal data, match
them with existing available data and predict in advance possible CVD conditions.

3 Procedure

The real challenge faced by the team was keeping the signals noise free and noise
filtering. Interference with the ambient sound and sound arising due to other
physiological activities like blood flow, surrounding electrical signals arising due to
intrinsic muscle activities. The signals were picked by the microphone needed to be
filtered and processed by using several filtering techniques to isolate the meaningful
electrical and acoustic signals generated due to the functioning of the heart and its
surrounding tissues. Similarly for the single electrode and multiple electrode ECG
signal acquisition, commercial adhesive conductor strips as well as clamps can be
used to get commercial grade signal.
Our indigenous CHMS combines the simplicity of acoustic stethoscope with
advanced electronics and information technology to facilitate better performance,
recording of the heart and lung sounds and analysis of the recorded signals. The
same CHMS has the capability of acquiring ECG signals through various leads,
which need to connected appropriately. It also sends the data to a computer either
by wireless or with a wired connection. The primary limitation of the analog or the
conventional stethoscope is that it is unable to capture several physiological sound
signals as they are below the threshold of hearing. This problem can be easily
overcome using the microphone aided digital stethoscope although the amplified
sound signal of interest contains enough noise, which interferes with the signal
generation process [10].
Digital stethoscopes available are intended for observation and recording of
heart sounds and murmurs as well as lung and airway sounds. There are other types
Reliable, Real-Time, Low Cost Cardiac … 285

of digital stethoscopes, which can be used, for monitoring heart and lung sounds
during anesthesia and some even combine with synchronous ECG monitoring. But
the CHMS was even more cost effective with same functional capability as that of
the Digital Stethoscope. The CHMS is a good tool for medical education also. Heart
sounds recorded with CHMS can be visually observed and listened to [10]. These
recording can be used to build multimedia tools to improve the quality of physical
exam and education. By providing connectivity solutions either wired or wireless, a
digital stethoscope can be used in telemedicine applications facilitating remote
diagnosis by specialists [11]. The recorded signals can further be spectrally ana-
lyzed and used for automated cardiac auscultation and interpretation. Our CHMS
coupled to a portable computer/tablet with suitable software and connected to
internet for automated or remote diagnosis by a specialist, will be the future trend.

4 Results

4.1 Design of the Prototype

In order to acquire real time Heart Sound signals, it is first essential to filter the
associated noise due environment and other physiologic activities like talking,
breathing, etc. which might be mixed with the heart sound. The design of the
prototype was done in two phases.

Phase 1: The Hardware system of CHMS with Projected Mobile Application

(Fig. 1)

The Cardiac Health Monitoring System ( CHMS)

Signal/Data
Acquisition

Acquiring audio
signal from the
prototype “digital”
stethoscope
Transmission to
Mobile Device
Conditioning
Signal

Signal Filtering Conversion of the Feeding the data Amplifier and

and

and removal of conditioned signal into the bluetooth Analog output to

noises by A/D Converter transceiver standard headset
Interpretation Application and

Historical Data
Storage of
Mobile

Executing the
Using the mobile
application – User Acquisition of the Storage of data
application to run
data entry for first digital signal datewise
FFT
time use
and guidance to
Automatic Data

Physician

Automatic Alarm
Mapping the data Indication to visit
Graphical Automatic Analysis Configuration for
with the “Gold” the Physician
representation on and report repeat sample
standard of incase of deviation
the mobile screen generation after select
heartbeat from standards
duration

Fig. 1 The process for the heart sound capture

286 M. Dutta et al.

Phase 2: Further Work involves signal processing and signal analysis using DSP
techniques and FFT for conversion of Electrical and Acoustic signals
into the frequency domain for further analysis.

4.2 Prototype of the CHMS Used to Capture the ECG Signals

in Real Time

The CHMS circuit uses a very sensitive common off-the-shelf Quad Op–Amp in a
Low Pass Filter [LPF] setup with differential voltage amplifier configured for
human interfacing with high impedance and high-noise environment (Fig. 2).

4.3 Frequency Response Analysis and Development

of Algorithm of the Signal Generated by CHMS

In this paper we would concentrate on acquisition of the various signals (ECG,

Heart Sound) through the custom designed front end analog hardware (Fig. 3) based
on the concept that appropriate digital and analog filter combination should allow
selective hearing of heart sounds (low frequency) and ECG signals in the frequency
spectrum (Fig. 4) and the development of the algorithm of the various signal
processing portions for removal of certain types of noises using the digital signal
processing techniques.
The frequency spectrum illustrated in Fig. 4 shows that the Frequency Response
of the normal heart beat sound signal captured by the CHMS. The frequency
content varying from around 50–250 Hz. Figure 4 shows that the basic frequency

Fig. 2 Prototype for Cardiac

Health Monitoring System
Reliable, Real-Time, Low Cost Cardiac … 287

Fig. 3 Acquired ECG

Signals on our CHMS

Fig. 4 Frequency response

components are obviously detected by the Fourier transform but not the time delay
between these components. Possibility lies that the interference of surrounding
noise as a result of other physiological activities may be one of the factors. Hence
removal of this noise using adaptive filtering and low pass filters prior to recording
of the said sound signal becomes very important. The two components (due to the
closure of the aortic valve) and (due to the closure of the pulmonary valve) are
obvious in Fig. 3. However, the FFT analysis cannot tell which of two, precedes the
other or the value of the time delay between them. These parameters (position and
time delay) are very important to detect some pathological cases.

5 Discussion

The impact of capturing and analyzing cardiac signals is realized when it is inte-
grated within a clinical information system. Probabilistic modeling is a requirement
to predict automated patient diagnosis and always requires accompanying clinical
history to optimize discrimination and calibration. The device stores the data in a
particular format and can be seamlessly uploaded to the hospital’s database using
the internet. If the mobile device does not have the connectivity for transmitted data
288 M. Dutta et al.

over the network, it can be paired with the hospital Bluetooth hotspot for easy
transmission. This requires customization for each hospital care system. The crucial
component in the success of these techniques is to get the universal model in place
and enable in providing sufficient signal quality to perform an accurate analysis of
the data.
It may be noted that while using various mobile devices to record heart signals
and sounds, we observed a high variance in quality between hardware, with some
units being completely unable to record useful data because of the low frequency
response characteristics of the Bluetooth audio section. This necessitated the
incorporation of the converting the analogue signal into a digitally sampled one
using an A/D converter. (This however had a big impact in the total cost of
ownership.) While studying the characteristics of some readily available branded
phones, we found that the Apple’s devices had very good low frequency response,
but the device itself was on the higher end of the cost spectrum with the ECG
attachment costing more than the smart phone itself.
An important factor in using digital auscultation techniques is the problem of
ambient noise and movements. Efforts have been made to remove such additional
signals from the actual heart sound for accurate analysis [10, 12]. There has been
use of LECG [12] for the use of adaptive filtering but has the problem of proper
placement in the chest, failure of which will result in the actual heart sound getting
cancelled out. The band pass filter is helpful to a certain extent after which the
differentiation of heart, lung and ambient noise becomes quite difficult for auto-
mated study and analysis. Since the noise overlaps in the time and frequency
domain, filtering is extremely difficult, [7] used a second off-body microphone to
record ambient noise to provide information for an adaptive filter.
Finally, we once again point out the fact that it is possible to reuse the existing
technology and hardware in countries where health care is expensive as well as not
easily accessible to provide low-cost reliable diagnostics. We also note that these
standalone systems would form a complimentary diagnostic framework that can be
leveraged by the higher health care mechanisms in providing timely and cost
effective treatment. We also note that the diagnostic capability of the stand-alone
advanced mobile device is not confined to heart rate analysis, but can be extended
to HRV, heart valve issues, lung function, infection, sleep structure and even
depression using various transducers.

6 Conclusion

There have been many attempts to bring advanced health monitoring devices at
affordable cost for the rural and masses below the poverty line. With the
advancement in microprocessor technologies, more and more computing power is
being provided at a comparatively lesser-cost month on month. This has led to
splurge in sophisticated mobile devices, which have become ubiquitous. The digital
auscultation procedure has been there for some while but has its own challenges of
Reliable, Real-Time, Low Cost Cardiac … 289

cost, accuracy etc. We have tried to address the existing problems and develop a
low-cost solution for the mass. So far the results have been satisfactory under
controlled conditions. Further work would include removal of extraneous noise,
optimum placement of electrical activity probe in the chest, improvement in getting
frequency response below 100 Hz and packaging of the device for ease of handling
and use.

Acknowledgments This work has been funded by AICTE under AICTE Research Proposal
Scheme.

References

1. Goldberger AL, Goldberger E (2006) Clinical electrocardiography: a simplified approach, 7th

edn. Elsevier/Mosby Inc, St Louis
2. American Heart Association (2008) Heart disease and stroke statistics-2008 update. AHA,
Dallas, Texas
3. WHO (2007) Prevention of cardiovascular disease. Guidelines for assessment and
management of cardiovascular risk. Geneva 47(12):1555–1569
4. Amin DSM, Fethi B-R (2008) Features for heartbeat sound signal normal and pathological.
Recent Pat Comput Sci 1:1–8
5. Mackay J, Mensah G (2004) Atlas of heart disease and stroke. World Health Organization,
Geneva
6. Mittra AK, Choudhari NK (2009) Development of a low cost fetal heart sound monitoring
system for home care application. JBiSE 2(6):380–389
7. He B, Cohen RJ (1992) Body surface laplacian ECG mapping. IEEE Trans Biomed Eng 39
(11):1179–1191
8. Oresko J, Cheng J, Jin Z, Duschl H, Cheng A (2009) Detecting cardiovascular diseases via
real-time electrocardiogram processing on a smart phone. In: Proceedings of the workshop on
biomedicine in computing: systems, architectures, and circuits (bic), June 2009, Austin, TX,
pp 13–16
9. Bhatikar SR, DeGroff C, Mahajan RL (2005) A classifier based on the artificial neural network
approach for cardiologic auscultation in pediatrics. Artif Intell Med 33(3):251–260
10. Zhang GQ, Zhang W (2009) Ageing Res Rev 8(1):52–60. doi:10.1016/j.artmed.2004.07.00
11. Fox K et al (2007) Resting heart rate in cardiovascular disease. J Am Coll Cardiol 50:823–830
12. Tan A, Masek M (2009) TT fetal heart rate and activity monitoring via smart phones. health
summit 2009. Washington DC. http://research.microsoft.com/en-us/events/mhealth2009/tan-
masek-presentation.pdf. Accessed 29–30 Oct 2009
 Part IV
Embedded Systems and Its Applications
in Healthcare
An Ultra-Wideband Microstrip Antenna
with Dual Band-Filtering for Biomedical
Applications

Subhashis Bhattacharyya, Amrita Bhattacharya

and Indranath Sarkar

Abstract Anultra-wideband (UWB) microstrip antenna with dual band filtering to

prevent interference from coexisting WLAN and downlink of X-band satellite
communication system is proposed for use in biomedical applications. Dual notch
bands are achieved by etching two nearly half wavelength inverted U shaped slots
on the radiating patch. Impedance bandwidth has been improved by asymmetric
optimization of patch width with respect to the feed line. The antenna achieves
−10 dB return loss bandwidth of about 10.45 GHz (2.35–12.796 GHz) while
rejecting 4.9–5.98 and 7.15–7.99 GHz bands. The antenna has wide bandwidth,
similar and stable radiation performance at the passband frequencies.

Keywords Half wavelength slot
Microstrip antenna
Notchband
Ultra-wide-

band (UWB)

1 Introduction

The approval and allocation of the frequency band 3.1–10.6 GHz for ultra-wide-
band (UWB) systems by Federal Communications Commission (FCC) [1], has
motivated researchers to design UWB antennas for short-range high data rate
wireless communication, high accuracy radar and imaging systems with low power
consumption. UWB indoor wireless communication devices can be utilized in
modern electronic-healthcare system and Wireless Body Area Network (WBAN).

S. Bhattacharyya (&)
A. Bhattacharya
I. Sarkar

Department of ECE, JIS College of Engineering (Autonomous Institution Under WBUT),
P III, A5, Kalyani, Nadia 741235, West Bengal, India
e-mail: [email protected]
A. Bhattacharya
e-mail: [email protected]
I. Sarkar
e-mail: [email protected]

© Springer India 2015 293

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_27
294 S. Bhattacharyya et al.

The printed planar monopole antennas [2, 3] have attractive features like wide
bandwidth, compact size, simple structure, low cost, ease of fabrication and
omnidirectional radiation pattern [4]. Owing to coexistence of the UWB system
with other wireless standards such as the WLAN systems (5.1–5.825 GHz) and
downlink of X-band satellite communication system (7.25–7.75 GHz), UWB
antennas with filtering property are required. To reject the two above mentioned
bands, two band stop filters are necessary which will increase the cost and com-
plexity of the system. A simpler way to solve this problem is to design an UWB
antenna with double band-rejection characteristics. UWB antennas with band-not-
ched function have been reported, mostly with one notched band [5–7] for WLAN
(5.15–5.825 GHz). Recently several antennas with dual notched bands [8–10] were
presented. Etching slots on the patch or on the ground plane and adding parasitic
elements are two widely used methods to obtain notched bands [11–13].
In this paper, a single layer microstripline-fed monopole UWB antenna with dual
notched bands for the rejection of WLAN and downlink of X-band satellite com-
munication systems is proposed. Dual notched bands have been achieved by
introducing two inverted U shaped slots of different lengths on the radiating patch.
Parametric study has been carried out through simulation and analyzed thoroughly.
Simulated return loss (S11) characteristics, surface current distribution and radiation
patterns have been illustrated and discussed. Surface current distributions analyze
the creation of notches.

2 Antenna Design

Figure 1a shows the configuration of the proposed antenna. Dimensions of the

antenna have been optimized using IE3D software [13] to achieve the desired result.
This antenna with dimensions of 40 × 34 × 1.6 mm3, has been fabricated on a PTFE
substrate with relative permittivity (ϵr) of 2.4. A rectangular beveled patch has been
created on the front side of the substrate while a ground plane with a rectangle slot
of dimension 30 × 16 mm2 etched on it has been designed on the back side of the
substrate. The width of the microstrip feed line is kept uniform of about 3.9 mm to
achieve 50-Ω characteristic impedance. The bevelled patch is asymmetrical with
respect to the microstrip feed line. Two bevelling angles have been created with
α1 = 10.56° and α2 = 26.19° on either side of the patch respectively.
The design of the slots on the patch involves guided wavelength λg = (λ0/√ϵeff)
[14] where ‘λ0’ given as λ0 = c/fc is the free space wavelength corresponding to the
notch centre frequency fc, and where c is the speed of electromagnetic wave in free
space. ϵeff = (ϵr + 1)/2 [14] is the effective dielectric constant averaged over two
mediums, the substrate and air. The longer inverted U slot (slot-1) has an optimized
length of 21.5 mm and uniform slot width of about 0.625 mm. This length of slot-1
is about 0.51λg with respect to the simulated notch centre frequency of 5.44 GHz.
The total length of the shorter inverted U slot (Slot-2) is 15.5 mm with slot width
t = 0.25 mm. The length of this slot-2 is about 0.51 λg at the simulated notch centre
An Ultra-wideband Microstrip Antenna … 295

Fig. 1 a Configuration of the proposed antenna. b Side view

frequency of 7.6 GHz approximately. Therefore both the slots become resonant at
those frequencies where their length is approximately 0.51 of the guided wave-
length. Slot-2 is etched 0.375 mm below slot-1.
The optimized parameters of the antenna are given as follows:
W = 40 mm, L = 34 mm, Wp = 15 mm, Lp = 8 mm, g = 1.5 mm, d1 = 8.05 mm,
d2 = 3.05 mm, α1 = 10.56°, α2 = 26.19°, Lf = 10 mm, Wf = 3.9 mm, Ls = 12 mm,
Ws = 4.75 mm, t = 0.625 mm, Ls1 = 10 mm, Ws1 = 2.75 mm, t1 = 0.25 mm,
Wgs = 30 mm, Lgs = 16 mm, Wge = 18.05 mm, We = 5 mm, g1 = 1 mm, g2 = 2 mm,
g3 = 0.375 mm, g4 = 10 mm, g5 = 5 mm, g6 = 3 mm.

3 Results and Discussion

3.1 Parametric Study

Several antenna parameters are varied with a range of values as part of the opti-
mization process.
Figure 2 shows that when the patch width increases asymmetrically due to d1
increase, the return loss improves throughout the UWB range and resonances get
prominence at lower frequencies also.
296 S. Bhattacharyya et al.

Fig. 2 Simulated S11 of the

antenna for different values of
d1

When only one slot is etched on the patch with total slot length of 21.5 mm and
slot width of 0.5 mm then only one notched band is achieved from 5.012 to
6.104 GHz with notch band centre frequency at 5.56 GHz. When another slot of
length 15.5 mm and width 0.25 mm is etched on the patch in addition to the
comparatively longer slot of length 21.5 mm as stated previously in this paragraph
and width changed to 0.625 mm, a second notched band is achieved from 7.15 to
7.99 GHz with notch band centre frequency of 7.57 GHz along with the first
notched band ranging from 4.9 to 5.98 GHz with centre frequency 5.44 GHz as
inferred from Fig. 3. So it can be concluded that etching of slots plays a vital role in
generation of notched bands. Each slot creates a notch with a centre frequency
which is related to the length of the respective slot.
Figure 4 shows the effect of variation of Ws1 on the second notch band centre
frequency and its notch bandwidth. As slot-2 dimension (Ws1) is increased, the total
slot length increases which implicitly corresponds to higher guided wavelength and
thus a corresponding lower notch centre frequency. From Table 1 it is observed that
with increasing Ws1, the notch centre frequency attains lower values with gradually

Fig. 3 Simulated S11 of the

antenna with and without slot
on patch
An Ultra-wideband Microstrip Antenna … 297

Fig. 4 Simulated S11 of the

antenna for various Ws1

Table 1 Effect of Ws1 on the

notch-2 characteristics Ws1 (mm) Notch-2 bandwidth Notch-2 centre
(%) frequency
(GHz)
2.25 8.85 7.91
2.75 11.096 7.57
3.25 12.81 7.18

increasing notch bandwidth. So it can be concluded that the notch band centre
frequency is controllable by varying the length of the slot.
Figure 5 shows the effect of variation of the vertical gap g3 between slot-1 and
slot-2. As the slot-2 position is moved on the patch vertically away from or towards
slot-1, the gap g3 varies. The details of this variation are listed in Table 2. It has
been observed that if any slot is shifted closer to the feed line (closer to the bottom

Fig. 5 Simulated S11 of the

antenna for different g3
298 S. Bhattacharyya et al.

Table 2 Effect of G3
g3 (mm) Notch-2 Notch-2 centre
bandwidth (%) frequency (GHz)
0.875 20.12 7.405
0.625 16.4 7.375
0.375 11.096 7.570

Fig. 6 Simulated S11 of the

antenna for different slot-2
width t1

edge of the patch), notch bandwidth, created by that slot, increases as well as the
notch peak level (in terms of Return Loss or S11 in dB) increases along with slight
decrease in notch centre frequency.
Figure 6 shows the effect of different widths of slot-2 on the S11 characteristics
of the antenna. With the increase in the slot-2 width ‘t1’, the second notch band-
width increases gradually as displayed in Table 3. Also the notch band centre
frequency increases by small amounts. So it is an important generic conclusion that
the notch bandwidth can be controlled by varying the slot width.
When slot-1 width ‘t’ is varied then the first notch higher cut off is changed as
seen from Fig. 7. For t = 0.525 mm the 1st notch ranges from 4.87 to 5.8 GHz and
for t = 0.625 it ranges from 4.9 to 5.98 GHz. So it again validates the control of
notch bandwidth by slot width.

Table 3 Effect of T1
t1 (mm) Notch-2 Notch-2 centre
bandwidth (%) frequency (GHz)
0.25 11.096 7.570
0.5 13.26 7.765
0.75 15.19 7.965
An Ultra-wideband Microstrip Antenna … 299

Fig. 7 Simulated S11 of the

antenna for different slot-1
width t

From the above slot width variations it has been observed that the higher cut off
frequency is affected more than the lower cut off value of the notched bands.

3.2 Simulation Results

The proposed antenna has been studied using IE3D software.

3.2.1 S11 Characteristics

The simulated impedance bandwidth of the antenna ranges from 2.35 GHz to about
12.79 GHz with dual notched bands, one ranging from 4.9 GHz to about 5.98 GHz
there by rejecting the WLAN band (5.15–5.825 GHz) completely and another
ranging from 7.15 to 7.99 GHz thereby completely rejecting the 7.25–7.75 GHz
band reserved for downlink of X-band satellite communication systems. This makes
it usable for WBAN applications within the UWB range and also making it
interference free from the two rejection bands. The details of the passbands are
provided in Table 4. It is observed from Fig. 8 that there are five prominent
resonances within the S11 < −10 dB range. A sharp resonance at 2.6 GHz (−40 dB)
and others at 3.85 GHz (−27.62 dB), 6.75 GHz (−12.52 dB), 8.82 GHz (−17.6 dB)
and 11.34 GHz (−21.8 dB) are observed.

3.2.2 Surface Current Distributions

The excited surface current distributions, simulated using Zeland’s IE3D simulation
software [13], for the proposed antenna at five different frequencies are presented in
300 S. Bhattacharyya et al.

Table 4 S11 performance of

the proposed antenna Lower cut Higher cut % Centre
off (GHz) off (GHz) Bandwidth frequency
(GHz)
2.35 4.88 70 3.6
5.99 7.14 17.517 6.565
8 12.796 46.12 10.4

Fig. 8 Simulated S11

performance of the proposed
dual band-notched UWB
antenna

Fig. 9. At the passband frequencies of 2.6, 6.76, 8.8 GHz the distribution of surface
current is uniform over the patch as well as the ground as shown in Fig. 9a, c and e
respectively. The current is spread over a greater area in the ground planet 2.6 GHz
compared to other passband frequencies, as observed from Fig. 9a which may
attribute to better radiation. So it may be the cause of the sharp resonance of
−40 dB, achieved at 2.6 GHz.
However when the antenna is operating at 5.44 and 7.6 GHz, which are the
centre frequencies of the two notched bands, respectively, it is observed that
stronger current is concentrated around the edges of the longer inverted U slot (slot-
1) only at the vicinity of 5.44 GHz as shown in Fig. 9b whereas stronger current
forms around the edges of the shorter inverted U slot (slot-2) only at the vicinity of
7.6 GHz as shown in Fig. 9d. This leads to impedance mismatch between the feed
line and the patch at these notch frequencies and thus radiation from the antenna is
much less in the two notched band regions. The current around the edges of these
slots is oppositely directed to the current emerging from the microstrip feed line at
each notch band. So the radiation fields generated by them are neutralised due to
destructive interference taking place. This leads to the desired high attenuation
around the notch centre frequency. So it can be derived that the longer etched slot is
responsible for creation of the notched band with lower notch centre frequency of
5.44 GHz and the shorter slot is responsible for the notched band with higher notch
centre frequency of 7.6 GHz.
An Ultra-wideband Microstrip Antenna … 301

Fig. 9 Simulated surface

current distributions of the
proposed antenna at
a 2.6 GHz, b 5.44 GHz,
c 6.76 GHz, d 7.6 GHz,
e 8.8 GHz

3.2.3 Radiation Pattern

The simulated radiation patterns in the elevation direction at different frequencies

are illustrated in Fig. 10. For Φ = 0°, EΦ represents the co polarization (co-pol) and
Eθ represents the cross polarization (x-pol) whereas for Φ = 90°, Eθ represents co
polarization and EΦ represents cross polarization.
At 2.6 GHz pass band frequency, the co-pol is dominating the cross-pol highly
in both Φ = 0° and Φ = 90° as shown in Fig. 10a, b respectively. We see similar
scenario at 8.8 GHz observed from the Fig. 10e, f. At the notch band centre
frequencies, the co-pol levels are usually lower than the x-pol levels. The patterns
for 5.44 GHz notch band frequency is illustrated in Fig. 10c, d.
The simulated patterns at the pass band frequencies are similar which is expected
from a wideband antenna. So it can be a general conclusion that in the notch band,
useful radiation that is the co-pol is less compared to the co-pols at all other pass
302 S. Bhattacharyya et al.

Fig. 10 Simulated radiation patterns of the proposed antenna in the elevation direction at
a Φ = 0°, b Φ = 90° at 2.6 GHz; c Φ = 0°, d Φ = 90° at 5.44 GHz; e Φ = 0°, f Φ = 90° at 8.8 GHz
An Ultra-wideband Microstrip Antenna … 303

band frequencies. This may provide tolerable link loss at the pass band frequencies
for a satisfactory on-body antenna required for biomedical applications.

3.3 Applications

3.3.1 Bio-medical Imaging Systems

In the UWB microwave imaging systems, a very narrow pulse is transmitted from a
UWB antenna to penetrate the body. As the pulse propagates through the various
tissues, reflections and scattering occur at the interfaces. A particular interest is in
the scattered signal from a small size denser tissue representing a tumour or other
abnormalities. The reflected and scattered signals can be received using an UWB
antenna, or array of antennas, and used to map different layers of the body.
Microwave Imaging is used in breast cancer detection which is a common diagnosis
in women. A typical detection system is illustrated in Fig. 11a. Normal breast tissue
is largely transparent to microwave radiation, whereas malignant tissues, which
contain more water and blood, cause microwave signal backscattering. This scat-
tered signal can be picked by an array of microwave antennas and analysed using a
computer [15]. Imaging result on an artificial breast model is shown in Fig. 11b.

3.3.2 Microwave Hyperthermia

Hyperthermia means repeated heating of the tumour to just over 40 °C. This
treatment is toxic for the tumour itself and has shown doubled cure rates when
combining hyperthermia with traditional cancer treatment. The microwave tech-
nique enables heat treatment of tumors which are deep-seated and/or relatively hard
to access in the body. The microwaves are transmitted from a number of antennas

Fig. 11 a Configuration of the breast cancer detection UWB radar system, b imaging result on an
artificial breast model
304 S. Bhattacharyya et al.

enclosing the relevant body part. The heat effect is developed in the tumor by the
transmission of microwaves which are adjusted in time, frequency and strength in
order to work together to form a focus in the desired location. This requires high
precision of the system, in order for the heating to be concentrated on just the
tumor, without heating surrounding healthy tissues.
Figure 12 shows scan reports of a 24-year-old man with grade III recurrent
glioma at three various stages as (A, B), (C, D), (E, F). In the hyperthermia group,
primary cases received hyperthermia treatment, and patients with recurrent tumors
were treated with hyperthermia in combination with radiotherapy and chemother-
apy. Electrodes were inserted into the tumor with the aid of a CT-guided stereo-
tactic apparatus and heat was applied for 1 h. During 3 months after hyperthermia,
patients were evaluated with head CT or MRI every month. Gliomas in the
hyperthermia group exhibited growth retardation or growth termination. Necrosis
was evident in 80 % of the heated tumor tissue and there was a decrease in tumour
diameter [16].

Fig. 12 a, b Brain tumor

showing an irregular
enhancing signal shadow
(arrows) on CT before
treatment; c, d CT scan
showing internal liquefaction
and tumor necrosis (arrows)
in low-density images at
7 days after hyperthermia
(hyperthermia + radiotherapy
+ chemotherapy); MRI scan
showing increased tumor
necrosis (arrows) and
surrounding tissue with no
obvious enhancement at
12 months after hyperthermia
An Ultra-wideband Microstrip Antenna … 305

Fig. 13 A modern medical

monitoring and detection
system

3.3.3 Medical Telemetry

Miniaturized sensors can be worn on the body or implanted inside the body or kept
at a distance to monitor a person’s physiological state continuously in his free living
conditions.
UWB telemetry systems are suitable for high data rate transmission such as
wireless endoscope and multi-channel continuous biological signal monitoring such
as Electroencephalography (EEG), Electrocardiogram (ECG) and Electromyogra-
phy (EMG). A telemetry system is shown in Fig. 13. Their benefits are low power
UWB transmitter to increase battery life, high data rate to increase the resolution
and performance, and less interference from other wireless systems. Generally a
UWB receiver consumes more power than narrow band systems. So an off shelf
receiver is placed at 0.5–10 m away to detect the transmitted signal from body. For
a UWB transmitter the FCC regulation requires the signal output to be −41 dBm/Hz
and lower [17]. However the power levels should not reach above the regulated in-
body tissue absorption levels and it is better to use microstrip antennas as external
nodes which are kept away from the body. This reduces long term exposure to
radiation.

4 Conclusions

An Ultra-wideband microstrip antenna with filtering property has been investigated

for use in important biomedical applications like microwave imaging, breast cancer
diagnosis, microwave hyperthermia, telemetry and many WBAN applications with
little or no interference from WLAN and downlink of X band satellite communi-
cation systems. Dual band notch characteristic has been achieved by embedding two
inverted U shaped slots on the patch. The designed antenna exhibits broad band-
width, good and similar radiation performance at Yucethe pass band frequencies.
306 S. Bhattacharyya et al.

References

1. Federal Communications Commission (2002) First report and order in the matter of revision of
Part 15 of the commission’s rules regarding ultra-wideband transmission systems. Federal
Communications Commission, ET-Docket, Washington, pp 98–153
2. Lee E, Hall PS, Gardner P (1999) Compact wideband planar monopole antenna. Electron Lett
35(25):2157–2158
3. Liang J, Chiau CC, Chen X, and Parini CG (2004) Printed circular disc monopole antenna for
ultra-wideband applications. Electron Lett 40(20):1246–1247
4. Chen ZN (2007) UWB antennas: from hype, promise to reality in IEEE Antennas Propagation
Conference. pp 19–22
5. Qu X, Zhong SS, and Wang W (2006) Study of the band-notch function for a UWB circular
disc monopole antenna. Microw Opt Technol Lett 48(8):1677–1670
6. Cho YJ, Kim KH, Choi DH, Lee SS, Park SO (2006) A miniature UWB planar monopole
antenna with 5-GHz band-rejection filter and the time-domain characteristics. IEEE Trans
Antennas Propag 54(5):1453–1460
7. Hong CY, Ling CW, Tarn IY, Chung SJ (2007) Design of a planar ultra wideband antenna
with a new band-notch structure. IEEE Trans Antennas Propag 55(12):3391–3397
8. Chu QX, Yang YY (2008) 3.5/5.5 GHz dual band-notch ultra wideband antenna. Electron Lett
44(3):172–174
9. Yin K, Xu JP (2008) Compact ultra wideband antenna with dual bandstop characteristic.
Electron Lett 44(7):453–454
10. Lee WS, Kim DZ, Kim KJ, Yu JW (2006) Wideband planar monopole antennas with dual
band-notched characteristics. IEEE Trans Microw Theory Tech 54(6):2800–2806
11. Chung Kyungho, Kim Jaemoung, Choi Jaehoon (2005) Wideband microstrip-fed monopole
antenna having frequency band-notch function. IEEE Microwave Wirel Compon Lett 15
(11):766–768
12. Trang ND, Lee DH, Park HC (2011) Design and analysis of compact printed triple band-
notched UWB antenna. IEEE Antennas and Wirel Propag Lett 10:403–406
13. Kelly James R, Hall Peter S, Gardner Peter (2011) Band-Notched UWB antenna incorporating
a microstrip open-loop resonator. IEEE Trans Antennas Propag 59(8):3045–3048
14. Zeland Software Inc. IE3D: MoM-Based EM Simulator. http://www.zeland.com/
15. Cohn SB (1969) Slot line on a dielectric substrate. IEEE Trans Microw Theory Tech 17
(10):768–778
16. Younis M. Abbosh (2014) Breast cancer diagnosis using microwave and hybrid imaging
methods. Int J Comp Sci Eng Surv 5(3):41
17. Sun J, Guo M, Pang H, Qi J, Zhang J, Ge Y (2013) Treatment of malignant glioma using
hyperthermia. Neural Regen Res 8(29):2775–2782
Design of Cryoprobe Tip for Pulmonary
Vein Isolation

B. Sailalitha, M. Venkateswara Rao and M. Malini

Abstract Pulmonary vein isolation is a method used for the treatment of

arrhythmias originating within pulmonary veins. Cryoablation is a technique which
employs extreme freezing to treat diseased or abnormal tissue. Cryoablation has
gained importance in the treatment of pulmonary vein isolation for the patients
suffering from atrial fibrillation. The proposed ellipsoidal ring shaped cryoprobe tip
has eliminated the need of continuous point to point lesion ablation by ablating the
entire circumferential area at a time. The aim of this paper is to discuss the design
and advantages of the proposed tip. The desired time for the entire tip to reach a
uniform temperature is ideally less than 2 s. But with the proposed tip the equi-
librium condition was obtained in much lesser time of 0.5 s.

Keywords Atrial fibrillation
Cryoablation
Pulmonary vein isolation

1 Introduction

The cardiovascular system involves heart and blood vessels which carry blood to and
away from the heart [1]. One of the commonly observed cardiovascular diseases is
cardiac arrhythmia. Fibrillation is one type of cardiac arrhythmia. On an average the
normal heart rate of a healthy person is 72 beats/min. When the electrical activity or
heart rate is irregular i.e., faster or slower than normal, then the person is said to be
suffering from cardiac arrhythmia (http://en.wikipedia.org/wiki/Electrical_activity).
Arrhythmias can occur in any part of the heart i.e., in atria or ventricles. They can
affect a person at any age and can cause a sudden cardiac arrest if they are not treated.
When fibrillation occurs in atria it is classified as atrial fibrillation. In atrial fibril-
lation, the electrical impulses are not only produced from SA node; instead many
impulses commence and spread chaotically through the atria. As a result, the

B. Sailalitha (&)
M. Venkateswara Rao
M. Malini

Department of Biomedical Engineering, Osmania University, Hyderabad, India
e-mail: [email protected]

© Springer India 2015 307

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_28
308 B. Sailalitha et al.

heartbeat is aberrant and there is a chance of formation of clots in the atria because
the blood received by the atria is not fully pumped into the ventricles. One of the
most commonly observed atrial fibrillations is arrhythmia originating within the
pulmonary veins. Doctors have discovered that there is a narrow band of muscle
tissue around each of the pulmonary veins near the opening of the left atrium that
may trigger these extra electrical signals as shown in Fig. 1.
In a human heart there are four pulmonary veins that carry oxygenated blood
from the lungs back to the left atrium. When the patient is diagnosed with atrial
fibrillation originating within the pulmonary veins the doctor suggests pulmonary
vein isolation as a mode of treatment [2–5]. Pulmonary vein isolation is an ablation
technique where the surgeon uses cryoablation method to destroy this small area of
abnormal tissue which results in a scar/lesion. These “cryo lesions” are introduced
in a pattern around each pair of pulmonary veins on each side (Fig. 2). As a result
the scar tissue blocks the fast and irregular electrical impulses produced by pul-
monary veins reaching the left atrium thus preventing atrial fibrillation.
To perform this ablation technique a cryoprobe is used to ablate the tissue. The
probe tip is made up of thermally conductive material to get cooled when the
cryogenic gases are allowed to circulate along the probe [6]. The size and shape of
the lesion created on the tissue depends on the size and shape of the tip. So
according to the size and shape of the scar tissue required, respective tip is selected
by the surgeon [7–10].

Fig. 1 Fibrillation
originating within pulmonary
vein

Fig. 2 The required ablating

area on side of pulmonary
veins
Design of Cryoprobe Tip for Pulmonary Vein Isolation 309

2 Methodology

The open heart surgery carried out for pulmonary vein isolation treatment lasts for
around 7 h. For 3 h during the treatment time the cardiopulmonary function of the
patient is maintained by a heart lung machine. When the blood is bypassed and the
heart is frozen the surgeon selects the set of points that are needed to be ablated.
The ablation is done along the circumferential area of the pulmonary veins using a
point to point ablation method. This method requires more than 30 min because the
selection of points has to be done carefully otherwise it may damage a healthy
tissue.
Presently, a 1 cm nipple shaped tip (Fig. 3) is being used by the cardiac surgeons
for creating the lesion. The main drawbacks of using such a shaped tip, is that it
takes more than 30 min for the surgeon to complete the procedure. This is because,
the surgeon has to select the location, ablate it and move to the next location and
repeat the same procedure till the required area is ablated. During the entire pro-
cedure, the patient is kept on a heart lung machine. Prolonged usage of heart lung
machine poses certain problems, and even the patient is prone to infection.
Considering the limitations of the presently used tip, a new ellipsoidal ring
shaped tip is designed. The tip shape is designed such that it almost corresponds to
the area to be ablated. So, it eliminates the need for point to point selection and
subsequent ablation.
The procedure performed is same but the tip is placed on pulmonary veins and
the total desired area is ablated at once. The time taken for the total ablation process
with the proposed tip is less than 10 min. As the tip is designed according to the
measurements of the pulmonary veins, only one time selection is required.
The major advantages of the newly proposed tip are (i) its simplifies the pro-
cedure as it requires only one time ablation (ii) reduces the time of the procedure
and (iii) Reduces the patients dependence on heart lung machine.

Fig. 3 Tip presently used for

pulmonary vein isolation
310 B. Sailalitha et al.

3 Results and Discussion

Different prototypes were designed and shown to the cardiovascular surgeon for
scrutiny. Based on their feedback, the ellipsoidal ring shape is finally selected the
prototype of the proposed tip (Fig. 4). By using this as a model, the evaluation and
design is performed using the mathematical tool (cad software ‘creo’). This is
shown in Fig. 5.

3.1 Design Details of the Tip

Considering the normal measurements of pulmonary veins the dimensions of the tip
were finalised as follows:
Major axis of the ellipsoid: 40 mm
Minor axis of the ellipsoid: 30 mm
Thickness of the ellipsoid: 1 mm
The time taken for the tip to reach equilibrium temperature after the application
of cryo gases as required by the surgeon is <2 s. The cryogenic gas chosen was
carbon dioxide which operates at −78 °C.

Fig. 4 Prototype of the tip

Fig. 5 The design of the tip

in ‘creo’ (CAD software)
Design of Cryoprobe Tip for Pulmonary Vein Isolation 311

The austenitic stainless steel 304 (1.4301) is considered as “tough” at cryogenic

temperatures. AISI 304 is called as “cryogenic steel”. Therefore AISI 304 can be
considered suitable for sub-zero ambient temperatures and hence is chosen to
design the proposed cryoprobe tip.
Volume of ellipsoid, v1 ¼ pða1 b1 ÞðtÞ
where
a1 = length axis radius
b1 = width axis radius
t = diameter of the ellipsoid

ð20Þ ð15Þ ð1Þ

¼p
1000 1000 1000
¼ 942  109 m3
v2 ¼ pða2 b2 ÞðtÞ

where
a2 = length axis radius
b2 = width axis radius
t = diameter of the ellipsoid
ð19Þ ð14Þ ð1Þ
¼p
1000 1000 1000
¼ 835  109 m3

∴ Volume of the ellipsoid ring (V), v1  v2 ¼ 187  109 m3

(1) Mass, m = volume × density = V × d
d ¼ density of 304 stainless steel
d ¼ 8;030 kg=m3
¼ 187  109  8030
¼ 0:000857282 kg

(2) Temperature drops from 21 to −78 °C

∴ Heat removed from the ring = m cp (DT)
where
cp = specific heat of 304
cp = 502 k/kg*k
¼ m ð502Þ ð21  ð78ÞÞ
¼ 42:60 J

If a solid CO2 (Dry Ice) in gases condition,

If Mass, m = 0.0083 kg/s
312 B. Sailalitha et al.

Temperature at −78 °C = 195 K

∴ Heat content of CO2 in gases from = m cp (DT) = 0.0083(1.84 × 103) (5)
In absolute 5 °C raise in CO2 (Dry Ice) = 76.36 J/s
76:36 ¼ 0:55 s
∴ Time taken ¼ 42:60
Through the desired time taken for the tip to reach equilibrium temperature is
less than <2 s. The calculated time for the designed tip is 0.5 s.

4 Conclusion

In the present work, an ellipsoidal ring shaped tip has been designed for the
cryoprobe used for the ablation of pulmonary veins. The limitations of the pres-
ently-being-used nipple shaped tip have been overcome by proposed tip. The
proposed tip also provides the advantages of simplified procedure, reduced time of
the procedure and reduced patient’s dependence on heart lung machine. The
presently designed tip is made of rigid material and the dimensions are based on
average normal values. To suit other dimensions, the tip may be made with flexible
material and tested.

Acknowledgments The authors would like to acknowledge the support from the UGC project
universities with potential for excellence awarded to Osmania University.

References

1. Etheridge ML, Choi J, Ramadhyani S, Bischof JC Methods for characterizing convective

cryoprobe heat transfer in ultrasound gel phantoms. doi:10.1115/1.4023237
2. Kuhne M, Sticherling C (2011) Cryoballoon ablation for pulmonary vein isolation of atrial
fibrillation: a better way to complete the circle? J Innov Cardiac Rhythm Manage 2:264–270
3. Wellens HJJ (2000) Pulmonary vein ablation in atrial fibrillation: hype or hope? Circulation
102:2562–2564. doi:10.1161/01.CIR.102.21.2562
4. Takase et al. (2004) Pulmonary vein dimensions and variation of branching pattern in patients
with paroxysmal atrial fibrillation using magnetic resonance angiography. Jpn Heart J 45:81–92
5. Fraunfelder FW (2008) Liquid nitrogen cryotherapy for surface eye disease. Trans Am
Ophthalmol Soc 106:301–324
6. Rostagno C, Berioli MB, Stefàno PL Treatment of atrial fibrillation in patients undergoing
mitral valve surgery. Dipartimento Area Critica Università Firenze, Cardiochirurgia AOU
Careggi Firenze, Italia
7. Iarocci M et al (1997) Advances in cryogenics engineering. In: Kittle P (ed) vol 43. Plenum
Press, New York, p 499
8. Kubota H, Sudo K, Takamoto S, Endo H, Tsuchiya H, Yoshimoto A, Takahashi Y, Inaba Y,
Furuse A Clinical result of epicardial pulmonary vein isolation (LAVIE) by cry ablation as
concomitant cardiac operation and clinical application of new ablation device(KIRC-119
infrared coagulator) to treat atrial fibrillation
Design of Cryoprobe Tip for Pulmonary Vein Isolation 313

9. Cabrera JA, Pizarro G, Sanchez-Quintana D (2010) Transmural ablation of all the pulmonary
veins: is it the Holy Grail for cure of atrial fibrillation? Eur Heart J 31:2708–2711. doi:10.
1093/eurheartj/ehq241 (online publish-ahead-of-print 6 Sept 2010)
10. Hurlich Low temperature metals, general dynamics astronautics, San Diego, California
Designing of a Multichannel Biosignals
Acquisition System Using NI USB-6009

Gaurav Kulkarni, Biswajeet Champaty, Indranil Banerjee, Kunal Pal

and Biswajeet Mohapatra

Abstract The current study delineates the designing of a three channel biosignals
acquisition system. The acquisition system was made using NI USB-6009.
A LabVIEW based graphical user interface (GUI) program was made to simulta-
neously acquire the biosignals. Electrocardiogram, spirogram and body surface
temperature signals were used as the representative signals for testing the device.
The device was tested successfully and may allow the acquisition of the signals
under ambulatory conditions.

Keywords Multichannel
Biosignal acquisition
NI USB-6009
LabVIEW

Ambulatory conditions

1 Introduction

There has been an increased effort in the development of ambulatory monitoring

devices. This may be attributed to the fact that the ambulatory devices provides the
healthcare providers in diagnosing the cause of the medical conditions at an early
stage [1]. This is specifically important if the condition of the patient is critical. In
general, the ambulatory monitoring devices should be user friendly, low cost, non-
invasive and light weight. Even though the ambulatory devices are expected to be
low cost, the accuracy and the sensitivity of the devices should not be compromised
[2]. This is because a lower accuracy and sensitivity of a medical device may
sometimes lead to wrong diagnosis of the medical condition. This, in turn, will lead
to wrong treatment of the patient thereby aggravating their medical condition.

G. Kulkarni
B. Champaty
I. Banerjee
K. Pal (&)

Department of Biotechnology and Medical Engineering,
NIT-Rourkela, Rourkela 769008, Odisha, India
e-mail: [email protected]; [email protected]
B. Mohapatra
Vesaj Patel Hospital, Rourkela 769008, Odisha, India

© Springer India 2015 315

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_29
316 G. Kulkarni et al.

Keeping this in mind, in this paper we have tried to develop an ambulatory device
capable of acquiring ECG, spirogram and body surface temperature. NI USB-6009
data acquisition device was used for the acquisition of the signals into a computer.
NI USB-6009 was chosen for efficient and accurate acquisition of the biosignals.
An acquisition program was made in LabVIEW.

2 Materials and Methods

2.1 Materials

NI USB-6009 (NI, US), ECG (Vernier, US), spirometer (Vernier, US) and body
surface temperature (Vernier, US) sensors were used in this study. LabVIEW 2010
(NI, US) was used for making the signal acquisition program.

2.2 Methods

The outputs from the ECG, spirometer and surface temperature sensors were
connected to the AI0 (pin no. 2), AI1 (pin no. 5) and AI2 (pin no. 8) analog input
terminals, respectively, of USB 6009 [3]. The sensors were powered with the 5 V,
acquired from the USB-6009. The DAQ was connected with the computer using
USB port. It drew power (5 V) from the USB port of the computer. It transmitted
the signal information serially to the computer using USB port. A LabVIEW
program was made to acquire the signals from the USB acquisition device. A
schematic diagram of the setup has been shown in Fig. 1.

Fig. 1 Schematic diagram of the setup

Designing of a Multichannel Biosignals Acquisition System … 317

3 Results and Discussions

3.1 Designing of the Multichannel Acquisition

The ECG, spirometer and surface temperature sensors were connected with the
USB-6009 using prototype connector board. The pictograph of the sensor con-
nection with the USB-6009 has been shown in Fig. 2. This device is compatible
with LabVIEW and does not require any extra drivers for interfacing. A LabVIEW
program for the simultaneous acquisition of the 3 biosignals was developed. The
schematic representation of the LabVIEW program has been shown in Fig. 3. The
program was made such that the signals from the analog input channels are acquired
simultaneously.
The controls for each of input channels was put in a while loop. The DAQ
assistant is configured to acquire analog input voltages from each channel.
Acquisition mode was set to continuous samples. The sampling rate was set to
1,000 Hz/s and the number of samples to be displayed were chosen to be 10,000
[4]. The signals were then spilt and processed based on the type of biosignals. The
output from ECG channel was processed using multiresolution wavelet analysis.
Since the ECG signal is considered as the time-varying signal, Wavelet trans-
form is being used in ECG signal analysis. The db 06 (Daubechies) wavelet was
used to decompose ECG signal into 8 levels. The signal was reconstructed using

Fig. 2 Sensor connection

with USB-6009
318 G. Kulkarni et al.

Fig. 3 LabVIEW program: a front panel, b block diagram

D6 + D7 levels as most of the energy of the ECG signal lies between 0.1 and 40 Hz
[5]. The signal from spirometer was smoothened using smoothing filter using tri-
angular moving average algorithm using 16 data points. The purpose of smooth-
ening is to eliminate high frequency noises and power line interference. The output
from body surface temperature was fed into express formula vi, which contained
formula for the conversion of the voltage signal into temperature (°C) [6, 7]. The
mean DC of this signal was extracted and displayed in a numeric indicator. The
waveform signals from all the 3 sensors were visualized in three different waveform
graphs on the front panel.
Designing of a Multichannel Biosignals Acquisition System … 319

3.2 Acquisition of the Signals

The acquisition of the ECG signals was carried out in the lead—I configuration. For
the purpose, red electrode was placed in left arm, green electrode in the right arm and
black electrode (right leg drive) was placed near the elbow [8]. The placement of the
electrode was done as per the manual provided by the manufacturer. The output of
the ECG sensor was fed into AI0 channel of the DAQ. The output of the spirometer
was fed into AI1 channel of USB-6009, whereas, output of the surface temperature
sensor was fed into AI2 channel of USB-6009 [9]. The spirometer was placed near
the mouth. The surface temperature sensor was placed on the forearm and was
secured using a white plaster. The representative output from the sensors have been
shown in Fig. 4. The results were found to be as reported in the literature [10].
After designing the ambulatory device, 5 volunteers were invited to test the
functioning of the device. The experimental details were described to the volun-
teers. A written consent was taken from the volunteers, if they agreed to participate
for the study. An ethical clearance was previously taken from the institute ethical
clearance committee to conduct biosignals acquisition studies. The volunteers were
asked to sit in the supine position and the sensors were placed at the predefined
position as described in the previous paragraph. The photograph of a volunteer with
the sensor placements has been shown in Fig. 5.
The developed device was able to acquire all the three signals in an efficient
manner without any noise or distortion. The device was found to be quite user
friendly. The complete setup of the device has been shown in Fig. 6.

Fig. 4 Picture showing representative output from the sensors

320 G. Kulkarni et al.

Fig. 5 Photograph of
volunteer with sensor
placement

Fig. 6 Complete setup of the

device
Designing of a Multichannel Biosignals Acquisition System … 321

4 Conclusion

The current study describe the designing of a multichannel biosignals acquisition

system using NI USB-6009. The software for controlling the hardware platform
was developed in LabVIEW. The biosignals were acquired efficiently and displayed
using LabVIEW program. This device can be used in ambulatory monitoring
applications for continuous acquisition of ECG, spirogram and body surface
temperature.

Acknowledgments Authors extend the acknowledgement for the logistical support provided by
National Institute of Technology, Rourkela during the completion of the study.

References

1. Corral-Peñafiel J, Pepin J-L, Barbe F (2013) Ambulatory monitoring in the diagnosis and
management of obstructive sleep apnoea syndrome. Eur Respir Rev 22(129):312–324
2. Branzila M, David V Wireless intelligent systems for biosignals monitoring using low cost
devices
3. Zhang J-G, Zhao X (2014) Design of the chaotic signal generator based on LABVIEW. Sens
Transducers 163:1726–5479
4. Shen L, Chaoran Y (2012) Identification of parameters for a DC-motor by LabVIEW
5. Selvakumar G, et al (2007) Wavelet decomposition for detection and classification of critical
ECG arrhythmias. In: proceeding of the 8th WSEAS international conference on mathematics
and computers in biology and chemistry pp 80–84
6. Granado Navarro M.Á (2012) Arduino based acquisition system for control applications
7. Sahoo S et al (2014) Wireless transmission of alarm signals from baby incubators to neonatal
nursing station. In: automation, control, energy and systems (ACES) 2014 first international
conference pp 1–5
8. Biel L et al (2001) ECG analysis: a new approach in human identification. IEEE Trans Instrum
Meas 50(3):808–812
9. Kumar U et al (2013) Design of low-cost continuous temperature and water spillage
monitoring system. In: IEEE 2013 international conference on information communication
and embedded systems (ICICES)
10. Heywood D et al (2009) Integrating the Vernier LabPro™ with Squeak Etoys. In: Society for
information technology and teacher education international conference
Arsenic Removal Through Combined
Method Using Synthetic Versus Natural
Coagulant

Trina Dutta and Sangita Bhattacherjee

Abstract As Arsenic contamination in drinking water has become a matter of

severe concern, researchers across the globe have been trying to develop an efficient
yet economic method for removal of arsenic from water. Authors of this paper
investigated combined treatment methodology namely, coagulation followed by
microfiltration (MF) to remove arsenic below permissible level of 10 ppb. Simu-
lated solutions of arsenate and arsenite salts (100 ppb) were prepared. A 95.83 %
reduction of arsenic for arsenate solution and 91.99 % reduction of arsenic in case
of arsenite solution were achieved in suitable pH range (8−10) when the solution
was subjected to coagulation using ferric chloride coagulant, followed by MF using
0.45 µm polyethersulphone membrane. In search of more cost effective and eco-
friendly yet viable route, authors explored the effectiveness of crushed Shelled
Moringa oleifera Seed (SMOS) as a natural coagulant. A 91.01 % decontamination
of arsenic for arsenate solution and 70.61 % for arenite solution were achieved in
suitable pH range 7−9. This alternative method of coagulation, bio-adsorption
(amino acid-arsenic interaction) followed by MF achieved in appreciable arsenic
removal efficiency compared to inorganic Ferric Chloride, a synthetic coagulant.
The sludge generated in case of ferric chloride was found to be toxic and highly
corrosive compared to that obtained with Moringa oleifera seeds.

Keywords Arsenic poisoning Coagulation followed by microfiltration
Ferric


chloride versus natural coagulant Moringa oleifera Arsenic removal

T. Dutta (&)
Chemistry Department, JIS College of Engineering, Kalyani, Nadia 741235, India
e-mail: [email protected]
S. Bhattacherjee
Chemical Engineering Department, Heritage Institute of Technology, Kolkata 700107, India
e-mail: [email protected]

© Springer India 2015 323

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_30
324 T. Dutta and S. Bhattacherjee

1 Introduction

Nowadays, arsenic poisoning has become one of the major environmental concerns
in the world as millions of human beings have been exposed to excessive arsenic
through contaminated ground water and surface water used for drinking. Arsenic, a
metalloid in group VA, is classified as Group 1 carcinogenic substance based on
powerful epidemiological evidence [1]. Hence, a continuous investigation is being
carried out across the world to develop an efficient yet economical technology for
reducing arsenic concentration below maximum contamination level (MCL) of
10 ppb as implemented by US Environment Protection Agency in the year 2001.
There are many possible routes of human exposure to arsenic from both natural
and anthropogenic sources. Arsenic occurs as a constituent in more than 200
minerals, although it primarily exists as arsenopyrite and as a constituent in several
other sulfide minerals. The introduction of arsenic into drinking water can occur as
a result of its natural geological presence in local bedrock. Arsenic-containing
bedrock formations of this sort are known in Bangladesh, West Bengal (India), and
regions of China, and many cases of endemic contamination by arsenic with serious
consequences to human health are known from these areas. Significant natural
contamination of surface waters and soil can arise when arsenic-rich geothermal
fluids encounter surface waters. When humans are implicated in causing or exac-
erbating arsenic pollution, the cause can usually be traced to mining or mining-
related activities [2].
The acceptable level as defined by WHO for maximum concentrations of arsenic
in safe drinking water is 0.01 mg/L. Arsenic contaminated water typically contains
arsenous acid (As III) and arsenic acid (As V) or their derivatives. Their names as
“acids” is a formality, these species are not aggressive acids but are merely the
soluble forms of arsenic near neutral pH. These compounds are extracted from the
underlying rocks that surround the aquifer. Arsenic acid tends to exist as the ions
[HAsO4]2− and [H2AsO4]− in neutral water, whereas arsenous acid is not ionized.
Arsenic removal from water is an important subject worldwide, which has
recently attracted great attentions. A variety of treatment processes has been
developed for arsenic elimination from water, including coagulation (precipitation),
adsorption, ion exchange, membrane filtration, electrocoagulation, biological pro-
cess, iron oxide-coated sand, high gradient magnetic separation and natural iron
ores, manganese greensand etc.
Investigators have also used modern technology such as membrane technology
which includes reverse osmosis, nanofiltration, ultrafiltration, microfiltration to
remove arsenic species from drinking water. Reverse osmosis, can reduce arsenic
limit below 10 ppb. But in developing countries like ours considering with low
annual income and low electric popularization, highly efficient RO and NF tech-
nology seems difficult to be applied due to its high-energy consumption.
Microfiltration (MF), is a low-pressure technique and can remove particles with
a molecular weight above 50,000. The pore size of MF membranes is too large to
effectively remove dissolved or colloidal arsenic species, however, the MF can
Arsenic Removal Through Combined Method Using Synthetic Versus Natural Coagulant 325

remove particulate forms of arsenic from water. Therefore, the arsenic removal
efficiency by MF membrane is highly dependent on the size distribution of arsenic
bearing particles in the water. Unfortunately, the percentage of particulate arsenic
normally is not too high in water. In order to increase the arsenic removal efficiency
by MF technique, the technique, which can increase arsenic particle size, such as
coagulation [3] and flocculation processes [4], can be used to assist the MF
techniques.
In this context, the authors have investigated the performance of a combined
method, namely coagulation followed by microfiltration in removal of arsenic from
contaminated water. In one study, ferric Chloride salt was selected as synthetic
coagulant and another study Moringa olifera was used as natural coagulant. Mi-
crofiltration in both the cases was used to remove arsenic particulates larger than
0.5 µm. Both the studies were carried out and results were reported.

2 Objective

In case of synthetic coagulant, Ferric Chloride, the coagulation process consisted of

the addition of an iron-based coagulant, ferric chloride that would get hydrolysed in
water to form Fe(OH)3 precipitate in the pH range of 4−10. Ferric hydroxide floc
having a net positive charge on the surface of the precipitate formed, would
favourably adsorb negatively charged arsenate by surface complexation. Arsenite
being neutral in charge, would also likely to be adsorbed on ferric hydroxide flocs,
was as effective and reliable synthetic coagulant due to its dense-discrete floc for-
mation over wide pH range and due to formation of less toxic and compact sludge.
But presence of synthetic coagulant (iron chloride, polyaluminum chloride etc.)
in processed water may have several serious drawbacks viz. Alzheimer’s disease,
health problems, carcinogenic effects if used for drinking purposes. Hence such
treatments have not been appreciated [5].
Therefore the necessity of natural coagulant instead of synthetic coagulant was
required to be used for arsenic removal. The process of water treatment using
natural coagulant is called bio-sorption. Biosorption has gained important credi-
bility during recent years because of its ecofriendly nature, excellent performance,
and low cost domestic technique for remediating even heavily metal loaded waste
water [6−8]. Recently, biological materials such as algae, bacteria, fungi, yeast
[9, 10], aquatic moss Fontinalis antipyretica [11] and agricultural waste products
like sunflower stalk [12], sargassum seaweed biomass [13], pinus cone biomass
[14], rice husk [15], olive pomace [16], cocoa shells [17], and grape stalk waste [18]
have been recognized as cheap natural sorbents for the removal of toxic metals.
However, there is still a strong challenge in developing economical and commonly
available biosorbents for Arsenic removal.
Moringa oleifera (Saihjan or drumstick) is one of the biosorbent, also known as
‘Miracle Tree’. It is a cosmopolitan tropical drought tolerant tree, available
throughout the year and is well documented for its various pharmacological
326 T. Dutta and S. Bhattacherjee

importances such as analgesic, antihypertensive activity, and anti-inflammatory

effects [19−21]. Further, the coagulating property of the seed powder of this plant
has been used for various aspects of water treatment such as turbidity, alkalinity,
total dissolved solids, and hardness [5, 22–25]. In Sudan, Northern Nigeria and
other developing countries, dried M. oleifera seeds are frequently used for removal
of turbidity in water as low technology domestic method [26]. However, its bio-
sorption behavior for the removal of toxic metals from water has not been given
attention. Moreover, the use of Moringa oleifera has an added advantage over the
chemical treatment of water because not only it is cheap and eco-friendly but also it
is biological and has been reported as edible. Moringa extract contain dimeric
cationic proteins or polypeptide (natural organic polymer) with various functional
groups, particularly low molecular weight amino acids [27]. These proteinacious
amino acids have a variety of structurally related pH-dependent properties of
generating either positively or negatively charged sites for attracting the anionic or
cationic species of metal ions, respectively [28]. It has molecular weight of 13 kDa
and isoelectric points in the range 4−8 [29]. Moreover its efficiency as coagulant is
not affected by pH of the solution.
In the light of the above, the major objective of this study was to investigate
removal efficiencies of arsenate and arsenite from simulated water by combination of
coagulation followed by microfiltration by varying dosages of the natural coagulant-
cum-polyelectrolyte M. oleifera as well as ferric chloride. The optimum dose has
been investigated to reach MCL standard and to make it cost effective as well.

3 Materials and Methods

Sodium arsenate salt (Na2HAsO4 · 7H2O), (Molecular Weight = 312) and Sodium
arsenite (NaAsO2) (Molecular Weight = 129.9) AR grade procured from Loba
Chemie Pvt. Ltd. Ferric Chloride salt, sodium hydroxide pellets were procured from
E. Merck, India. The required solutions were prepared with ultrapure deionised
water. 1 N Hydrochloric acid and 1 N Sodium Hydroxide solutions were prepared
for pH adjustment.
Moringa oleifera Lamarck seeds were collected during summer. Seeds were
washed thoroughly with double-distilled water to remove the adhering dirt, dried at
60 °C for 24 h. For shelled seeds (SMOS), the husks enveloping each seed were
removed and the kernel was ground to a fine powder using a blender. No other
chemical or physical treatments were used prior to experiments.
In the study, polyethersulfone (PES) membrane, a hydrophilic membrane con-
structed from pure polyethersulfone polymer was used during MF. PES membrane
filters are designed to remove particulates during general filtration. The polye-
thersulfone membrane filters have excellent flow speeds and, connected to it, a high
filterable volume. Biologic and pharmaceutical solutions can be filtered, in the wide
pH-range of pH 1−14, because of their low protein adsorption.
Arsenic Removal Through Combined Method Using Synthetic Versus Natural Coagulant 327

4 Experimental Procedure

Accurate amount of sodium arsenate was weighed and dissolved in 500 ml of

ultrapure water to prepare 100 µg/l of arsenic solution. In 1st iteration 50 ml of this
solution was taken in each of 4 different conical flasks. 10, 15, 20, 25 % FeCl3
solutions were prepared 0.10 ml of FeCl3 solution was added to each of the 4 flasks
respectively. The pH of the solution was increased and kept in the range of 8−10 by
adding 1 N NaOH solution. These were stirred magnetically using a magnetic stirrer
for 15 min at 30 °C. Flocs are formed upon increasing the pH. The solution was kept
for half an hour to settle down the flocs and was filtered using Whatman filter paper
(Grade 41) to remove suspended coarse particles (20−25 µm). For removal of finer
particles, the supernatant solution was then microfiltered using a “all glass vacuum
filtration unit” (make: Sartorius A.G., Göttingen, Germany), fitted with portable
vacuum pump (Sartorius, A.G., Göttingen, Germany), with polyether sulfone (PES)
membrane (47 mm diameter, pore size 0.45 µm) being used as filter media. In the
2nd set of study, 50 ml of similar simulated arsenate solution was taken in four
different conical flask. 0.1, 0.5, 1 and 1.5 g of powdered Moringa oleifera Seed
(SMOS) were mixed with the solution respectively and pH value adjusted in the
range 7−9. Then the samples were kept on the shaker for 45 min at 110−120 rpm.
After settling down big flocks, the soutions were filtered by whatman filter paper
(Grade 41) and supernatant is followed by microfiltration (Figs. 1 and 2).
The arsenic concentration of permeate collected from both the processes were
then measured by Perkin-Elmer Atomic absorption spectroscopy (AAS). Similar
procedure has been followed for arsenite salt also and results were noted.

Fig. 1 FeCl3 conc (%) on

arsenate removal (%)
328 T. Dutta and S. Bhattacherjee

Fig. 2 FeCl3 conc (%) on

arsenite removal (%)

5 Result and Discussion

The removal of As(V) and As(III) from simulated water has been investigated by
coagulation/adsorption followed by microfiltration process.
In 1st iteration, FeCl3 has been taken as coagulant here because Iron based
coagulants are generally more effective than aluminum based coagulants for arsenic
removal from water on a weight basis [30]. Ferric chloride is a primary coagulant.
Primary coagulants are used to cause particles destabilised and begin to clump
together. When ferric chloride is dissolved in water the solution becomes strongly
acidic as a result of hydrolysis.
As the pH was adjusted, ferric hydroxide precipitated from the solution. As the
pH decreased, the number of positively charged sites on the ferric hydroxide par-
ticles increased. The precipitated ferric hydroxide had a net positive charge on the
surface and the negatively charged arsenate anions got adsorbed on the surface of
ferric hydroxide precipitate by surface complexation models. In this way, the
arsenate got removed from the solution. The sedimentation and filtration processes
then remove arsenic particulates.
It has been found that both arsenate and arsenite could be effectively removed
with 20 % FeCl3 dosage followed by microfiltration.
In the subsequent of treatment, crushed, powdery natural coagulant Moringa
oleifera seed was selected. As it contains dimeric cationic polypeptide it can easily
coagulate with negatively charged arsenic salt. It had been reported that an aqueous
solution of the Moringa seeds is a heterogeneous complex mixture, having cationic
polypeptides with various functional groups, especially low molecular weight
amino acids [27]. Amino acids have been reported as efficient phytochelators that
work at even at low concentrations and having the tendency to interact with metal
ions and significantly enhance their mobility. The proteinacious amino acids,
depending upon the pH, possess both the negatively and positively charged ends
Arsenic Removal Through Combined Method Using Synthetic Versus Natural Coagulant 329

and are, thus, capable of generating the appropriate atmosphere for attracting the
anionic or cationic species of the metal ions.
Here pH value was maintained between 7 and 9. In this pH range, As(V) exists
in monovalent (H2AsO4−), and divalent (HAsO42−) anionic species and As(III)
exists in (H2AsO3− and HAsO32−) anionic species [8]. The positively charged
group of amino acids can hold the negatively charged monovalent arsenic species.
The majority of amino acids present in the target biomass have isoelectric points
in the range 4.0−8.0 [29]. In this range of pH, over 90 % of the amino acid
molecules are in the ionized state. The negatively charged monovalent arsenate or
arsenite species may be held by the positively charged group of amino acids. With
the increase in pH range, the carboxylic group of the amino acids would pro-
gressively be deprotonated as carboxylate ligands, simultaneously protonating the
amino group. Such positively charged amino groups facilitate the SMOS-Arsenic
binding. SMOS contains proteins and has a net positive charge at pH below 8.5
(pKa > 8.6) [31]. These positively charged proteins are also considered to be active
moieties for binding metal ions.
Here the optimum dose has been found as 0.5 g per 50 ml means 10 g/L for both
arsenate and arsenite salt. The removal of arsenic resulted from precipitation, co-
precipitation and adsorption mechanisms. With further increase in coagulant dos-
age, arsenic removal had not improved significantly. On the other hand a further
increase in coagulant dose causes restabilization of the particles as the charge
reversal on the colloids occurs [32]. So it can be concluded that lower dosage of
investigated natural coagulants was better than higher ones. This is very important
not only for process economy but also for lower organic matter load in processed
water because it is known that high organic load might cause microbial growth [33].
Arsenic removal was calculated by the formula

% Removal ¼ ðCi  Cf Þ=Ci  100

Ci is the initial concentration (μg/L) and Cf is the final arsenic concentration

(μg/L) (Figs. 3 and 4).

Fig. 3 SMOS dose

(gm/50 ml) on arsenate
removal (%)
330 T. Dutta and S. Bhattacherjee

Fig. 4 SMOS dose

(gm/50 ml) on arsenite
removal (%)

6 Conclusion

From the present study, it can be concluded that arsenic contamination can be
brought down below MCL limit of 10 ppb by using natural biosorbent, Moringa
oleifera. A comparable removal of 91.01 % for arsenate salt and 70.61 % for
arsenite salt have been obtained using SMOS compared to synthetic coagulant,
FeCl3. More over natural coagulant is cheaper has no toxicity and corrosiveness as
in case with synthetic coagulant. Moreover the produced sludge can be used as bio-
fertilizers and it becomes an added advantage of this treatment.
Shelled M. oleifera seeds (SMOS) provides an exciting opportunity under the
domain of Green Processes for domestic, environment friendly, low-cost methods
for the decontamination of toxic metals from aqueous systems. It is believed that, in
the near future, SMOS could be a potential challenger of synthetic coagulants for
the treatment of contaminated drinking water. As a coagulant, Moringa is non-toxic
and biodegradable. It is environmentally friendly, and unlike FeCl3, does not sig-
nificantly affect the pH and conductivity of the water after the treatment. Sludge
produced by coagulation with Moringa is not only innocuous but also four to five
times less in volume than the chemical sludge produced by FeCl3 coagulation. So,
as a coagulant, Moringa oleifera can be a potentially viable substitute of FeCl3.

References

1. IARC (International Agency for Research on Cancer) (1987) IARC monographs on the
evaluation of carcinogenic risk to humans: overall evaluations of carcinogenicity: an updating of
IARC monographs, 1–42(7). International Agency for Research on Cancer Lyon, pp 100–206
2. Garelick H, Jones H, Dybowska A, Valsami-Jones E (2008) Rev Environ Contam Toxicol
197:17−60 (Department of Natural Sciences, School of Health and Social Sciences, Middlesex
University, The Burroughs, London NW4 4BT, UK)
Arsenic Removal Through Combined Method Using Synthetic Versus Natural Coagulant 331

3. Ghurey G, Clifford D, Tripp A (2004) Iron coagulation and direct microfiltration to remove
arsenic from groundwater. J AWWA 96(4):143–152
4. Han B, Runnells T, Zimbron J, Wick-ramasinghe R (2002) Arsenic removal from drinking
water by flocculation and microfiltration. Desalination 145:293–298
5. Okuda T, Baes AU, Nishijima W, Okuda M (2001) Coagulation mechanism of salt solution-
extracted active component in Moringa oleifera seeds. Water Res 35:830–834
6. Chandrasekhar K, Kamala CT, Chary NS, Anjaneyulu Y (2001) Removal of heavy metals
using a plant biomass with reference to environmental control. Int J Miner Process 68:37–45
7. Basu A, Kumar S, Mukherjee S (2003) Arsenic reduction from environment by water lettuce
(Pistia stratiotes L.). Ind J Environ Health 45:143–150
8. Ghimire KN, Inoue K, Makino K, Miyajima T (2002) Adsorptive removal of arsenic using
orange juice residue. Sep Sci Technol 37:2785–2799
9. Matuschka B, Strauba G (1993) Biosorption of metals by a waste biomass. J Chem Technol
Biotechnol 37:412–417
10. Pagnanelli F, Papini PM, Toro L, Trifoni M, Veglio F (2000) Biosorption of metal ions on
Arthrobacter sp.: biomass characterization and biosorption modeling. Environ Sci Technol
34:2773–2778
11. Martins RJE, Pardo R, Boaventura RAR (2004) Cadmium(II) and Zinc(II) adsorption by the
aquatic moss Fontinalis antipyretica: effect of temperature, pH and water hardness. Water Res
38:693–699
12. Sun G, Shi W (1998) Sunflower stalks as adsorbents for the removal of metal ions from waste
water. Ind Eng Chem Res 37:1324–1328
13. Kratochvil D, Pimentel P, Volesky B (1998) Removal of trivalent and hexavalent chromium
by seaweed biosorbent. Environ Sci Technol 32:2693–2698
14. Ucun H, Bayhan YK, Kaha Y, Cakici A, Algur OF (2002) Biosorption of Chromium(VI) from
aqueous solution by cone biomass of Pinus sylvestris. Bioresour Technol 85:155–158
15. Khalid N, Rahman A, Ahmad S, Kiani SN, Ahmmad J (1998) Adsorption of cadmium from
aqueous solutions on rice husk. Plant Soil 197:71–78
16. Pagnanelli F, Sara M, Veglio F, Luigi T (2003) Heavy metal removal by olive pomace:
biosorbent characterization and equilibrium modeling. Chem Eng Sci 58:4709–4717
17. Tyagi RD, Blais JF, Laroulandie J, Meunier N (2003) Cocoa shells for heavy metal removal
from acidic solutions. Bioresour Technol 90:255–263
18. Isabel V, Nuria F, Maria M, Nuria M, Jordi P, Joan S (2004) Removal of copper and nickel
ions from aqueous solutions by grape stalks wastes. Water Res 38:992–1002
19. Caceres A, Saravia A, Rizzo S, Zabala L, Leon ED, Nave F (1992) Pharmacologic properties
of Moringa oleifera: screening for antispasmodic, anti-inflammatory and diuretic activity.
J Ethnopharmacol 36:233–237
20. Marugandan S, Srinivasan K, Tandon SK, Hasan HA (2001) Anti-inflammatory and analgesic
activity of some medicinal Plants. J Med Arom Plants Sci 22:56–58
21. Dangi SY, Jolly CI, Narayanan S (2002) Anti-hypertensive activity of the total alkaloids from
the leaves of Moringa oleifera. Pharm Biol 40:144–148
22. Suleyman A, Muyibi SA, Evison LM (1995) Moringa oleifera seeds for softening hard water.
Water Res 29:1099−1105
23. Ndabigengesere A, Narasiah KS (1998) Quality of water treated by coagulation using Moringa
oleifera seeds. Water Res 32:781–791
24. Kalogo Y, Verstraete W (2000) Technical feasibility of the treatment of domestic waste water
by CEPS–UASB system. Environ Technol 21:55–65
25. Megat J (2001) Moringa oleifera seeds as a flocculants in waste sludge treatment. Int J Environ
Stud 58:185–195
26. Ndabigengesere A, Narasiah KS, Talbot BG (1995) Active agents and mechanism of
coagulation of turbid water using Moringa oleifera. Water Res 29:703–710
27. Jose TA, Oliveira, Silveria BS, Vasconcelos LM, Cavada BS, Moriera RA (1999)
Compositional and nutritional attributes of seeds from the multipurpose tree Moringa
oleifera Lam. J Sci Food Agric 79:815−820
332 T. Dutta and S. Bhattacherjee

28. Costa G, Michant JC, Guckert G (1997) Amino acids exuded form cadmium concentrations.
J Plant Nutr 20:883–900
29. Delvin S (2002) Amino Acids and Proteins, 1st edn. IVY, New Delhi
30. Yuan T, Luo Q, Hu J, Ong S, Ng W (2003) A study on arsenic removal from household
drinking water. J Environ Sci Health A 38:1731−1744
31. Makker HPS, Becker K (1997) Nutrients and antiquality factors in different morphological
parts of Moringa oleifera tree. J Agric Sci (Cambridge) 128:311−322
32. Duan J, Gregory J, (2003) Coagulation by hydrolysing metal salts. Adv Colloid Interface Sci
100−102:475−502
33. Šciban M, Klašnja M, Antov M, Škrbic B (2009) Removal of water turbidity by natural
coagulants obtained from chestnut and acorn. Bioresour Technol 100:6639–6643
Development of Novel Architectures
for Patient Care Monitoring System
and Diagnosis

M.N. Mamatha

Abstract Designing a highly efficient patient care and monitoring system which
can handle multiple patients and multiple parametric measurements from every
single patient in real time will lead in improvising the data handling capability at
Central Nurse Stations (CNS) and Decentralized Nurse Stations (DCNS). The Bio
signal Data Acquisition Systems have been designed to suit patients located at CNS
and DCNS in a hospital. The RTL design of Bio signal Data Acquisition System
was successfully simulated using Model sim. The design was placed and routed
using Xilinx ISE 8.2i tool and the bit stream generated is used for downloading into
the targeted FPGA. The FPGA used is XC3S400-5ft256, common for both schemes
shown. At 50 MHz operation, the system is capable of wireless communication of
32 numbers of Bio signals up to 400,000 bits/s although the maximum frequency of
operation of this design is 89 MHz reported by Xilinx ISE tool. This system is truly
upgradeable, be it in extending the capabilities to more number of patients or in
improving throughput.

Keywords PCM
CNS
DCNS
DAS
FPGA
ZIGBEE

1 Introduction

The availability of prompt and expert medical care can meaningfully improve
health care services particularly at understaffed rural or remote areas [1]. Continents
facing continuous threats due to spread of infectious diseases, high levels of infant
and maternal mortality, low level of life expectancy and deteriorating health care
facilities are the greatest beneficiaries of continuous patient care monitoring and
assisted with quick diagnosis techniques whenever required. To handle emergency
situation, the main requirement is to continuously monitor intensive care parameters

M.N. Mamatha (&)

Department of Electronics and Instrumentation, B.M.S.C.E, Bangalore, India
e-mail: [email protected]

© Springer India 2015 333

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_31
334 M.N. Mamatha

of patients and simultaneously display all captured information to the competent or

decision making experts (doctors) anywhere at any time [2–4]. A closer look at the
patient monitoring system reveals that it is the doctor alone who is remotely located
whereas the emergency site is at a stable point.
The advent of wireless systems has enabled hospitals to customarily measure
and analyze the vital signs of surgical, trauma, and other critical care patients
throughout their stay in the hospital. Patient Monitoring devices have now been
developed for pulse rate, heart rate, temperature, Oxygen Saturation (SpO2), Central
Venous Pressures (CVP), respiration rate, Pulmonary Pressures, EEG, ECG and
blood pressure, among other parameters [5, 6]. These systems are able to interface
with hospital information technology to provide caregivers with immediate and
regular observation of patient status.
Recent advances in embedded computing systems have led to the emergence of
wireless sensor networks consisting of small, battery powered devices with limited
computation and radio communication capabilities. Sensor networks permit data
gathering and computation to be deeply embedded in the physical environment
[7, 8]. This technology has the potential to impact the delivery and study of
resuscitative care by allowing vital signs to be automatically detected and fully
integrated into the patient care used for real time storage, correlation of hospital
records, long term observation, etc.

2 Bio Signal Data Acquisition for Multi-patient Care

and Monitoring

The present work focuses on implementing a Bio signal Data Acquisition system
with an FPGA which can cater to numerous physiological parameters of the patient
that are acquired by the sensors or electrodes as shown in Fig. 1. In spite of the
system catering to multi patients, it processes the acquired data rapidly in real time
since its design is heavily pipelined and massively parallel. This design may be
expressed in two different schemes: One as Decentralized Nurse Station without
wireless communication and the other as DCNS with Zigbee or LAN. For the latter
scheme, we also need to design a Centralized Nurse Station for receiving the
acquired data, storing data and Diagnostics. Since this system is exact inverse of
DCNS and involved, only DCNS design has been undertaken in the present work.

3 Scheme Proposed for Patient Care Monitoring System

The proposed work comprises the design of Single as well as Multi Patient Care
Acquisition Systems involving the following developments.
Development of Novel Architectures for Patient Care Monitoring … 335

Fig. 1 Block diagram of the proposed bio-signal data acquisition system

1. Development of novel algorithms.

2. Design of architectures for patient care and monitoring for different schemes.
(i) A single FPGA which can store a number of physiological parameters from
each patient and may be equipped in semi-special ward which houses up to 8
patients in a decentralized nurse station is used. A localized data acquisition
is proposed to make it cost effective.
(ii) In this scheme, a higher capacity FPGA would be used to acquire bio signals
from multi patients and communicate them to a Central Nurse Station in real
time. The system shall be endowed with wireless duplex communication
ability.
336 M.N. Mamatha

4 Architectures of Schemes Proposed

EEG, EOG, EMG and Temperature signals from multi patients will be continuously
acquired, stored and displayed at DCNS. The acquired signals at the DCNS are
transmitted to CNS XBee Pro, also popularly known as Zigbee at 400,000 bits/s.
This means that the throughput of the system is 40,000 Bytes/s since there are 10
bits to be transmitted per character. The Bio signals from the patient bedside are
hard wired to DCNS.
The Multi patient care and monitoring system for Scheme 1 and 2 are shown in
Figs. 2 and 3 respectively. The two schemes are the same except for the Zigbee
controller design and XBee Pro module, which is incorporated only in Scheme 2.
As shown in the Figs. 1 and 2, 32 numbers of bio signals are acquired using
sensors and signal conditioning circuits at the bedside of the patient. These signals
are hardwired to amplifiers at the Data Acquisition Module of the DCNS. The
signal conditioning circuits use Instrumentation amplifiers in order to get a faithful
reproduction of the original signals. The amplified signals are fed to 4 Numbers of
ADC 0809, which acquire signals concurrently, all activities being coordinated by a
Centralized Controller.
The acquired signals are stored in 4 K Bytes of Dual RAM Bank. The Patient as
well as the System Monitoring is managed by the Controller. Using the keyboard,
Patient information such as ID and signals used can be programmed in a non
volatile storage. An LCD display facilitates the viewing of these inputs in addition
to displaying alarm conditions such as the Temperature is very high, reminder for
taking medicines etc. A sound alarm accompanies such displays to alert the patients
or nursing staff concerned. The features mentioned so far are common to both the
schemes.

Fig. 2 Multi patient care monitoring system

Development of Novel Architectures for Patient Care Monitoring … 337

Fig. 3 Multi patient care monitoring transmission

As shown in Fig. 3, a Zigbee Controller is designed in order to communicate the

acquired Bio signals to a wireless XBee Pro module. The Zigbee transmitter and
receiver are connected to DI and DO pins respectively of the XBee Pro.

5 Proposed Bio Signal Data Acquisition System

The architecture for Scheme 2 only is described in Fig. 4, here since the Scheme 1
is a subset of Scheme 2. The top Architectural design of Bio Signal Data Acqui-
sition System as implemented in this work was presented in order to facilitate the
development of the Algorithm. Figures 2 and 3 shows the overall architecture of
Bio-DAS. The top design module “bio_das” instantiates three other sub modules
“bio_da” which controls 4 Nos. of ADCs, Dual RAM Bank (dualram_bank) and
Zigbee Controller (zigbeec).
The detailed Architecture of Bio signal data acquisition system has been realized
on a single FPGA in this work.

5.1 Detailed Architecture of Dual RAM Bank

The ADC outputs are stored in intermediate buffers “din1” to “din4” acquired from
ADC signals “oadc1” to “oadc4” respectively. These signals are fed to Dual RAM
Bank. It contains 32 Nos. of Dual RAMs, which in turn consists of two RAMs,
338 M.N. Mamatha

reset_n ale

clk
sc
oadc1 [7:0]

adc_clk
oadc2 [7:0] BIO SIGNAL
DATA
ACQUISTION chno [2:0]
oadc3 [7:0] SYSTEM
zb_reset_n
( bio_das) To/From
oadc4 [7:0] zb_di XBee Pro

zb_rts_n Onlyfor
eoc1
Scheme 3
zb_do
eoc2

eoc3 zb_cts_n

eoc4

Fig. 4 Architecture of bio signal data acquisition system

RAM1 and RAM2 of size 64 Bytes each. Each location in RAM is of 8 bits size.
The control input signal “rnw” configures one of the two RAMs in write only mode
and the other in read only mode alternately. A ‘high’ at “rnw” configures RAM1
in write mode and RAM2 in read mode and vice versa. Once configured, the RAM
in write mode gets all the data until “rnw” toggles. The memory size of the RAM in
the proposed work is 32 × 2 × 64 Bytes.
The signal “rnw” is connected to RAM 1 directly, whereas its inverted signal is
connected to RAM2. The outputs of each RAM “dout1” and “dout2” are multi-
plexed and issued as the output “dout”.
As shown therein, the data bus “din1” to “din4” are input to the Dual RAMs
marked for Patients P1/P2, P3/P4, P5/P6 and P7/P8. For each Patient, 4 Nos. of
Dual RAMs are allocated so that they may store 4 × 2 × 64 Bytes of samples at the
most for the four types of signals: EEG, EOG, EMG and Temperature put together.
The validity of this data input signal (“din_valid1” to “din_valid32”) for all the
32 Dual RAMs are asserted individually by the “bio_da” module. It may be noted
that four valid signals such as “din_valid1”, “din_valid9”, “din_valid17” and
“din_valid25” are activated simultaneously so that the corresponding samples may
be stored concurrently. While reading, “dout_sel1” to “dout_sel32” are activated
one after another for every 64 Bytes read.
The XBee Pro module of Rhydolabz is engineered to meet IEEE 802.15.4
standards and support the unique needs of low-cost, low-power, wireless sensor
Development of Novel Architectures for Patient Care Monitoring … 339

networks. This module requires minimal power and provides reliable delivery of
data between devices.

5.2 Detailed Architecture of Zigbee Controller

The XBee Pro Wireless Module interfaces to a host device such as the proposed
single FPGA chip DCNS through a logic level, asynchronous serial port. Through
its serial port, the module can communicate with any designed logic. The module
has a transmitter and a receiver. The interface between the DCNS and the XBee Pro
Module is shown in Fig. 5. At the other end of wireless communication, another
XBee Pro Module interfaces with a CNS, which design is beyond the scope of the
present work.

(a)
di
di

DCNS cts CNS

XBee XBee cts
FPGA Pro Pro FPGA
do do

rts
rts

(b)

Fig. 5 a XBee pro interfaces to DCNS and CNS, b XBee, c Simulated output
340 M.N. Mamatha

Fig. 5 (continued)

The architecture of XBee Pro Interfaces to DCNS and CNS is shown in

Fig. 5a–c.

6 Results

The experimental set up of Multi PCM with DAC and ADC cards are as shown in
Fig. 6
• Developed fast, novel algorithms suitable for FPGA/ASIC implementation.
designed a massively parallel and highly pipelined architecture suitable for
FPGA implementation for multi patient care and monitoring system.
• The complete bio signal data acquisition system was coded in verilog con-
forming to RTL coding guidelines and successfully simulated using modelsim.

Fig. 6 a Hardware set up of scheme 1 and 2 multi PCM system, b DAC, c ADC FPGA
implementation
Development of Novel Architectures for Patient Care Monitoring … 341

• The designed system was synthesised, placed and routed with XCS 400-5ft256
FPGA. The real time system is capable of transmitting up to 32 numbers of bio
signal data of patients at 400,000 bits/s using a wireless communication channel.

7 Conclusion

• Bio signal DAQ has been designed for multi patient care and monitoring and is
suitable for FPGA implementation.
• Acquire four different bio signals such as EEG, EOG, EMG and temperature
from eight patients simultaneously using designed sensors.
• Conditioning circuits were also designed to meet the exacting specifications of
amplifying low amplitude bio signals.
• Novel algorithms have been developed for acquiring and monitoring the bio
signals for decentralized nurse stations. These algorithms were designed as
architectures and coded in Verilog.
• The analog signals from bio sensors were digitized by a “bio signal data
acquisition” module and were stored in “dual ram bank”. These were trans-
mitted serially through a controller module “zigbeec” to a CNS and the signals
acquired may be analyzed and diagnosed.

8 Scope for Future Work

Following are some of the suggestions related to this work, which can be under-
taken by enthusiastic researchers in near future.
• A DCNS was designed. The system can be further enhanced by designing and
integrating a lossless compression encoder before the XBee Pro Wireless
module.
• A Centralized Nurse Station can also be designed. It may also incorporate
compression decoder in its design.
• The patients’ beds may be made adjustable for flat, reclining or sitting posture
using motors controlled by patients themselves using eyewinks, Flex etc. or
even thought.
• Pressure ulcers or bedsores are one of the most common complications of
patients who cannot change positions in a bed all by themselves. By splitting the
bed into two halves, provide posture adjustment using motors, driver by
patients.
342 M.N. Mamatha

References

1. Gomez H, Camacho J, Yelicich B, Moraes L, Biestro A, Puppo C (2010) Development of a

multimodal monitoring platform for medical research. In: Annual international conference of
IEEE, engineering in medicine and biology society, Buenos Aires, Argentina, pp 2358–2361
2. Tanha AR, Popovic D (1997) An intelligent monitoring system for patient health care. In: 19th
international conference IEEE, engineering in medicine and biology society, Chicago, IL, USA,
pp 1086–1087
3. Reilent E, Loobas I, Pahtma R, Kuusik A (2010) Medical and context data acquisition system
for patient home monitoring. In: 12th biennial baltic electronics conference, Estonia,
pp 269–272
4. Zhan Y, Silvers CT, Randolph AG (2007) Real-time evaluation of patient monitoring
algorithms for critical care at the bedside. In: Proceedings of the 29th annual international
conference of the IEEE engineering in medicine and biology society cite international, France,
pp 2783–2786
5. Takizawa K, Li H-B, Hamaguchi K, Kohno R (2007) Wireless patient monitoring using IEEE
802.15.4a WPAN. In: IEEE, Singapore, pp 235–240
6. Dilmaghani RS, Bobarshad H, Ghavami M, Choobkar S, Wolfe C (2011) Wireless sensor
networks for monitoring physiological signals of multiple patients. IEEE Trans Biomed Circuits
Syst 5(4):347–356
7. Tafa Z, Stojanović R (2006) Bluetooth–based approach to monitoring biomedical signals. In:
Proceedings of the 5th WSEAS international conference on telecommunications and
informatics, Istanbul, Turkey, pp 415–420
8. Kogure Y, Matsuoka H, Kinouchi Y, Akutagawa M (2005) The development of a remote
patient monitoring system using java-enabled mobile phones. In: Proceedings of the 2005 IEEE
engineering in medicine and biology 27th annual conference, Shanghai, China, pp 2157–2260
Review on Biocompatibility of ZnO Nano
Particles

Ananya Barman

Abstract ZnO nano-particles have some unique properties like piezoelectric,

semiconducting, catalytic properties and antibacterial activities. Thus these particles
are widely used in optoelectronics, sensors, transducers, energy conversion and also
in medical sciences. Zinc oxide nano-particles have the potential to function as
natural selective killers of all highly proliferating cells, whether cancerous or not. The
application of zinc oxide nano-particles in cancer therapy looks intriguing and
exciting, specific tumor cell targeting will be essential (e.g., by nano-particle func-
tionalization with cell ligands) because these nano-particles are killers of all rapidly
proliferating cells, irrespective of their benign or malignant nature. So the cellular
level biocompatibility and biosafety of ZnO is studied here. Hela cell line showed a
complete biocompatibility to ZnO nanostructures from low to high NW concentra-
tions beyond a couple of production periods. The L929 cell line showed a good
reproduction behavior at lower NW concentration, but when the concentration was
close to 100 μg/ml, the viability dropped to *50 %. Our study shows the biocom-
petability and biosafety of ZnO NWs when they are applied in biological applications
at normal concentration range.

Keywords Zinc oxide nanoparticles
Dispersion
Selective cytotoxicity

Mesenchymal stem cells

1 Introduction

Nanostructures of zinc oxide (ZnO) have attracted much interest because of their
unique piezoelectric, semiconducting, and catalytic properties [1, 2] and a wide range
of applications in optoelectronics, sensors, transducers, energy conversion and
medical sciences [3–10]. Recently, utilizing the coupled piezoelectric and semi-
conducting properties, ZnO nanowire (NW) arrays and nanobelts have been applied

A. Barman (&)
Chemistry Department, JIS College of Engineering, 741235 Kalyani, Nadia, India
e-mail: [email protected]

© Springer India 2015 343

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_32
344 A. Barman

for converting mechanical energy into electricity and building piezotronic devices
[3, 5, 11].
Again ZnO nano particles have potential biomedical and lifescience applications.
zinc oxide nanoparticles offer significant benefits, and are used in several products
and systems, including sunscreens, biosensors, food additives, pigments, rubber
manufacture, and electronic aterials [1]. In the biomedical field, one of the most
important attributes of these nanoparticles is their antibacterial activity, with pub-
lished reports confirming the efficacy of zinc oxide nanoparticle-based preparations
as prophylactic agents against bacterial infections [11, 12].
Few reports had been published on the cytotoxicity of ZnO nano particles on
mammalian cells. Other reports suggest that these nanoparticles are nontoxic to
cultured human dermal fibroblasts [13] but exhibit toxicity towards neuroblastoma
cells [14] and vascular endothelial cells [15] and induce apoptosis in neural stem cells
[16]. It has been reported that nanoparticle size influences cell viability. Jones et al.
pointed out that zinc oxide particles of 8 nm in size were more toxic than larger zinc
oxide particles (50–70 nm) in Staphylococcus aureus [17]. Recently, Hanley et al.
observed that there is an inverse relationship between nanoparticle size and cyto-
toxicity in mammalian cells, as well as nanoparticle size and reactive oxygen species
production [18], while Deng et al. showed that zinc oxide nanoparticles manifested
dose-dependent, but no size-dependent, toxic effects on neural stem cells [16].
Many of the reported studies on zinc oxide nanoparticle toxicity have significant
limitations for two reasons. Firstly, some reports used zinc oxide nanoparticles with
different characteristics (size, shape, and purity) and, secondly, the dispersion
protocols used have often been unsuitable for biological use. Thus, the first
requirement for the biomedical use of zinc oxide nanoparticles is an effective
biocompatible protocol for aqueous dispersion.
The potential application of zinc oxide nanoparticles in medicine as a follow-on
from the report by Hanley et al. that these nanoparticles induce toxicity in a cell-specific
and proliferation-dependent manner. This group demonstrated that zinc oxide nano-
particles exhibit a strong preferential ability to kill rapidly dividing cancerous T cells
(28–35x) but not normal cells [19]. A recent report confirmed that these nanoparticles
exert a cytotoxic effect on human glioma cells, but not on normal human astrocytes. The
mechanisms of toxicity appear to involve the generation of reactive oxygen species,
with cancerous cells producing higher inducible levels than normal cells.

2 Materials and Methods

2.1 Preparation of ZnO Nano Particles

Due to its vast areas of application, various synthetic methods have been employed
to grow a variety of ZnO nanostructures, including nanoparticles, nanowires,
nanorods, nanotubes, nanobelts, and other complex morphologies [20–33].
Review on Biocompatibility of ZnO Nano Particles 345

There are many methods we have, to synthesize ZnO nano particles, like met-
allurgical process, mechanical, chemical, mechanochemical process, controlled
precipitation, sol-gel method and solvothermal and hydrothermal method. Mainly
sol-gel method or hydrothermal method present low cost and environment friendly
synthetic route, most of the literature for ZnO nano-particles is based on the
solution method. In addition, synthesis of ZnO nano-particles in the solution
requires a well defined shape and size of ZnO nano-particles.
In this regards, Monge et al. [34] reported room-temperature organometallic
synthesis of ZnO nano-particles of controlled shape and size in solution. The prin-
ciple of this experiment was based on the decomposition of organometallic precursor
to the oxidized material in air. It was reported [35] that when a solution of dic-
yclohexylzinc(II) compound [Zn(c-C6H11)2] in tetrahydrofuron (THF) was left
standing at room temperature in open air, the solvent evaporated slowly and left a
white luminescent residue, which was further characterized by X-Ray diffraction
(XRD) and transmission electron microscopy (TEM) and confirmed as agglomerated
ZnO nanoparticles with a zincite structure having lack of defined shape and size.
The ZnO NWs were grown by a vapor-liquid-solid (VLS) process in a horizontal
tube furnace, as reported previously [36, 37]. Gold nanoparticles were used as the
catalyst and the NWs were supported by a polycrystalline alumina substrate.
In some of the reported study 0.2 mmol of zinc nitrate hexahydrate, Zn (NO3)2·
6H2O, and 2 mmol oleic acid (OA) were dissolved in 20 mL triethylene glycol
(TREG). The mixture was loaded into a three-neck flask and was stirred at room
temperature for 30 min. The original colorless clear solution started to turn milk
white. Then, the mixture was heated to 240 °C with continuous stirring for 120 min.
A mass of bubbles appeared when the temperature was up to 150 °C, which
demonstrated the reaction of ester elimination. After being cooled to room tem-
perature and separated by centrifugation, the nanocrystals were first washed using
toluene and then washed using absolute ethanol to remove the unreacted molecules.
The final particles were collected by centrifugation and redispersed in deionized
water under agitation for further characterization.
Again 1.86 mmol of zinc acetate dihydrate, also Zn (Ac)2 · 2H2O, was dissolved in
10 mL ethanol in a flask under vigorous stirring and refluxed for 90 min at 68 °C. The
obtained Zn (Ac)2 solution was then cooled down to room temperature. 4.11 mmol of
KOH was dissolved in 5 mL of ethanol and kept in an ultrasonic bath for 40 min at
room temperature. The obtained KOH solution was slowly added to Zn (Ac)2 solution
and was stirred at room temperature for 60 min. Then 0.5 mL of deionized water and
0.34 mmol of 3-aminopropyltriethoxysilane (APTES) which was dissolved in 5 mL
ethanol were added into the reaction system. After 120 min of constant stirring at
room temperature, the precipitate was isolated by centrifugation. The nanocrystals
were first washed using toluene and then washed using absolute ethanol to remove the
unreacted molecules. The final particles were collected by centrifugation and re-
dispersed in de-ionized water under agitation for further characterization.
After synthesis, the ZnO nano-particles are characterized by DLS, UVand FT-IR
Spectra, SEM, TEM, X-RAY Diffraction, EDAX and the optical properties also
observed.
346 A. Barman

3 Biocompatibility of ZnO Nano Particles

3.1 Cell Culture and Interaction with Zno

For the biocompatible study, 5 mg of ZnO NWs were used as the test sample, which
were removed from the polycrystalline alumina substrate, then put into a poly-
propylene centrifuge tube, which was filled with 5 mL sterile phosphate buffer
solution (PBS, pH 7.2). The average diameter of the NWs is 1 µm and the average
length is 200 µm. The NWs were dispersed by ultrasonication for 10 min. Then
samples with concentrations of 1000, 100, 10, and 1 µg/ml, respectively, were
prepared (Fig. 1a).
Two cell lines from different origins of tissues were utilized [38]. One was Hela
cell line (American type Culture Collection, ATCC, CCL-2, Homo sapiens), a kind
of epithelial cell. The other one was L-929 cell line (ATCC, CCL-1, Mus mus-
culus), from subcutaneous connective tissue. The physical interaction of NWs and

Fig. 1 Effect of ZnO Nws on the growth and reproduction of cells as a function of time. a As-
grown ZnO nanowires on an alumina substrate. b Hela cell have been cultured for 4 h. c–h Hela
cells cultured with ZnO nanowires in solution after growing for 0, 6, 12, 18, 24 and 48 h
respectively. The Cells grew and reproduced with the presence of ZnO nanowires. Panels
b–h were recorded at the same magnification so that the number of cells in each image represents
its concentration. The scale bar is 100 μm
Review on Biocompatibility of ZnO Nano Particles 347

Fig. 2 SEM images of Hela cells on ZnO NWs arrays. a Two Hela cells are growing on the
surface of ZnO NWs arrays. b Cells are upheld by the NWs. Some ZnO NWs are phagocytosed
into the Hela cell (pointed out by the red arrow). The diameter and the length of the nanowires
are *100 nm and *1.5 μm respectively

Hela cells was investigated by inverted microscopy and scanning electron

microscopy (SEM). ZnO NWs array was in high density and the average diameter
and the length of the nanowires were *100 nm and *1.5 µm, respectively. After
seeding Hela cells on the ZnO NWs arrays, the substrate was incubated in 37 °C,
5 % CO2 for 24 h with DMEM cell culture medium. As the cells in the culture
medium subsided on the NWs substrate, the cells were upheld by ZnO NWs
(Fig. 2a). To maintain the morphology of the cells for SEM, immobilized process
will be actualized to the cells adhered on the substrate. SEM images of the samples
are shown in Fig. 2a, b. Figure 2a shows two Hela cells began to extend and grow
on the surface of ZnO NWs. Hela cells phagocytosed some of the broken ZnO NWs
into the cell membrane (Fig. 2b). No external force or inducement was necessary
for the phagocytosis, owing to the self-activity of the cell. The results indicate that
ZnO nanowires have fine biocompatibility and can be readily phagocytosed by the
cells. he Hela cells were still growing and reproducing even though the NWs were
directly in contact with them (Fig. 1c–h). Also, we can see the degradation of ZnO
NWs in cell culture medium with the time changing (Fig. 1c–h).
Some study were made on human neuroblastoma cell line and bone marrow cells
from tibia and femur. These cells are the biological model of proliferating cells. The
study demonstrated that zinc oxide nano-particles induce formation of excess
reactive oxygen species resulting in alteration and damage of cellular proteins,
DNA, and lipids, which can cause cell death [39–42]. We used carboxy-H2DCFDA
fluorescence as a reporter of intracellular oxidant production.
It was reported that initial evaluation of the biocompatibility of ZnO nano
particles was performed using heamolysis assays. This results further confirm the
biocompatibility [43].

4 Results and Discussion

ZnO NWs are biocompatible and biosafe to the two cell lines (Fig. 3). The vi-
abilities of Hela cells cultured with NWs for 12 and 24 h showed no difference. The
48 h cultured cells showed only a slight reduction in viability at a high concen-
tration of 100 µg/ml (Fig. 3a).
348 A. Barman

Fig. 3 Cell viability tested by MTT as a function of ZnO NW concentration and time. a Cell
viability tested by MTT method as a function of ZnO NWs for 12, 24, 48 h. b Viability of L929
cell line in MTT test, cultured with different concentrations of ZnO NWs for 12, 24, 48 h

More than 95 % of the Hela cells were alive after the test, and there was no
significant difference in viability among the plates of three time groups (SPSS,
Paired-sample t test) (Table 1). In the 48 h MTT experiment with the highest NW
concentration of 100 µg/ml, the viability of the Hela cells was a little lower than
that of the sample without NWs, but the viability was still more than 75 % (Fig. 3a).
Table 1 shows the statistical analysis of Paired-sample t tests using SPSS. If the
values of paired-sample t tests are larger than 0.05, it means that there is no
significant difference between the compared groups. Most of the values shown are
larger than 0.05, indicating adding ZnO NWs did not affect the viability of the cells.
For 96-well plates of planting L929 cells, the viabilities showed some variations
(Fig. 3b). For the 12 h plate, the viability of L929 cells was lower than those of the
other two 24 and 48 h time-sequence plates (Table 1), indicating that the L929 cell
was in a frail period and more sensitive to ZnO NWs when it was first cultured for
less than 12 h. After the culturing time exceeded 12 h, the viability of the cells
remained strong and was better than 95 % even at relatively high NWs concen-
tration. However, the viability of the cells dropped significantly when the NW
concentration reached 100 µg/ml. Therefore, the NWs are considered to be com-
pletely biocompatible and biosafe at NWs concentration lower than 100 µg/ml.

Table 1 Paired-sample
T-test, SPSS Hela/concentration (µ g/ml) 0.1 1 10 100
12 h Sig. (2 tailed) 0.087 0.103 0.541 0.059
24 h Sig. (2 tailed) 0.471 0.346 0.524 0.060
48 h Sig. (2 tailed) 0.124 0.736 0.131 0.000
L929/concentration (µ g/ml) 0.1 1 10 100
12 h Sig. (2 tailed) 0.039 0.155 0.007 0.001
24 h Sig. (2 tailed) 0.725 0.059 0.287 0.000
48 h Sig. (2 tailed) 0.182 0.545 0.142 0.000
Review on Biocompatibility of ZnO Nano Particles 349

Other zinc oxide nanoparticle dispersions have already been used for cytotox-
icity studies. Some published reports indicate that zinc oxide nanoparticles can be
toxic to mammalian cells [14–16]. Hanley et al. proposed that the mechanism of
zinc oxide nanoparticle toxicity involves the generation of reactive oxygen species.
These authors also suggested cell-specific behavior, with cancer cells producing
higher inducible levels of reactive oxygen species than their normal counterparts
following exposure to zinc oxide nano-particles [44].
Cell proliferation assays performed in a human neuroblastoma cell line dem-
onstrate that the cytotoxicity induced by zinc oxide nano-particles is dose-depen-
dent. No significant changes in cell viability of nano-particle concentrations of
10 µg/mL was observed. At these concentrations, we observed a spurious small
increase in viability compared with the control due to nanoparticle absorbance at
the working wavelength. Cell viability dropped significantly at a nanoparticle
concentration of 20 µg/mL, and was associated with production of reactive oxygen
species. Based on published work, we have developed a model of zinc oxide
nanoparticle cytotoxicity. Essentially the process involves internalization of the
nano-particles via receptor-mediated endocytosis, hydrolysis of the nanoparticles
via zinc ions within the lysosomes, and release of zinc ions into the cytosol.
When zinc oxide nano-particles are added to growth medium, the particles are
coated by proteins in the medium. Cell surface receptors bind the protein adsorbed
onto the nanoparticles, and the nano-particles enter cells via receptor-mediated
endocytosis. When nano-particle containing endosomes fuse with lysosomes, the
pH drops dramatically, approaching 5. The rate of zinc oxide hydrolysis was seen
as a function of pH; whereas under physiological conditions the fraction hydrolyzed
is negligible (0.02 %), the hydrolysis of zinc oxide is complete at pH 5.75. The zinc
ions induce lysis of the lysosomal membrane, and the ions are released into the
cytosol. Cytosolic accumulation of zinc ions triggers pathways which ultimately
cause cell death.
Specifically, Xia et al. demonstrated that zinc oxide dissociation disrupts the
cellular homeostasis of zinc, leading to lysosomal and mitochondrial damage, and
ultimately cell death by inhibiting cellular respiration through interference with
cytochrome bc1 in complex III and with α-ketoglutarate dehydrogenase in complex
I [39]. Other researchers have under-lined that zinc ion-mediated production of
reactive oxygen species promotes two important mechanisms, i.e., cytoplasmic
release of calcium ions and interaction with the cytoplasmic membrane, causing
loss of membrane integrity and leading to calcium influx through membrane
channels [40].
Over the last two decades, mesenchymal stem cell-based therapy has progressed
rapidly from preclinical to early clinical Phase I and II studies in a range of human
diseases (www.clinicaltrialgov). Osteocytes used in this work were obtained via
osteogenic differentiation of mesenchymal stem cells. These two populations,
although biologically related, show opposite proliferation rates. We evaluated cell
death induced by zinc oxide nano-particles using flow cytometry. The increased
fluorescent intensity in the presence of zinc oxide nano-particles was more sig-
nificant in mesenchymal stem cells than in the differentiated osteogenic lineage.
350 A. Barman

The different behavior of the two cell types is likely due to the different cellular
interaction with zinc oxide nano-particles rather than to any differences in uptake of
the nano-particles. The strong interaction between osteocytes and the nano-particles
is reflected by the parallel strong increase of side scatter compared with the control
cultures, indicative of increased cellular complexity due to nano-particle cell
interaction [45].
Based on these findings, we confirm that zinc oxide nano-particles have the
potential to function as natural selective killers of all highly proliferating cells,
whether cancerous or not. In conclusion, although the application of zinc oxide
nanoparticles in cancer therapy looks intriguing and exciting, specific tumor cell
targeting will be essential (e.g., by nano-particle functionalization with cell ligands)
because these nano-particles are killers of all rapidly proliferating cells, irrespective
of their benign or malignant nature.

5 Conclusion

In conclusion all these studies show the biocompatibility and biosafety of ZnO nano
particles when they are applied in biological applications at normal concentration
range. This is an important conclusion for their applications in in vivo biomedical
science and engineering. The threshold of intracellular ZnO NP concentration
required to induce cell death in proliferating cells is 0.4 ± 0.02 mM. Finally, flow
cytometry analysis revealed that the threshold dose of zinc oxide nano-particles was
lethal to proliferating pluripotent mesenchymal stem cells but exhibited negligible
cytotoxic effects to osteogenically differentiated mesenchymal stem cells. These
results confirm the ZnO NP selective cytotoxic action on rapidly proliferating cells,
whether benign or malignant.
In addition, relatively small size, ease of transport within tissues/organs, ability
to cross plasma membranes, and potential targeting of biologically active molecules
will facilitate biomedical applications of nanoparticles in the field of medicine.

References

1. Lieber CM, Wang ZL (2007) MRS Bull 32:99–104

2. Norton DP, Heo YW, Ivill MP, Pearton SJ, Chisholm MF, Steiner T (2004) Mater Today 7
(6):34–40
3. Wang ZL, Song JH (2006) Science 312:242–246
4. Law M, Sirbuly D, Johnson J, Goldberger J, Saykally R, Yang P (2004) Science 305:1269
5. Wang XD, Song JH, Liu J, Wang ZL (2007) Science 316:102–105
6. Johnson JC, Yan HQ, Schaller RD, Petersen PB, Yang PD, Saykally RJ (2002) Nano Lett 2
(4):279–283
7. Arnold MS, Avouris P, Pan ZW, Wang ZL (2003) J Phys Chem B 107(3):659–663
8. Lao CS, Kuang Q, Wang ZL, Park MC, Deng Y (2007) Appl Phys Lett 90:262107
Review on Biocompatibility of ZnO Nano Particles 351

9. Dorfman A, Kumar N, Hahm S (2006) J Adv Mater 18:2685

10. Liu TY, Liao HC, Lin CC, Hu SH, Chen SY (2006) Langmuir 22(13):5804–5809
11. Qin Y, Wang XD, Wang ZL (2008) Nature 451:809–813
12. Huang ZB, Zheng X, Yan DH et al (2008) Toxicological effect of ZnO nano-particles based on
bacteria. Langmuir 24(8):4140–4144
13. Reddy KM, Feris K, Bell J, Wingett DG, Hanley C, Punnoose A (2007) Selective toxicity of
zinc oxide nanoparticles to prokaryotic and eukaryotic systems. Appl Phys Lett 90
(21):2139021–2139023
14. Jeng HA, Swanson J (2006) Toxicity of metal oxide nanoparticles in mammalian cells.
J Environ Sci Health A Tox Hazard Subst Environ Eng 41(12):2699–2711
15. Gojova A, Guo B, Kota RS, Rutledge JC, Kennedy IM, Barakat AI (2007) Induction of
inflammation in vascular endothelial cells by metal oxide nanoparticles: effect of particle
composition. Environ Health Perspect 115(3):403–409
16. Deng XY, Luan QX, Chen WT et al (2009) Nanosized zinc oxide particles induce neural stem
cell apoptosis. Nanotechnology 20(11):115101
17. Jones N, Ray B, Ranjit KT, Manna AC (2008) Antibacterial activity of ZnO nanoparticle
suspensions on a broad spectrum of microorganisms. FEMS Microbiol Lett 279(1):71–76
18. Hanley C, Thurber A, Hanna C, Punnoose A, Zhang J, Wingett DG (2009) The influences of
cell type and ZnO nanoparticle size on immune cell cytotoxicity and cytokine induction.
Nanoscale Res Lett 4(12):1409–1420
19. Hanley C, Layne J, Punnoose A et al (2008) Preferential killing of cancer cells and activated
human T cells using ZnO nanoparticles. Nanotechnology 19(29):295103
20. Song RQ, Xu AW, Deng B, Li Q, Chen GY (2007) Adv Funct Mater 17:296
21. Gao PX, Ding Y, Mai W, Hughes WL, Lao C, Wang ZL (2005) Science 309:1700
22. Hu P, Liu Y, Wang X, Fu L, Zhu D (2003) Chem Commun 50:1304
23. Zhang J, Sun L, Liao C, Yan C (2002) Chem Commun 3:262
24. Tang Q, Zhou W,Shen J, Zhang W, Kong L, Qian Y (2004) Chem Commun 10(6):2004
25. Greene LE, Law M, Goldberger J, Kim F, Johnson JC, Zhang Y, Saykally RJ, Yang P (2003)
Angew Chem 42:3031
26. Kong XY, Ding Y, Yang R, Wang ZL (2004) Science 303:1348
27. Qian D, Jiang JZ, Hansen PL (2003) Chem Commun 9(9):1078
28. Zhong X, Knoll W (2005) Chem Commun 9(9):1158
29. Li F, Ding Y, Gao P, Xin X, Wang ZL (2004) Angew Chem 43:5238
30. Ding GQ, Shen WZ, Zheng MJ, Fan DH (2006) Appl Phys Lett 88:103106
31. Shpeizer BG, Bakhmutov VI, Clearfield A (2006) Micro Meso Mater 90:81
32. Wang X, Summers CJ, Wang ZL (2004) Adv Mater 16:1215
33. Polarz S, Orlov AV, Schüth F, Lu AH (2007) Chem Eur J 13:592
34. Monge M, Kahn ML, Maisonnat A, Chaudret B (2003) Angew Chem Int Ed 42:5321
35. Carnes CL, Klabunde KJ (2000) Langmuir 16:3764
36. Wang XD, Summers CJ, Wang ZL (2004) Nano Lett 4:423–426
37. Yang P, Yan H, Mao S, Russo R, Johnson J, Saykally R, Morris N, Pham J, He R, Choi H
(2002) Adv Func Mater 12:323
38. Hanks CT, Wataha JC, Sun ZL (1996) Dent Mater 12:186–193
39. Xia T, Kovochich M, Liong M et al (2008) Comparison of the mechanism of toxicity of zinc
oxide and cerium oxide nanoparticles based on dissolution and oxidative stress properties.
ACS Nano 2(10):2121–2134
40. Huang CC, Aronstam RS, Chen DR, Huang YW (2010) Oxidative stress, calcium
homeostasis, and altered gene expression in human lung epithelial cells exposed to ZnO
nanoparticles. Toxicol In Vitro 24(1):45–55
41. Florianczyk B, Trojanowski T (2009) Inhibition of respiratory processes by overabundance of
zinc in neuronal cells. Folia Neuropathol 47(3):234–239
352 A. Barman

42. Gazaryan IG, Krasnikov BF, Ashby GA, Thorneley RNF, Kristal BS, Brown AM (2002) Zinc
is a potent inhibitor of thiol oxidoreductase activity and stimulates reactive oxygen species
production by lipoamide dehydrogenase. J Biol Chem 277(12):10064–10072
43. Dobrovolskaia MA, Clogston JD, Neun BW, Hall JB, Patri AK, McNeil SE (2008) Nano Lett
8:2180–2187
44. Wang ZL (2004) Zinc oxide nanostructures: growth, properties and applications. J Phys
Condens Matter 16(25):R829–R858
45. Cai D, Blair D, Dufort FJ et al (2008) Interaction between carbon nano-tubes and mammalian
cells: characterization by flow cytometry and application. Nanotechnology 19(34):1–10
Tailoring Characteristic Wavelength
Range of Circular Quantum Dots
for Detecting Signature of Virus in IR
Region

Swapan Bhattacharyya and Arpan Deyasi

Abstract Characteristic wavelength carrying the signature of virus is analytically

determined through its match with the radiating wavelength comes out from
computation of intersubband transition energies of different circular quantum dots,
namely quantum ring and quantum disk. Time-independent Schrödinger equation is
solved subject to the applied electric field along the axis, and first and second order
Bessel functions are considered for computation of energy subbands. Non-mono-
tonic spacing of quantized energy states have been observed by changing different
dimensions of the quantum dots. Three lowest confinement states along with
subband energies are plotted with different structural parameters, and also with
external field. Comparative study reveals that better tuning of intersubband tran-
sition energy can be achieved in quantum ring than quantum disk having similar
structural parameters; which reveals the fact that characteristic wavelength from
quantum ring can track wider rage of virus signature. Tailoring of wavelength can
be revealed by notifying the blueshift/redshift in absorption spectra in the choice of
frequency region.

Keywords Characteristic wavelength Quantum ring Quantum disk

Electric



field Intersubband transition energy Signature of virus

S. Bhattacharyya (&)
Dept of Electronics and Communication Engineering, JIS College of Engineering,
Kalyani 741235, India
e-mail: [email protected]
A. Deyasi
Dept of Electronics and Communication Engineering, RCC Institute of Information
Technology, Kolkata 700015, India
e-mail: [email protected]

© Springer India 2015 353

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_33
354 S. Bhattacharyya and A. Deyasi

1 Introduction

Low dimensional quantum structures have already attracted the interest of

researchers due to its possibility in designing novel electronic [1–3] and optoelec-
tronic [4–6] devices which have been physically realizable due to the advancement of
growth techniques [7, 8]. Quantum well, wire and dot are the new devices where
carrier motion is restricted along one, two or three dimensions so that it becomes
comparable with de- Broglie wavelength, and it leads to the quantization of energy
levels. Different geometrical variations of quantum dots are already proposed for
computation of eigenstates for specific application purposes [9, 10], among which the
ring and disk-shaped dots attract special attention because they are proposed to be
utilized in designing Qubit [11], to realize the dream of quantum computers [12].
A ring-shaped quantum dot may be termed as quantum ring, which can also be
considered as a “hole” created at the middle of a circular quantum dot; whereas a
solid cylindrical-shaped one-dimensional quantum dot may be termed as quantum
disk. This particular class of circular quantum dot attracted quantum physicists due
to the possibility of experimental observations of the Aharonov-Bohm effect [13].
Kish et al. [14] demonstrated the candidature of heterostrucutre quantum ring as
laser with good current confinement and photon confinement. Han et al. [15]
experimentally investigated the same with square geometry. Filikhin et al. [16]
calculated the quantized energy levels of semiconductor quantum ring using non-
linear Schrödinger equation with the consideration of energy-dependence of
effective mass. Llorens et al. [17] also calculated electronic states of a quantum ring
under lateral electric field and compared it with the results of a quantum disk.
Energy level of the disk is also computed by Lamouche and Lépine [18] in absence
of any field. Peeters and Schweigert [19] calculated the same in presence of
magnetic field. Hassanien et al. [20] calculated the same for parabolic potential
profile in presence and absence of electric field. Kikuchi et al. [21] fabricated
heterostructure quantum disk for designing novel LED. Susa [22] theoretically
established that quantum disk can be utilized in designing distributed feedback laser
with very precise control of threshold current density.
Knowledge of optical transition between two subbands can be used in special-
ized applications e.g., non-invasive medical diagnosis, environment monitoring etc.
It can also be utilized to monitor and control food quality. The characteristic
wavelength of a few viruses generally lies in the IR range; hence its identification
can only be possible with its proper matching with the radiating/absorbing wave-
length of similar range from a device. The most important feature of these appli-
cations is the characteristic wavelength due to the intersubband transition energy,
originated due to the transition form a particular quantum state to another quantum
state. This, in turn, depends on quantized states; which are function of structural
parameters of the device. Addition of field in precise direction will add flexibility to
the purpose as it can help to tune the emitted/detected wavelength in either of
blueshift/redshift directions. Hence computation of intersubband transition plays the
vital role in determining signature of the virus.
Tailoring Characteristic Wavelength Range … 355

In this paper, characteristic wavelength range of two different circular quantum

dots is analytically computed from the result of intersubband transition energy.
Eigenstates are obtained by solving time-independent Schrödinger equation. Elec-
tric field is applied along the axis of the structure, and radius as well as thickness is
varied to observe the effect on lowest three quantization states. Quantum ring and
quantum disk are considered for vis-à-vis comparative analysis of eigenstates with
variation of structural parameters, and also with electric field up to a moderate
range. Dimensions and field are chosen in such a way transition energies lay in IR
region, so that signature of virus fall inside that range can suitably be traced out.

2 Mathematical Modeling

2.1 Calculating Eigenstate of Quantum Disk

We consider the schematic structure of a cylindrical quantum disk of radius ‘b’ and
thickness ‘d’ as shown in Fig. 1a under external electric field (F) applied perpen-
dicular to the plane of the disk. Using the cylindrical co-ordinate system, the time-
independent Schrödinger equation for the electron wavefunction W can be written as

 

2 1 @
h @W 1 @2W @2W
 q þ 2 2 þ 2  eF W ¼ E W ð1Þ
2 m q @q @q q @q @z

where ρ ≤ b and 0 ≤ z ≤ d, F being the externally applied electric field, E is the

energy eigenvalue, m* is the effective mass of electron, e the electronic charge and
ħ the Dirac constant. Here we assume parabolic band structure.
The left hand side of Eq. (2) may be conceived as effective Hamiltonian operator
(Ĥ) acting on wavefunction ψ. We can break up the Hamiltonian operator (Ĥ) into
two parts as

H ^ 0 þ H^0
^ ¼H ð2Þ

The above equation satisfies the condition

^ 0 W0 ¼ E0 W0
H ð3Þ

where

 
_ h
2 1 @ @W 1 @2W @2W
H0 ¼  q þ þ ð4Þ
2 m q @q @q q2 @q2 @z2

where E0 and ψ0 are the unperturbed values of energy eigenvalue and the wave-
function respectively.
356 S. Bhattacharyya and A. Deyasi

Fig. 1 a Schematic structure

of a cylindrical quantum disk,
b Cylindrical quantum ring

Choosing Ψ0 = R(ρ) Θ (θ)Z(z) and applying the method of separation of vari-

ables, the θ-dependent component of the wavefunction Θ has solution given by

H  expðimhÞ ð5Þ

for m = 0, 1, 2,…
The ρ-dependent component of the wavefunction, R, can be determined by
solving the differential equation

@ 2 R @R
q2 þ þ ðk2 q2  m2 ÞR ¼ 0 ð6Þ
@q2 @q

for 0 ≤ ρ ≤ b, where

2 m ðE0  En Þ
k2 ¼ ð7Þ

h2
Tailoring Characteristic Wavelength Range … 357

En is the quantized energy due to motion in the z-direction. The general solution
of Eq. (7) is obtained in terms of m-th order Bessel functions Jm(λ) and Ym(λ) of
first kind and second kind respectively. With the application of boundary condi-
tions, R = 0 at ρ = 0 (ρ = b), it can be seen that only some discrete values of λ are
allowed. These discrete values can be obtained by solving the following
determinants
 
 Jm ðk1 Þ Ym ðk1 Þ 
 ¼0 ð8Þ
 Jm ðk2 Þ Ym ðk2 Þ 

where λ1(λ2) corresponds to the value of λ at ρ = 0 (ρ = b).

The solution of (9) requires two quantum numbers, m and l (say), which indi-
cates the l-th zero (solution) of the m-th order Bessel function. Thus, the unper-
turbed energy eigenvalue E0 can be obtained using Eq. (8) and, the energy
eigenvalue is quantized with three quantum numbers (m, l, n).
The exact eigenvalue can be approximately derived first by operating Ĥ on the
wavefunction Ψ0 and subsequently by using the relation

E 0 ¼ E0 þ Hss0 ð9Þ

where
R _
W0s H 0 W0s dV
Hss0 ¼ V
R  ð10Þ
W0s W0s dV
V

W0s being the complex conjugate of the unperturbed wavefunction W0s of state
‘s’. Using various substitutions and simplifications, the energy eigenvalue of
electron in the disk is written as
2 3

h2 np2 eFd

16
6 2 ms d 2 ffi7
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
7
Ezn ¼ 6 7 ð11Þ
24
h2 np2
h2 np2 5
þ  
 eFd
2m d 2m d

2.2 Calculating Eigenstate of Quantum Ring

We consider the schematic structure of the cylindrical quantum ring of inner radius
‘a’, outer radius ‘b’ and thickness ‘d’ as shown in Fig. 1b. Using the cylindrical co-
ordinate system, the time-independent Schrödinger equation for the electron wave-
function W is same as Eq. (1) where the domain is changed as a ≤ ρ ≤ b and 0 ≤ z ≤ d.
358 S. Bhattacharyya and A. Deyasi

The final expression of energy eigenvalue of quantum ring with approximation

of not very large F is given by

h2 2
Elmn ¼ k
2 m2 ml 3

h2 np2 eFd
6 2 m d  7 ð12Þ
16
6 s
2
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
7
7
þ 6 7
24
h2 np2
h2 np2 5
þ  eFd
2 m d 2 m d

3 Results and Discussion

Using Eqs. (11) and (12), the energy eigenvalues of an n-GaAs quantum disk and
ring are computed as function of different structural parameters and also of external
electric field.
Figure 2 shows the variation of eigenenergies for the lowest three states with
outer diameter of the disk in presence and absence of electric field. From the plot, it
is observed that by increasing the diameter of the disk, energy monotonically
decreases. This is due to the fact that with increase in dimension, quantum con-
finement decreases, which lowers the eigenenergy. The rate of decrement is sig-
nificant with lower diameter, whereas the rate reduces with higher size of the disk.
Application of the axial electric field further lowers the eigenvalue for same
dimension. Similar feature may be observed for quantum ring when plotted with

Fig. 2 Lowest three energy 60

states with diameter of a=10(nm), d=20(nm)-QD
quantum disk in presence and -------- EL=0
_____ EL=200kV/m
absence of electric field 50
______ E
111
E 112
Energy(meV)

E 113
40

30

20

10
20 25 30 35 40
b(nm)
Tailoring Characteristic Wavelength Range … 359

Fig. 3 Lowest three energy 90

states with outer diameter of a=10(nm), d=20(nm)-QR
quantum ring -------- EL=0
_____ EL=200kV/m
70 --------- E111
E112

Energy(meV)
E113

50

30

10
20 25 30 35 40
b(nm)

Fig. 4 Comparative study of 90

lowest three energy states a=10(nm)-QR
80 d=20(nm)-QD,QR
with outer diameter of the ring
and disk in presence of -------- Disk _____ E111
70 _____Ring
electric field E112
EL=100kV/m E113
60
Energy(meV)

50

40

30

20

10

0
20 25 30 35 40
b(nm)

thickness, as shown in Fig. 3. The notable feature observed from the plot is that the
curvature of graph for any eigenstate remains same in absence of field compared to
the plot generated in presence of field.
Figure 4 shows the comparative analysis for lowest three eigenstates with outer
diameter for quantum disk and ring having similar dimensions in presence of axial
electric field. It may be seen form the result that higher magnitude of eigenstates can
be obtained form quantum ring than that of the disk. The reason behind the dif-
ference in magnitude for any eigenstate is that the ring possesses air dielectric in the
core region which has lower dielectric constant than the solid GaAs material present
in the disk. By virtue of the heterostructure formed between air and semiconductor,
quantum confinement becomes higher for the ring than that of the disk. The
360 S. Bhattacharyya and A. Deyasi

Fig. 5 Comparative study of 35

intersubband transition energy a=10(nm)-QR
d=20(nm)-QD,QR
with outer diameter of the ring 30
and disk in presence of -------- Disk ______ ΔE
21
_____ Ring
electric field 25
ΔE31
EL=100kV/m ΔE32

ΔEn(meV)
20

15

10

0
20 25 30 35 40
b(nm)

Fig. 6 Comparative study of 25

lowest three energy states a=10(nm)-QR
with thickness of the ring and b=40(nm)-QD,QR

disk for constant electric field 20 - - - - - - QDisk _____ E111

______ QRing
E112
EL=100kV/m E113
Energy (meV)

15

10

0
20 25 30 35 40
d(nm)

difference is significant for lower diameter, but the gap decreases with higher
dimension. Corresponding intersubband transition energy is plotted in Fig. 5.
From the plot, it is observed that transition energy variation is significant for
quantum ring than that of quantum disk, which speaks about the fact that quantum
ring can be used as a better optoelectronic transmitter than that of disk with similar
shape and size in presence of external electric field.
Similar comparative analysis is made between ring and disk with thickness of
the structure, shown in Fig. 6. From the plot, it is observed that with increase of
thickness, energy monotonically decreases. The rate is nonlinear for very lower
thickness value, but it becomes almost linear thereafter. The physics behind the
nature of energy profile is identical to the previous explanation, and so is not
Tailoring Characteristic Wavelength Range … 361

Fig. 7 Comparative study of - - - - - - QDisk

_____ E
111
lowest three energy states ______ QRing E112
E113
with electric field in the 24
n-GaAs quantum ring and
disk

Energy (meV)
20

16

a=10(nm)-QR
b=40(nm)-QD,QR
d=20(nm)-QD,QR
12
0 50 100 150 200
EL(kV/m)

Fig. 8 Comparative study of 8

intersubband transition energy
with electric field in the _____ ΔE a=10(nm)-QR
21
n-GaAs quantum ring and ΔE31 b=40(nm)-QD,QR
6 ΔE32 d=20(nm)-QD,QR
disk
- - - - - - QRing
______ QDisk
ΔEn(meV)

0
0 50 100 150 200
EL(kV/m)

mentioned again to avoid the repetition. Hence intersubband transition energies

calculated for the lowest three states become almost constant throughout the range
of interest which indicates the fact that tuning is not possible in this case.
Longitudinal electric field adds further dimension for tuning of energy for both
the structures. By increasing the strength of field, it may be seen that energy
decreases monotonically, as shown in Fig. 7. The rate of reduction is almost linear
in low-to-medium field range. Thus eigenstates of circular quantum dots can be
externally tuned by applying suitable electric field. Variation of intersubband
energy is plotted in Fig. 8. From the figure, it may be concluded that transition
energy slowly increases with increase of electric field, which helps electrical tuning
of optoelectronic properties for lasing applications.
362 S. Bhattacharyya and A. Deyasi

4 Conclusion

Characteristic wavelength range of quantum ring and quantum disk are determined
by analytically computing intersubband transition energies. Different structural
parameters are varied along with low-to-moderate axial electric field to calculate the
range. Comparative studies are made considering lowest three eigenstates of the
two quantum dots of similar size subject to similar external conditions, which
reveals the fact that quantum ring provides higher magnitude of eigenenergies than
quantum disk. Thus intersubband transition energy can be better varied in case of
quantum ring than the disk over a very small range of dimensional change. Effect of
electrical tuning is quite similar on these two quantum dots within our choice of
interest. Results help to design nanostructure optical detector in required wave-
length range for detecting signature of virus which emits characteristic wavelength
within the region of choice. Appropriate tuning of dimensions of the structure and
axial electric field helps to find the exact signatory wavelength of the virus.

References

1. Zhuang L, Guo L, Chou SY (1998) Silicon single-electron quantum-dot transistor switch

operating at room temperature. Appl Phys Lett 72:1205–1207
2. Pigorsch C, Wegscheider W, Klix W, Stenzel R (1997) 3D-Simulation of novel quantum wire
transistor. Phys Status Solidi B 204:346–349
3. Urban D, Braun M, König J (2007) Theory of a magnetically controlled quantum-dot spin
transistor. Phys Rev B 76:125306
4. Cui D, Jian X, Sheng-Yong X, Paradee G, Lewis BA, Gerhold MD (2006) Infrared
photodiode based on colloidal PbSe nanocrystal quantum dots. IEEE Trans Nanotechnol
5:362–367
5. Gvozdic DM, Schlachetzki A (2005) Modulation response of V-groove quantum-wire lasers.
IEEE J Quantum Electron 41:842–847
6. Kunz M et al (2005) High-speed quantum dot lasers and amplifiers for optical data
communication. Appl Phys A 80:1179–1182
7. Somaschini C, Bietti S, Sanguinetti S, Koguchi N, Fedorov A (2010) Self-assembled GaAs/
AlGaAs coupled quantum ring-disk structures by droplet epitaxy. Nanotechnology 21:125601
8. Kuramochi E, Temmyo J, Kamada H, Tamamura T (1999) Spatial ordering of self-organized
InGaAs/AlGaAas quantum disks on GaAs (311)B substrates. J Electron Mater 28:445–451
9. Li Y, Voskoboynikov O, Lee CP, Sze SM (2001) Electron energy state dependence on the
shape and size of semiconductor quantum dots. J Appl Phys 90:6416–6420
10. Li Y, Voskoboynikov O, Lee CP, Sze SM (2001) Computer simulation of electron energy
levels for different shape InAs/GaAs semiconductor quantum dots. Comput Phys Commun
141:66–72
11. Dong QR, Li SS, Niu ZCh, Feng SL, Zheng HZ (2004) Electronic structure of self-assembled
InAs quantum disks in an axial magnetic field and two-electron quantum-disk qubit. J Appl
Phys 96:3277–3281
12. Loss D, DiVincenzo DP (1998) Quantum computation with quantum dots. Phys Rev A
57:120–126
13. Mühle A, Wegscheider W, Haug RJ (2007) Coupling in concentric double quantum rings.
Appl Phys Lett 91:133116
Tailoring Characteristic Wavelength Range … 363

14. Kish FA, Caracci SJ, Maranowski SA, Holonyak N, Smith SC, Burnham RD (1992) Planar
native-oxide AlxGa1−xAs-GaAs quantum well heterostructure ring laser diodes. Appl Phys
Lett 60:1582–1584
15. Han H, Forbes DV, Coleman JJ (1995) InGaAs-AlGaAs-GaAs strained-layer quantum-well
heterostructure square ring lasers. IEEE J Quantum Electron 31:1994–1997
16. Filikhin I, Vlahovic B, Deyneka E (2006) Modeling of InAs/GaAs self-assembled
heterostructures: quantum dot to quantum ring transformation. J Vac Sci Technol A: Vac
Surf Films 24:1249–1251
17. Llorens JM, Trallero-Giner C, García-Cristóbal A, Cantarero A (2002) Energy levels of a
quantum ring in a lateral electric field. Microelectron J 33:355–359
18. Lamouche G, Lépine Y (1995) Ground state of a quantum disk by the effective-index method.
Phys Rev B 51:1950–1953
19. Peeters FM, Schweigert VA (1996) Two-electron quantum disks. Phys Rev B 53:1468–1474
20. Hassanien HH, Abdelmoly SS, Elmeshad N (2006) Exact solution of finite parabolic potential
disk-like quantum dot with and without electric field. FIZIKA-A 15:209–218
21. Kikuchi A, Kawai M, Tada M, Kishino K (2004) InGaN/GaN Multiple quantum disk
nanocolumn light-emitting diodes grown on <111> Si substrate. Jpn J Appl Phys 43:L1524–
L1526
22. Susa N (1998) Feasibility study on the application of the quantum disk to the gain-coupled
distributed feedback laser. IEEE J Quantum Electron 34:1317–1324
Methodology for a Low-Cost Vision-Based
Rehabilitation System for Stroke Patients

Arpita Ray Sarkar, Goutam Sanyal and Somajyoti Majumder

Abstract Stroke is a life threatening phenomenon throughout the world caused due
to the blockage (by clots) or bursting of arteries. As a result permanent or semi-
permanent neurological damage may occur that requires proper rehabilitation to
overcome the deficiency of communication or communication disorder of the stroke
patient with the outer world. This causes delay in overall recovery and affects the
general hygiene of the patient. Computer vision based interaction using gazes may
be helpful for such cases. In all those methodologies as a mandatory step eye
tracking has to be performed. Present work uses a low cost web camera for eye
tracking using Haar feature-based cascade function in comparison with the costlier
eye tracking systems available in the market. This method easily detects the eye
balls from the video online with less computational load. Several experiments have
been carried out to evaluate the performance in different background, lighting
conditions and quality of images.

Keywords Computer vision
Object detection
Face recognition
Patient

rehabilitation

A.R. Sarkar (&)
G. Sanyal

Department of Computer Science and Engineering, National Institute of Technology,
Durgapur, India
e-mail: [email protected]
G. Sanyal
e-mail: [email protected]
S. Majumder
Surface Robotics Lab, CSIR—Central Mechanical Engineering Research Institute,
Durgapur, India
e-mail: [email protected]

© Springer India 2015 365

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_34
366 A.R. Sarkar et al.

1 Introduction

The term stroke is well known to the world, especially to the aged people, as it is
the second common cause of death throughout the world [1]. Stroke is caused due
to bursting or blockage of arteries by clots [2]. Stroke leads to malfunction of the
brain as the arteries that carry oxygen and nutrients are damaged. As a result, either
death or permanent neurological damage will occur based on the severity and
location of the damage. Research has proved that different parts of the brain are
responsible for different sensory and motor functions. Damage to a particular
location will cause a disability in respective function. Patient may lead almost
normal life after undergoing some therapies, if the stroke is not severe. But constant
support from the family and friends and intensive rehabilitation by the healthcare
professionals are required, if the stroke is severe and the disability is permanent.
Stroke affects physical, cognitive and emotional functioning. The followings are the
most common after-effects of stroke found in the patients [2].
• Vascular Dementia: Loss of thinking ability
• Aphasia: A communication disorder
• Memory: Short-term or long-term memory loss
• Depression: Biological, behavioural or social factors cause such depression
• Pseudobulbar Affect (PBA): A medical condition which causes sudden crying or
laughing
All the above disabilities in terms lead to cause lack of communication or
communication—disorder between the patient and outer world including family
members, attending healthcare professionals and doctor. Lack or limited commu-
nication delays the recovery of the patient as well as affects the general hygiene of
the patient.
This situation can be overcome by introducing an automatic rehabilitation device
using computer vision-based system which will provide 24 h constant support.
Such system may be operated by the patient either through gazes or gestures, if
possible or through only nodding head or moving the iris. The input system for such
devices can be customized based on the type and degree of disability of the patient.

2 State-of-the-Art

Each year approximately 20 million people are affected by stroke and out of them
only 15 million people survive [3]. These lucky 15 million people may need a
communication system for rehabilitation. Several research organizations, academic
institutions, professional companies are engaged in developing low-cost rehabili-
tation systems for stroke patients. Most of these systems use mainly vision-based
methodology. Some other devices use robotic systems or tactile sensing. Successful
Methodology for a Low-Cost Vision-Based Rehabilitation … 367

attempts have been made to use smart phones, tablets and other similar gadgets to
implement those rehabilitation systems. Jack et al. has developed a desktop PC
based virtual reality system for rehabilitating hand function for the stroke patients
[4]. The proposed system uses a Cyber Glove and a Rutgers Master II-ND (RMII)
force feedback gloves as input devices. An efficient gesture based system for
impaired dexterity has been developed by Ushaw et al. using a tablet [5]. Sucar
et al. has presented the development of a cheaper vision-based system for intensive
movement training [6]. Reinkensmeyer et al. have proposed a web-based rehabil-
itation system that eliminates the always-present therapist and it consists of web-
based library of status test, several therapy games as well as progress charts [7].
As already mentioned robotic therapy is also popular for rehabilitation for
chronic stroke patients [8–10]. MANUS is such a specially designed robot devel-
oped by MIT being used for rehabilitation of stroke patients [8]. Another portable,
low-cost robotic system has been developed by Huq et al. for post-stroke upper
limb rehabilitation [9, 10]. Most of these robots are commonly used in place of a
human therapist. They use easy and interactive GUI for both manual and automatic
control. Some of them have force feedback options for better usage.
University of Michigan has successfully developed different support systems
using iPhone and iPad to overcome various types of communication challenges for
the people with aphasia [11]. These support systems include text-to-speech con-
verter, talking picture dictionary, phonemic cues, speech-to-text converter, video
calls, talking dictionary system etc. and these can be easily installed on smart
phones and tablets. Microsoft Kinect sensor is well known for its use in gaming and
equipped with a depth sensor and camera. Research team from Roke Manor
Research Ltd in association with Southampton University has developed a gesture
recognition software using Microsoft Kinect sensor for supervision of patients with
stroke by a physiotherapy clinic over the internet [12].
As the first step, most of the above mentioned devices/systems use suitable eye
detection and tracking technology. Some researchers are using commercially
available eye tracker system for the purpose and developing own (software)
modules to upgrade the performance [13]. A few of the popular commercial eye
trackers are Mirametrix S2 eye tracker [14], EyeTech VT2 [15], Gaze Point GP3
desktop eye tracker [16], Grinbath Eye Guide Mobile Tracker System [17], SMI
eye tracking glasses [18], Tobii eye tracking glasses [19]. These systems include
both hardware and software as a single package. The approximate costs (excluding
shipping and handling charges, taxes) of the popular eye tracking systems are
presented in the Table 1.
Every year, it has been reported 90–222 cases of stroke per 100,000 in India [1].
Such costlier systems for rehabilitation purpose are difficult to afford by the com-
mon people. If a low-cost system can be designed and developed with the available
cheapest components/technologies, a large number people will be benefitted. At
first stage, an efficient eye tracking system has been developed that will help the
stroke patient to indicate his/her requirements through the icons displayed on a
368 A.R. Sarkar et al.

Table 1 Table styles

Popular commercial eye tracking systems Price
Mirametrix S2 eye tracker $4,995
EyeTech VT2 $5,000
Gaze point GP3 eye tracker with gazepoint analysis $1,395
Grinbath complete eye guide® mobile tracker system $8,999
SMI eye tracking glasses 2.0 $30,800
Tobii eye tracking glasses $15,800

wall-mounted monitor/TV/display placed in front of the patient to his/her family

members/attendant through gazes/gestures. Initially a low resolution camera of US$
8 was used and subsequently a high resolution HD camera of approximate US$ 50
[INR 3000] and a desktop PC of approximate US$ 650 [INR 40000] have been
used for implementation.

3 Concept Model

The proposed system will be used for rehabilitation purpose of the stroke patients. It
will consist of a display unit/monitor or presently available high end television
placed in front of the patient at suitable location considering the lighting condition
and clear view of the face of the user. Several large size icons will be displayed on
the screen. These icons may include ‘Call the nurse’, ‘Self-guide for Physical
Exercises’, ‘Read Books’, ‘Listen Music’, ‘Watch Movies’, ‘Ask for Water’, ‘Ask

Fig. 1 The concept model of

the proposed system with the
patient lying halfway on the
bed along with the display
and camera unit placed on the
wall at the front
Methodology for a Low-Cost Vision-Based Rehabilitation … 369

for Food’ etc. The patient needs to locate the desired icon using gazes and if
necessary follow the onscreen instructions. A high definition camera will be
installed at a fixed location to track the eye balls from a specified distance. Night
vision camera is preferred to other normal cameras for viewing even in the dark.
Otherwise proper illumination has to be arranged especially during the night.
A processing unit will be employed for capturing, storing and processing of images.
Windows based processors are preferred due to use of Microsoft Visual Studio
2013 and Open CV software. Initially a desktop PC has been recommended. But
with the advancement of technologies small processor or low-cost single board
computer or any other windows based gadgets can be used. As shown in Fig. 1 the
patient may completely lie down on the bed or may lie down halfway on the bed
and the system will work from a distance. The proposed concept model has been
presented in Fig. 1.

4 Adopted Methodology

The objective of the present work is to detect the eyeball of the user and track it for
activating certain commands displayed on the computer screen placed in front of a
stroke patient. To achieve the goal a few steps have to be followed properly.
Firstly the face has to be detected amidst of cluttered background from the captured
image. Then the eye regions are extracted from the face region. Next step is to
detect the eyeball centre from the eye region using eye centre localization. Cali-
bration is another important step to be followed to identify the exact position of the
eye centre on the camera window with reference to the interface screen. This will
help to select the icons to activate specific commands and proceed accordingly. The
generalised system architecture of the proposed system has been shown in Fig. 2.
Till date detection of face region and detection of eye region from the detected
face region have been achieved. This method is preferred to detection of eye regions
directly as most of the available cheap cameras (webcam) are not of higher reso-
lution. The major disadvantages are that the captured images are blurry and the

Fig. 2 Generalized system architecture for proposed system

370 A.R. Sarkar et al.

Fig. 3 The flow diagram of

the proposed methodology for
face detection and eye
detection for stroke patients

computational load is also more. So it is better to apply the necessary processing

steps on the extracted eye regions rather than processing the whole image to locate
the eye centres. This reduces the computational load. In the current work Haar
classifiers have been used to detect faces and eye regions. Paul Viola and Michael
Jones proposed the Haar feature-based cascade classifier to detect objects effec-
tively in 2001 [20]. In this method a window of preferred size is moved over the
input image and the Haar-like feature for each sub-section of the image is calcu-
lated. This value is compared to a predefined threshold that differentiates objects
Methodology for a Low-Cost Vision-Based Rehabilitation … 371

from non-objects. This is a machine learning technique where the function is trained
using positive (contains object) and negative (contains non-object) images.
Microsoft Visual Studio 2013 with Open CV libraries has been used as the
programming platform. Main advantages of Open CV are that it is easily available
and it contains many pre-trained classifiers for detection of face, eye, smile etc. The
captured video by the camera is loaded frame by frame and the desired xml files are
loaded and applied to each of these frames. The three consecutive detections are
merged as one. Pruning is done to exclude the region of less interest i.e. where
chance of appearance of face is very less. The smallest window to detect face can be
defined based on the need or application. Similar approach is followed to detect
eyes. The detected region is then marked by a rectangular box.
The final step is to detect eye centre from the extracted images of eye regions.
Several methods are available. A few of them are colour based filtering, shape
based filtering, use of Hough circles. Another novel method is to train classifiers for
eyeballs like that of faces and eye regions. Presently colour based filtering has been
used to detect eye centres. The schematic diagram of the above methods has been
presented in Fig. 3.

5 Experiments

Image capturing and processing are the two fundamental steps associated with the
present work. The quality of the image plays a vital role in this case. There are
several factors those affect the quality. A few of them are type of the camera (CCD
or CMOS), resolution of the camera, illumination of the environment, type and
position of light, distance between the camera and object etc. These factors should
be taken care of to capture a good quality image leading to successful detection.
The experiments have been carried out in normal room (dimension 12 Ft × 12
Ft) as the application of the proposed system will be limited to indoor only i.e. in
home or hospital. The isometric view of the experimentation room has been pre-
sented above in Fig. 4. There are two doors, one window and one fluorescent light
in the room. Though the application of the system will be for the patients either
completely lying or lying halfway on the bed, here for the ease of initial experi-
ments user sits on the chair with the window at the back. The display unit along
with the camera is placed on a table at a distance of 2 Ft in front of the user. Only a
fluorescent light has been used during the experiments in night.
Two different types of USB cameras have been used. Initially a low cost CMOS
camera with 300 K pixels resolution (Make: Frontech, Model: JIL 2244) [21] has
been used and that has been replaced later by a 720 pixel HD camera (Make:
Logitech, Model: HD C525) [22] with zoom and auto focus. The quality of images
captured by these two cameras are presented in Fig. 5a, b. Simultaneously a lux-
meter has been employed on a plane co-planner with the face to measure the
average illumination of the surroundings. The data have also been recorded for
reference.
372 A.R. Sarkar et al.

Fig. 4 Isometric CAD view of the experimental system showing that the user sitting in the middle
of a room with the display and camera unit in front

Fig. 5 The online face and eye detection as well as marking from the frame captured by a the low
resolution camera without spectacles, b the high resolution camera of the user with spectacles at
same location in different orientation

Continuous videos have been recorded using both the cameras in both day and
night with different position of the user and lighting conditions. The programme
also constantly detects and marks faces and eyes automatically online. As the user
moves the eyeballs, the eye centres are detected and marked online. Both the cases
with spectacles and without spectacles have been considered during the experi-
ments. The saved videos and associated data have been analysed and presented in
the tables below. The relevant processed data for the low resolution camera with
and without spectacles have been given in Tables 2 and 3 presents the processed
data for high resolution camera.
Methodology for a Low-Cost Vision-Based Rehabilitation … 373

Table 2 Video data from the

low resolution camera Total frames Correct detections Semi-correct detection
Without spectacles
1,157 1,110 1,149
1,113 1,025 1,110
1,158 1,143 1,155
1,043 688 1,026
With spectacles
1,043 473 819
1,158 830 1,064
1,158 609 1,030
1,149 851 1,017

Table 3 Video data from the

high resolution camera Total frames Correct detections Semi-correct detection
Without spectacles
1,161 1,036 1,146
1,042 941 997
1,139 1,080 1,138
1,157 1,152 1,153
With spectacles
862 715 856
810 717 802
1,158 961 1,143
832 611 821

6 Results and Discussion

The efficiency of detection is calculated for the above mentioned data with reference
to the total number of detected frames. Thus the efficiencies of detection for the low
resolution camera with and without spectacles have been shown in Fig. 6a, b
respectively. Similarly Fig. 7a, b present the efficiencies of detection for the high
resolution camera with and without spectacles respectively.
Here the following guideline has been followed to consider the correct detection
of eyeballs. The main aim is to detect correctly the eyeballs of the opened eyes. So
when the eyes are closed there should be no detection or in other sense if the eyes
are closed and no detection is there, it has been considered as correct detection.
However detection has also been assumed or manipulated by tracking the single eye
and knowing the distance between the centres of the two eyeballs for common
normal people. This may fail for squint-eyed people.
The efficiency of correct detection has been denoted by the solid red line and the
dotted blue lines represent the semi-correct detection that includes the detection of
single eye, double eyes and closed eyes. The performance of the low resolution
374 A.R. Sarkar et al.

Fig. 6 Efficiency of detection for low resolution camera. a Without spectacles. b With spectacles

camera without spectacles is found better than that of with spectacles. All the values
except one are lying above the range of 80 %. The efficiencies of detection for the
low resolution camera with spectacles lie between 40 and 75 % which seems to be
very poor for the intended application. This might happened due to improper
illumination during the experimentation in the night time.
The efficiencies of detection for the high resolution camera without spectacles
always lie above 80 %. This has been achieved due to no glare of the spectacles and
experiments are carried out in uniform day light. Also the efficiencies of detection
for the high resolution camera with spectacles have been found to be present above
80 %. There has been no definite patterns followed in the above graphs and hence
no inference can be drawn. This may be due to the fact that each video is different
Methodology for a Low-Cost Vision-Based Rehabilitation … 375

Fig. 7 Efficiency of detection for low resolution camera. a Without spectacles. b With spectacles

from the other with respect to the orientations of the user, lighting conditions,
background etc. However the illumination of the surroundings has a relationship
with the efficiency of detection as shown in Fig. 8. The efficiency of detection will
increase with the increase in surrounding illumination.
To provide active online support to the stroke patients 100 % efficiency is
desirable. There are several reasons for hindrances. The most identified issues are
low illumination, low resolution, glaring from the spectacles, incidence of light on
the camera, cluttered background etc. These difficulties can be overcome by using
high resolution camera, placement of the camera in the direction of light (not
opposite), using proper and uniform lighting, clean and clear spectacles (glasses) of
the user.
376 A.R. Sarkar et al.

Fig. 8 Efficiency of detection without spectacles for low resolution camera against different
environmental illuminations

7 Conclusion

The objective of the present work is to develop a low cost eye tracking system to be
used by the stroke patients for rehabilitation. The proposed methodology uses the
well known machine learning technique, Haar Classifiers. Haar classifiers provide
better performance than other eye tracking algorithm due to its training capability.
Microsoft Visual Studio 2013 along with Open CV has been used to carry out the
work. Experiments have been performed in a normal living room using low and
high resolution camera in both day and night with and without spectacles. Effi-
ciency of detection for the high resolution camera without spectacles has better
performance than all others. Several factors are responsible for not achieving 100 %
efficiency. Work is in progress to achieve 100 % efficient, robust, rugged system
desirable for the rehabilitation of stroke patients [23]. As the future scope, move-
ment of the eyeballs in the camera window is being mapped on the wall-hang
monitor or any other suitable display.

Acknowledgments Authors are grateful to the Head and other faculty members of Department of
CSE, NIT-Durgapur and SR Lab, CSIR—CMERI Durgapur for their continuous help and tired
less support. Authors also thank Dr. D.N. Ray for his continuous suggestions and advices, without
which this work would not have completed.
Methodology for a Low-Cost Vision-Based Rehabilitation … 377

References

1. Taylor FC, Suresh Kumar K (2012) Stroke in India fact sheet (updated 2012)
2. National Stroke Association (2013) Explaining stroke. http://www.stroke.org/site/PageServer?
pagename=explainingstroke. Accessed 19 July 2013
3. Dalal P, Bhattacharjee M, Vairale J, Bhat P (2007) UN millennium development goals: can we
halt the stroke epidemic in India? Ann Indian Acad Neurol 10(3):130–136
4. Jack D, Boian R, Merians AS, Tremaine M, Burdea GC, Adamovich SV, Recce M, Poizner H
(2001) Virtual reality-enhanced stroke rehabilitation. IEEE Trans Neural Syst Rehabil Eng 9
(3):308–318
5. Ushaw G, Ziogas E, Eyre J, Morgan G (2013) An efficient application of gesture recognition
from a 2D camera for rehabilitation of patients with impaired dexterity. School of Computing
Science Technical Report Series. http://www.cs.ncl.ac.uk/publications/trs/papers/1368.pdf.
Accessed 22 Jan 2013
6. Sucar L, Luis R, Leder R, Hernandez J, Sanchez I (2010) Gesture therapy: a vision-based
system for upper extremity stroke rehabilitation. In: Proceedings of IEEE engineering medical
biology society, 2010, pp 107–111
7. Reinkensmeyer DJ, Pang CT, Nessler JA, Painter CC (2002) Web-based tele rehabilitation for
the upper extremity after stroke. IEEE Trans Neural Syst Rehabil Eng 10(2):102–108
8. Fasoli SE, Krebs HI, Stein J, Frontera WR, Hogan N (2003) Effects of robotic therapy on
motor impairment and recovery in chronic stroke. Arch Phys Med Rehabil 84:477–482
9. Huq R, Wang R, Lu E, Lacheray H, Mihailidis A (2013) Development of a fuzzy logic based
intelligent system for autonomous guidance of poststroke rehabilitation exercise. In:
Proceedings of 13th international conference on rehabilitation robotics, 24–26 June 2013, WA
10. Huq R, Lu E, Wang R, Mihailidis A (2012) Development of a portable robot and graphical
user interface for haptic rehabilitation exercise. In: Proceedings of 4th IEEE/RAS-EMBS
international conference on biomedical robotics and biomechatronics, June 2012, Italy
11. Block M, Mercado L (2013) Talking tech: technology expands communication opportunities
for people with aphasia, everyday survival. Springer, Heidelberg, pp 18–19
12. Roke Manor Research Ltd (2013) Microsoft kinect gesture recognition software for stroke
patients, Inside Technology, Issue 9. http://www.ttp.com. Accessed 07 June 2013
13. Arrington Research (2013) Eye tracker prices. http://www.arringtonresearch.com/prices.html.
Accessed 20 Dec 2013
14. Mirametrix, S2 Eye Tracker (2014) http://mirametrix.com/products/eye-tracker. Accessed 25
Feb 2014
15. iMotions A/S Denmark, Quotation no. 884836000000801041, 21 February 2014
16. Gazepoint Products (2014) http://gazept.com/products. Accessed 15 Feb 2014
17. EyeGuide Mobile Tracking Price (2014) http://www.grinbath.com/content/eyeguider_mobile_
tracker_pricing. Accessed 13 June 2014
18. Aerobe Medicare Pvt. Ltd., New Delhi, Quotation no. nil, 07 April 2014
19. Tobii Glasses 2 (2014) http://www.tobii.com/en/eye-tracking-research/global/landingpages/
tobii-glasses-2/our-offering. Accessed 25 May 2014
20. Viola P, Jones M (2001) Rapid object detection using a boosted cascade of simple features. In:
Proceedings of IEEE conference computer vision and pattern recognition, 2001
21. Product—Frontech (2012) E-brochure. http://www.frontechonline.com/product.php. Accessed
25 May 2012
22. Logitech HD Webcam C525 (2013) http://www.logitech.com/en-us/product/hd-webcam-c525.
Accessed 12 Oct 2013
23. Sarkar AR, Sanyal G, Majumder S (2013) Hand gesture recognition systems: a survey. Int J
Comput Appl 71(15):25–37
Coacervation—A Method for Drug
Delivery

Lakshmi Priya Dutta and Mahuya Das

Abstract The present review outlines recent advances in coacervate based

research, their historical background and area of diversification. Methods of their
preparation, encapsulation, theoretical overview, coacervation induced nano parti-
cle formation, applications in various fields have been covered. Chemically mod-
ified coacervates used in drug delivery research are discussed critically to evaluate
the usefulness of these system in delivering bioactive molecules. From literature
survey, it is realized that coacervate based research in drug delivery as well as in
proto cellular biology have increased rapidly. Hence the present review is timely.

Keywords Coacervate
Coacervation
Nanomedicine
Biogenesis

1 Introduction

One of the most unsolved, difficult and interesting mystery surrounding the science
lies behind the mechanism of life formation via the interaction of bio macro mol-
ecules. In relation with this Pasteur have done one experiment on biogenesis [1, 2]
which revealed only that the life can not arise spontaneously under conditions that
exist on earth today and the pre-biotic environment must be different billions of
years ago. In 1929, J.B.S. Haldane presented a brief, paper in The Rationalist
Annual that reflects the modern concept on protocell theory according to which life
arose on earth as membrane bound system [1–12]. According to them- There is no
fundamental difference between a living organism and lifeless matter. The complex
combination of manifestations and properties so characteristic of life must have

L.P. Dutta
Department of Nanoscience and Technology, JISCE, Kalyani, Nadia, West Bengal, India
M. Das (&)
Department of Chemistry, Department of Nanoscience and Technology,
JISCE, Kalyani, Nadia, West Bengal, India
e-mail: [email protected]

© Springer India 2015 379

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_35
380 L.P. Dutta and M. Das

arisen in the process of the evolution of matter [4–6]. Oparin have proposed the
solution leading to the idea of coacervate which is termed as microspheres of
assorted organic bio molecules that are associated by weak interactions [4–10]. All
the newer research throughout the world to solve the mystery has been limited
towards the evolution of macro molecules like DNA, RNA, protein [1, 3, 4] from
the past decades. On the basis of these ideas the interaction between these types of
biomacromolecules has broadened the area of investigation starting from the origin
of life to the recent advancement in drug delivery [4] including encapsulation.

2 Concept Behind Coacervation

The coacervates are spherical aggregation of organic macro molecules making up a

inclusion which is held by hydrophobic forces [1, 3–8]. The boundaries of the
coacervates allow the selective adsorption of simple organic molecules from the
surrounding medium [13]. The drop that forms the coacervate particle is often a
liquid oil and the drops will have a shell of the hydro colloid mixture around them
and under correct conditions, the colloids will set to form a solid or semipermeable
shell [14, 15]. Coacervate formation is both entropy and enthalpy driven [4, 5] and
so a minimum temperature have to be maintained. Coacervate are basically the cell
like encapsulator model [12, 16, 17] to host the cellular components and bio
molecules. Though Lipid vesicles are considered the prototypical protocell as they
can form functional microscopic spherical assemblies due to their amphiphilic
nature, but there are alternative models based on colloid phase separation that lead
to the emergence of compartments. In many such encapsulation processes like food
flavoring [14, 15, 18–20], drug delivery etc where it is desirable to encapsulate as
well as compartmentalize a particle, enzymatic materials or drop, or polymeric
stimuli responsive drug moiety to protect the interior contents of that particle, to
provide controlled release [21] or to experiment the taste of the interior phase and
thus coacervation are employed.

3 Mechanism

The coacervation process involves the formation of microscopic droplets of the

coacervate phase in the stirred liquid phase resulting usually in unstable colloidal
dispersion showing tendency towards coalescence. The droplet size can be pre-
served and maintained by chemical crosslinking or physical gelation of the polymer
in the coacervate phase. The coacervation mechanism is generally subdivided into
two groups: simple coacervation and complex coacervation. The simple coacer-
vation (segregative phase separation) involves partial desolvation of the polymer
caused either by changing the solvent quality for the polymer in the solution or by
changing the solution temperature, the pH in the case of weak PEs, or by adding a
Coacervation—A Method for Drug Delivery 381

Fig. 1 Coacervation: multi phase distribution phenomenon

nonsolvent for the polymer to the solution. In medical technology, simple coac-
ervation is often used for entrapping drugs into microcapsules.
When solutions of two colloids are mixed at a specific concentration, phase
separation [14, 15, 18–20] between the colloidal system have been termed as
complex coacervation. Complex coacervation typically possesses very low inter-
facial free energy [4, 5] in aqueous solution and this property enable the coacervate
to engulf a variety of particles in solution. Hence the complex coacervation refers to
the associative phase separation that occurs between two oppositely charged
polyelectrolytes (PE), leading to the formation of polyelectrolyte complexes, or
between a PE and di- and trivalent counterions resulting in so-called “inotropic”
hydrogels. In rare cases, complex coacervation may occurs between polymers via
hydrogen bonds leading to the formation of hydrogen-bonded complexes.
Fabrication methods of coacervation involves formation of stable micro and
nanoparticles under mild conditions without using organic solvents, surfactants, or
steric stabilizers. The chemical phases present in coacervate system are as follows
(i) a liquid manufacturing vehicle phase (ii) a core material phase and (iii) a coating
material phase. To form the three phases, the core material is dispersed in a solution
of the coating material; the solvent for the polymer is the liquid manufacturing
vehicle phase. The polymer coating material phase can be formed by changing the
temperature of the polymer solution or by adding a salt or incompatible polymer
[15, 18–20] (Fig. 1).

4 Coacervates Induced Nano Particle Formation

The remarkable achievements attained in the fields of nanoscience and nanotech-

nology have led to applications in the most diverse fields of medicine. Nanomed-
icine is the application of nanotechnology (the engineering of tiny particles) to the
prevention and treatment of disease. In principle, nanomedicine focuses on two main
fields: (a) development of new diagnostic systems like nanobiosensors, nanochips
and (b) Development of highly efficient delivery vehicles to incorporate the drug
induced nano particles within human body. Complex coacervation induced nano
382 L.P. Dutta and M. Das

particle formation was investigated by Hu et al. Hu et al. prepared chitosan–PAA

nanoparticles by template polymerization of PAA in a chitosan solution at 70 °C
using K2S2O8 as initiator [22]. Chitosan nanoparticles encapsulating siRNA were
prepared by complex coacervation of chitosan and polyguluronate [23]. Because of
their ease of preparation and extensive modification opportunities, complex coac-
ervate core micelles (C3Ms) are used as a good alternative for diffusional nanop-
robes. Carboxymethylcellulose and alginate have also been complexed via the
coacervation technique with chitosan to prepare nanoparticles. There are numerous
reports on the preparation of chitosan–DNA nanoparticles and chitosan–heparin
nanoparticles via phase induced coacervation technique. Thus complex coacervation
is a very efficient and useful procedure for obtaining nanoparticles. The speculative
subfield of nano technology have emerged a new dimension towards incorporation
of nanomedicines within micro capsules as well as formation of nano particles via
the microsphere based techniques.

5 Factor Affecting Coacervation

The coacervation procedure is the stimuli responsive system that largely depends on
various factors like concentration of the added macro ions, pH, temperature and
ionic strength. The concentration of added micro ions are varied with the total
polyion mixing concentration increasing with an increase in the polyion concen-
tration up to a maximum at a polyion concentration of approximately 1 % (w/v) and
then decreasing. The micro ion concentration at which maximum coacervation
occurs changes with poly ion concentration as does the value of this maximum.
Kaibara et al. saw no temperature dependence for pHc and pHφ for PDADMAC and
BSA. On the other hand, because coacervation of PE/micelle system can be
entropy-driven, by release of counterions, temperature can induce coacervation.

6 Theoretical Approach

In accordance with the view of oparin-Haldane though coacervation theory is

typically based on protocell formation within the coast of archaic water basin some
hypothesis (coacervate in coacervate) also made on the basis of coacervate as
“culture media”. The approach concerns coacervates as suitable culture media from
which the first polynucleotide have ever manufactured and around the replicating
DNA molecule secondary DNA molecules accumulated which developed gradually
to the first prokaryotic cells. A theoretical approach on complex coacervation was
first demonstrated by overbeek and voorn on the basis of experimental studies of de
Jong [24] and his coworkers. The voorn-overbeek theory practically based on the
configurational status of coacervates in multiple phases [15, 25] though they have
neglected the Huggins interaction or solute-solvent interaction. A second theory
Coacervation—A Method for Drug Delivery 383

was parallely proposed by Veis [18]; “the dilute phase aggregation model” that
correlates the electrostatic attraction phenomenon with simple thermodynamics
according to which entropy driven aggregation drives the complex coacervation.
There are several basic differences between these two theories. Voorn-overbeek
theory have neglected the thermodynamic approach along with solute-solvent
interactions and they have utilized the gelatin-acacia system as a major one whereas
Veis-Aranyi group deals with gelatin-gelatin system as a major one. However later
the voorn-overbeek model was modified by Veis by including the Huggins
parameter whereas replacing the debye-huckel parameter [18].

7 Application

Coacervates have been taken over in various field of application of medical tech-
nology with their astonishing versatility. The nature of coacervation as described is
fully dependent on weak electrostatic interaction and entropy dependent phase
separation mainly. Over the course of the years, the research on Coacervates have
been shifted from prebiotics to its application in modern medicinal chemistry. In
this regard, Scientists have taken different types of oppositely charged polyelectr-
olites to monitor their self assembling behavior.
The polymer coacervation is a process of liquid–liquid phase separation of a
polymer solution into a polymer-rich phase (coacervate phase) and a polymer-lean
phase (equilibrium phase) which is generally distinct from precipitation or coagu-
lation—a phase separation process in colloidally unstable systems resulting in the
formation of compact aggregates (coagula). Coacervation involves formation of
stable micro and nanoparticles or spheres under mild conditions without using
organic solvents, surfactants, or steric stabilizers. The fabrication being a reversible
process, finds a potential application as vehicles for triggered drug delivery.
The complex coacervation is an increasingly popular approach for the fabrica-
tion of particulate drug delivery systems. Stable water-soluble colloidal particles are
often formed by oppositely charged PEs mixed together under certain conditions in
a ratio by which a nonstoichiometric electrostatic complex is formed in the fabri-
cated coacervate. The existing studies on complex coacervate particles is actually
based on the early works of Tsuchida’s and Kabanov’s groups [19].
In the vast majority of studies in relation to application of complex coacervate,
chitosan, a naturally occurring weak polycation, and its derivatives are used as one
of the components. Besides biocompatibility, biodegradability, and low toxicity, the
popularity of chitosan is explained by its mucoadhesive properties, making it useful
for transmucosal drug delivery as well as its ability to open tight junctions between
epithelial cells, enhancing the transport of macromolecules across epithelia [19].
Colloids based on electrostatic chitosan–DNA, chitosan–protein, and chitosan–
polysaccharide complexes along with chitosan hydrogels crosslinked with polyion
tripolyphosphate have drawn much interest due to the formation of strong inter-
action between high positive charge on the backbone of chitosan and those
384 L.P. Dutta and M. Das

negatively charged biomacromolecules with them. Because of the electrostatic

nature of these complexes, they are intrinsically pH- and ionic strength responsive
and are modern tool for drug delivery.
Complex coacervation phase separation behavior via protein-polysaccharide
interaction like gelatin gum acacia system was first reported by Oparin [1]. Fol-
lowed by him different types of protein-polysaccharide system have been taken into
account to study the coacervation. As depicted chitosan-gelatin, chitosan-gum
acacia, chitosan-DNA system are granted as the major ones for the experimental
purpose. Chitosan the deacetylated product of chitin, chemically known as β (1,4)
2-amino-2-deoxy-D-glucose is a hydrophilic polysaccharide renders special cat-
ionic property in solution chemistry [20]. Due to it’s excellent biodegradable and
biocompatible property, this naturally occurring biopolymer act a major role
towards medicinal chemistry. Gelatin, gum acacia [14, 15, 18–20, 26] and DNA are
anionic biomolecules that interacts with positively charged moieties via supramo-
lecular interactions [14, 15, 18–21, 26, 27]. Formation of lysozyme-heparin com-
plex coacervates [21] at a pH near the physiological pH (pH-7.2) largely resembles
the proto cellular type coordination. Among different scenarios, inter molecular
complexation between oppositely charged macro molecules like lanthanide based
[27] heavy metal, ATP-Polylysine [28–31], polynucleotide-ATP [32] coacervates
have emerged a new field of research on microsphere based drug encapsulation as
well as the self sustaining nature have emerged a new paradigm in the area of origin
of life. Spontaneous self assembling behavior of fatty acid entrapped membrane on
the surface of preformed coacervate micro droplets has also been investigated.
These types of membrane bounded coacervate microdroplets have a broad range of
sequestration property as compared to uncoated coacervate micro droplets. Parti-
tioning of different types of dyes, nano particles, enzymes and coenzymes within
the membrane bound [22, 23, 32, 33] as well uncoated coacervate micro droplet
inferior have been well established. Work in progress is focused on incorporating
more versatile types of nano- and micro phases into complex coacervates to further
improve encapsulation properties. Complex coacervation was proposed to play a
role in the formation of the underwater bioadhesive of the Sandcastle worm
(Phragmatopoma californica) [22].

8 Conclusion

Superamphiphiles or supramolecular chemistry based on coacervation have

emerged a new scope of research towards various field of science. Today the
comprehensive technology of coacervation enables it to satisfy the need as en-
capsulator. However the most important aspect of coacervate research is their
diversity towards application in pharmaceutical industry as well as in food and
polymer industry. Moreover electrochemical moieties can be attached through the
procedure via noncovalent interactions greatly reducing the need for organic
chemical synthesis. Since the coacervate moieties are expected to possesses stimuli
Coacervation—A Method for Drug Delivery 385

responsiveness, they can able to compartmentalize. Thus condensation of wide

range of poly electrolytes in aqueous media produces microdroplets that are capable
of sequestering wide range of compounds, dyes, drug molecules and nano particles.
Thus coacervation is an extremely facile method for non covalent construction of
stimuli responsive moieties.

References

1. Oparin AI (1953) The origin of life, 2nd edn. Dover Publications, New York
2. Mansy SS et al (2008) Template-directed synthesis of a genetic polymer in a model protocell.
Nature 454:122–125
3. Rasmussen S et al (eds) (2009) Protocells: bridging nonliving and living matter. MIT Press,
Cambridge
4. Luisi PL (2006) The emergence of life. Cambridge University Press, Cambridge
5. Hargreaves WR, Deamer DW (1978) Liposomes from ionic, single-chain amphiphiles.
Biochemistry 17:3759–3768
6. Szostak JW, Bartel DP, Luisi PL (2001) Synthesizing life. Nature 409:387–390
7. Meierhenrich UJ, Filippi JJ, Meinert C, Vierling P, Dworkin JP (2010) On the origin of
primitive cells: from nutrient intake to elongation of encapsulated nucleotides. Angew Chem
Int Ed 49:3738–3750
8. Dzieciol AJ, Mann S (2012) Designs for life: protocell models in the laboratory. Chem Soc
Rev 41:79–85
9. Deamer DW, Dworkin JP (2005) Chemistry and physics of primitive membranes. Top Curr
Chem 259:1–27
10. Apel CL, Deamer DW, Mautner MN (2002) Self-assembled vesicles of monocarboxylic acids
and alcohols: conditions for stability and for the encapsulation of biopolymers. Biochim
Biophys Acta 1559:1–9
11. Oberholzer T, Wick R, Luisi PL, Biebricher CK (1995) Enzymatic RNA replication in self-
reproducing vesicles: an approach to a minimal cell. Biochem Biophys Res Commun
207:250–257
12. Chen IA, Szostak JW (2004) Membrane growth can generate a transmembrane pH gradient in
fatty acid vesicles. Proc Natl Acad Sci USA 101:7965–7970
13. Hyman AA, Simons K (2012) Cell biology. Beyond oil and water—phase transitions in cells.
Science 337:1047–1049
14. Burgess DJ (1960) Complex coacervates of gelatine. J Phys Chem 64 1203–1210
15. Overbeek JTG, Voorn MJ (1957) Phase separation in polyelectrolyte solutions. Theory of
complex coacervation. J Cell Comp Physiol 49(Supp I):7
16. Zhu TF, Adamala K, Zhang N, Szostak JW (2012) Photochemically driven redox chemistry
induces protocell membrane pearling and division. Proc Natl Acad Sci USA 109:9828–9832
17. Adamala K, Szostak JW (2013) Competition between model protocells driven by an
encapsulated catalyst. Nat Chem 5:495–501
18. Veis A (1961) Phase separation in polyelectrolyte solutions. II. Interaction effects. J Phys
Chem 65:1798–1803
19. Motornov M, Roiter Y, Tokarev I, Minko S (2010) Prog Polym Sci 35:174–211
20. Sionkowska A, Wisniewski M, Skopinska J, Kennedy CJ, Wess TJ (2004) The photochemical
stability of collagen-chitosan blends. Biomaterials 162:545–554
21. Schmitt C, Sanchez C, Thomas F, Hardy J (1999) Complex coacervation between h-
lactoglobulin and acacia gum in aqueous media. Food Hydrocoll 13:483–496
22. Stewart RJ, Wang CS, Shao H (2011) Complex coacervates as a foundation for synthetic
underwater adhesives. Adv Colloid Interface Sci 167:85–93
386 L.P. Dutta and M. Das

23. Hu Y, Jiang X, Ding Y, Ge H, Yuan Y, Yang C (2002) Biomaterials 23:3193–3201

24. Bungenberg de Jong HG, Kruyt HR (1929) Coacervation (partial miscibility in colloid
systems). Proc K Ned Akad Wet 32:849–856
25. Overbeek JTG, Voorn MJ (1957) Phase separation in polyelectrolyte solutions. Theory of
complex coacervation. J Cell Comp Physiol 49(1):7–26
26. Burgess DJ, Carless JE (1984) Microelectrophoretic studies of gelatin and acacia for the
prediction of complex coacervation. J Colloid Interface Sci 98(1):1–8
27. Wang J, Velders AH, Gianolio E, Aime S, Vergeldt FJ, Van As H, Yan Y, Drechsler M, de
Keizer A, Cohen Stuarta MA, van der Guchta J (2013) Controlled mixing of lanthanide(III)
ions in coacervate core micelles. Chem Commun 3736
28. Seyrek E, Dubin PL, Tribet C, Gamble EA (2003) Ionic strength dependence of protein
polyelectrolyte interactions. Biomacromolecules 273–282
29. Poon W, Pusey P, Lekkerkerker H (1996) Colloids in suspense. Phys World 55:3762
30. Mattison KW, Brittain IJ, Dubin PL (1995) Protein–polyelectrolyte phase boundaries.
Biotechnol Prog 11:632–637
31. Weinbreck HS, Rollema RH (2004) Tromp, diffusivity of whey protein and gum arabic in their
coacervates. Langmuir 20:6389–6395
32. Koga S, Williams DS, Perriman AW, Mann S (2011) Peptide–nucleotide microdroplets as a
step towards a membrane-free protocell model. Nat Chem 3(9):720
33. Lee DW, Yun K-S, Ban H-S, Choe W, Lee SK, Lee KY (2009) J Control Rel 139:146–152
34. Fulton AB (1982) How crowded is the cytoplasm? Cell 30:345–347
35. Bakker MAE, Galema SA, Visser A (1999) Microcapsules of gelatin and carboxy methyl
cellulose. European Patent Application EP 0 937 496 A2, Unilever NV, NL; Unilever PLC,
GB (Bangs WE, Reineccius GA 1981)
36. Tiebackx FWZ (1911) Gleichzeitige Ausflockung zweier Kolloide. Chem Ind Kolloide
8:198–201
A Simulation Study of Nanoscale
Ultrathin-Body InAsSb-on-Insulator
MOSFETs

Swagata Bhattacherjee and Subhasri Dutta

Abstract In this paper, we report, a simulation study of nanoscale ultra thin body
InAsSb channel n-MOSFETs. Our work is based on numerical simulation using
ATLAS, a 2-D device simulator. Accuracy of the model has been verified by
comparing simulation results with the reported experimental data. The proposed
model has been employed to calculate the drain current, transconductance of InAsSb
channel MOSFETs for different gate and drain voltages and also to compute Short
Channel Effects.

Keywords Analog circuit applications InAsSb DG MOSFETs Transconductance




Threshold voltage roll-off Subthreshold slope

1 Introduction

Silicon is on the verge of its fundamental limits due to the aggressive scaling of the
modern complementary metal oxide semiconductor field effect transistor (MOS-
FET) [1]. Existing technology trends indicate that more non-Si elements are being
added to the Si transistor for enhancing its scalability and performance for the future
CMOS technology. Recently high mobility channel materials has attracted exten-
sive attention due to its improved carrier mobility [2]. Among all known semi-
conductors, InAsxSb1−x has one of the highest electron mobilities [2–9] and
saturation velocities. However, due to its smaller band gap and larger permittivity,

S. Bhattacherjee (&)
Department of Physics, JIS College of Engineering, Block A, Phase III,
Kalyani, India
e-mail: [email protected]
S. Dutta
Department of Nano Science and Technology, JIS College of Engineering,
Block A, Phase III, Kalyani, India
e-mail: [email protected]

© Springer India 2015 387

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9_36
388 S. Bhattacherjee and S. Dutta

InAsSb channel MOSFETs suffer from high value of leakage current and reduced
electrostatic integrity in short channel devices.
There have been a few approaches available in the literature [4, 5] in order to
evaluate the current voltage characteristic of InAsSb channel devices pertaining to
DG MOSFETs. In [4], an experimental work on the fabrication of ultrathin-body
InAsSb-on-insulator n-type field effect transistors (FETs) with ultrahigh electron
mobilities has been reported. Another group [4, 5] worked with depletion and
enhancement-mode InAsSb MOSFETs by integrating a composite high-κ gate stack
considering ultrathin InAs0.7Sb0.3 quantum well structure. But till date no work has
been delt with short channel effects of InAsSb channel MOSFETS.
In the present paper first time we have studied the simulation work on short
channel effects of InAsSb channel MOSFETs. Also the performance of nanoscale
InAsSb channel MOSFETs over the entire region of operation have been studied.
Further our predicted results have been compared with reported experimental data
[8] to ensure the validity of our proposed model.

1.1 Model Development

The cross section of the n-type InAsSb channel MOSFET considered in our study is
shown in Fig. 1. The details of the process flow for fabrication of such a device are
reported in [8]. The device comprises a 500-nm n-type ultrathin InAs0.7Sb0.3 layers
of different thicknesses (TInAsSb = 7 and 17 nm) with a doping concentration of
1 × 1017 cm−3. The gate insulator consists of a 10-nm thick ZrO2 dielectric layer,
while Ni is used for formation of ohmic source and drain contacts. For top-gate
FETs, a 10-nm-thick ZrO2 while for bottom gate a 50 nm SiO2 gate dielectric is
used. In the investigations we have employed 2-D numerical device simulator

Fig. 1 Schematic diagram of

a InAs0.7Sb0.3 channel
MOSFET
A Simulation Study of Nanoscale Ultrathin-Body … 389

ATLAS to simulate the InAsSb MOSFET. Two-dimensional device simulations are

done for the InAsSb-channel MOSFET, as shown in Fig. 1. As the channel material
is an alloy, its dielectric constant and bandgap are computed using linear interpo-
lation among the corresponding reported parameters of the constituents [9]. The
computed values of dielectric constant and bandgap of 17.7 and 0.174 eV,
respectively, for InAs0.7Sb0.3 are used in our simulation. An intrinsic carrier con-
centration of 1.8 × 1016 cm−3 for InAsSb is used in our study. Experimental
findings reveal that the InAsSb/high-k interface contains appreciable amount of
interface trap charge density in the range of 1013 eV−1 cm−2 [8] which is incor-
porated in our simulation. Physical models that were taken into account in the
simulation included the electric field dependent carrier mobility and the velocity
saturation. The measured value of electron mobility in InAsSb channel n MOSFETs
are used to obtain empirically the field dependent relationship for mobility [9]. This
functional variation of mobility with respect to the effective field has been incor-
porated as a library file in the simulator by employing the C-interpreter tool.
Various electrical parameters like electron concentration, surface potential, electric
field, drain current, etc. have been obtained from the device simulation of InAsSb
channel MOSFETs. Figure 1 shows Schematic diagram of an InAsSb channel n-
MOSFET displaying different regions such as source, drain and gate along with
mesh points within the grid as obtained from ATLAS. All dimensions are in μm.

2 Results and Discussions

Figure 2 shows the surface potential variation of the structure referred in Fig. 1. A
comparison of the simulated device characteristics with the experimental results as
reported in [8] for the InAsSb MOSFET is shown in Fig. 3 for three different values

Fig. 2 Variation of surface

potential with channel length.
Other parameters are
l = 500 nm, tInAsSb = 7 nm,
Dit = 1 × 1013 eV−1 cm−2,
teq = 1.5 nm, Vgs = −0.2 V
and Vds = 50 mV
390 S. Bhattacherjee and S. Dutta

Fig. 3 Dependence of drain 500

current with drain bias for Vg = 0.2 V
three different gate biases.
400 Vg = 0.4 V
Other parameters are
Vg = 0.6 V
l = 500 nm, tInAsSb = 7 nm,

Drain current (uA/um)

Dit = 1 × 1013 eV−1 cm−2 and 300
teq = 1.5 nm. Solid symbols
represent simulation data and
open symbols for 200
experimental results
100

0.0 0.2 0.4 0.6 0.8

Drain voltage (V)

of the drain bias. It is followed from the graph that the results obtained from our
model matches well with the experimental data which ensures validity of our
model. Figure 4 shows the variation of drain current in both the linear and log scales
with gate to source voltage for InAsSb and Si channel devices considering two
different channel lengths. For lower channel length the drain current is higher as
expected but it is observed that instead of having higher values of ON current in
InAsSb channel device, a large amount of OFF current relative to the Si is also
present. High OFF current and high value of subthreshold slope for InAsSb channel
MOSFETs (161 mV/dec for 90 nm channel length) makes the device inferior for
low power applications. Figure 5 depicts the transconductance curves of InAsSb
control devices for a wide range of gate bias voltages to capture the subthreshold,
linear and saturation regions of operation. As follows from the curve that trans-
conductance shows a higher value for InAsSb channel device. But in case of

Fig. 4 Variation of drain 0.0005

current with gate bias for two 1E-4
different channel lengths.
0.0004
Open and closed symbols are 1E-6
Drain current (A/um)

Drain current (A/um)

for Si and InAsSb channel

1E-8
devices respectively. Other 0.0003 l= 90 nm
l = 45 nm
parameters are tInAsSb = 7 nm, 1E-10
Vds = 0.05 V and teq = 1.5 nm 0.0002
1E-12

0.0001 1E-14

1E-16
0.0000
1E-18
-0.5 0.0 0.5 1.0 1.5
Gate voltage(V)
A Simulation Study of Nanoscale Ultrathin-Body … 391

Fig. 5 Plot of
transconductance with gate 0.0005
bias for different channel l = 90 nm
l = 45 nm

Transconductance (S/um)
lengths. Other parameters are 0.0004
tInAsSb = 7 nm, Vds = 0.05 V
and teq = 1.5 nm. Open and 0.0003
closed symbols are for Si and
InAsSb channel devices,
0.0002
respectively
0.0001

0.0000

-0.0001
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2
Gate voltage (V)

InAsSb channel device, transconductance reduces very fast with increasing gate
bias while Si channel device can retain it for a longer time.
Figure 6 depicts variation of threshold voltage roll-off with channel length for
InAsSb channel deices and its Si counterparts. From the graph it is found that short
channel effects is higher for InAsSb channel devices due to its high dielectric
constant and beyond 65 nm it is become dominating. So 65 nm is the scaling limit
for the current structure.

Fig. 6 Variation of threshold

voltage roll-off with channel 0.6
length. Other parameters are
tInAsSb = 7 nm, Vds = 0.05 V 0.5 InAsSb channel
Threshold voltage roll-off

and teq = 1.5 nm. Open and Si channel

closed symbols are for Si and 0.4
InAsSb channel devices,
respectively 0.3

0.2

0.1

0.0

50 60 70 80 90 100
Channel length (nm)
392 S. Bhattacherjee and S. Dutta

3 Conclusion

In this paper first time we have investigated simulation study of short InAsxSb1−x
channel MOSFETs. Predicted simulation results have been verified with reported
experimental datas and a good match has been found among the two which con-
firms the validity of the proposed model. Our investigations show that InAsSb
channel devices yield improved results for ON current and transconductance. But at
the same time short channel effects is also a dominating factor for InAsSb channel
devices specially below 65 nm channel length. InAsSb has a much smaller direct
band gap as compared to Si which gives rise to a higher leakage current. Its higher
dielectric constant degrades electrostatic integrity in short channel devices. Thus
replacement of Si channel by InAsSb channel devices requires a few critical issues
to be addressed in InAsSb channel MOS technology.

References

1. Kim SH, Yokoyama M, Taoka N, Nakane R, Yasuda T, Ichikawa O, Fukuhara N, Hata M,

Takenaka M, Takagi S (2011) Enhancement technologies and physical understanding of
electron mobility in III–V n-MOSFETs with strain and MOS interface buffer engineering. In:
IEEE IEDM Tech Dig 13.4.1
2. Bennett BR, Magno R, Boos JB, Kruppa W, Ancona MG (2005) Antimonide-based compound
semiconductors for electronic devices: a review. Solid State Electron 49:1875–1895
3. Ali A et al (2010) Advanced composite high-k gate stack for mixed anion arsenide antimonide
quantum well transistor. In: IEEE IEDM Tech Dig 6.3.1
4. Ali A, Madan H, Misra R, Agrawal A, Schiffer P, Boos JB, Bennett BR, Datta S (2011)
Experimental determination of quantum and centroid capacitance in arsenide–antimonide
quantum-well MOSFETs incorporating nonparabolicity effect. IEEE Trans Electron Devices
58:1397–1403
5. Ali A, Madan H, Agrawal A, Ramirez I, Misra R, Boos JB, Bennett BR, Lindemuth J, Datta S
(2011) Enhancement-mode antimonide quantum-well MOSFETs with high electron mobility
and gigahertz small-signal switching performance. IEEE Electron Device Lett 32:1689–1691
6. Pal HS, Low T, Lundstrom MS (2008) NEGF analysis of InGaAs Schottky barrier double gate
MOSFETs. In: IEEE IEDM Tech Dig 1
7. Xuan Y, Wu YQ, Shen T, Yang T, Ye PD (2007) High performance submicron inversion-type
enhancement-mode InGaAs MOSFETs with ALD Al2O3, HfO2 and HfAlO as gate dielectrics.
In: IEEE IEDM Tech Dig 637
8. Ali A, Madan H, Barth MJ, Boos JB, Bennett BR, Datta S (2013) Effect of interface states on
the performance of Antimonide nMOSFETs. IEEE Trans Electron Devices 34:360–362
9. Fang H, Chuang S, Takei K, Kim HS, Plis E, Liu C-H, Krishna S, Javey A (2012) Ultrathin-
body high-mobility InAsSb-on insulator field-effect transistors. IEEE Trans Electron Device
Lett 33:2012–2014
Author Index

A Datta, Shreyasi, 235

Aditya, S., 109 Dey, Aritra, 149
Agarwal, Sanjit, 49 Deyasi, Arpan, 353
Ahmed, Rosina, 49 Dutta, Debeshi, 185
Arun, Indu, 49 Dutta, Deepneha, 281
Dutta, Himadri Sekhar, 85
B Dutta, Lakshmi Priya, 379
Bagh, Niraj, 271 Dutta, Meghamala, 281
Balaji, V. Sri, 59 Dutta, Sourav, 281
Banerjee, Annwesha, 3 Dutta, Subhasri, 387
Banerjee, Indranil, 185, 315 Dutta, Trina, 323
Barman, Ananya, 343
Basavaprasad, B., 75 G
Bhattacharjee, Karabi Ganguly, 149 Ghosh, Poulami, 125
Bhattacharjee, Rajat, 175 Ghosh, Sayani, 13
Bhattacharya, Amrita, 293 Gupta, Somsubhra, 3
Bhattacharyya, Saugat, 125, 249
Bhattacharyya, Subhashis, 293 H
Bhattacharyya, Swapan, 353 Harish, V., 109
Bhattacherjee, Sangita, 323
Bhattacherjee, Swagata, 387 J
Bhavanam, S. Nagakishore, 263 Jadhav, D.V., 27
Bhowmik, Prabir, 101
Bhowmik, Shilpi Pal, 215
Biswas, Gopa Roy, 67 K
Kalbande, Dhananjay, 193
Kapgate, Deepak, 193
C Kawadiwale, Ramish B., 27
Chakrabarti, Bipasha, 215 Kewate, Prachi, 225
Chakraborty, Chandan, 39, 49 Khasnobish, Anwesha, 125, 235, 249
Champaty, Biswajeet, 185, 271, 315 Koley, Subhranil, 39
Chatterjee, Sanjoy, 49 Konar, Amit, 235, 249
Chatterjee, Shamba, 161 Kulkarni, Gaurav, 315
Chowdhury, Sayanta Pal, 161
M
D Maity, Swarup, 215
Dan, Tarun Kumar, 207 Majee, Narayan Chandra, 67
Das, Mahuya, 379

© Springer India 2015 393

S. Gupta et al. (eds.), Advancements of Medical Electronics,
Lecture Notes in Bioengineering, DOI 10.1007/978-81-322-2256-9
394 Author Index

Majee, Sutapa Biswas, 67 R

Majumder, Somajyoti, 365 Ravi, M., 75
Malini, M., 307 Routray, Aurobinda, 137
Mamatha, M.N., 333
Mane, Vijay M., 27 S
Manna, Nilotpal, 13 Sadhu, Anup Kumar, 39
Mazumder, Ankita, 125 Sadhu, Pradip Kumar, 101
Midasala, Vasujadevi, 263 Saha, Monjoy, 49
Mishra, Rakesh Kumar, 207 Sailalitha, B., 307
Mishra, Subodh, 137 Sanyal, Goutam, 365
Mitra, Pabitra, 49 Sarkar, Arpita Ray, 365
Mohapatra, Biswajeet, 315 Sarkar, Indranath, 293
Mukherjee, Aroma, 149 Sasmal, Milan, 175
Mukherjee, Nilarun, 85 Sharma, Ashika, 281
Sharma, Gayatri, 281
N Shaw, Laxmi, 137
Nag, Manas K., 39 Siddaiah, P., 263
Nandy, Tanaya, 13 Sikdar, Swati, 281
Nayak, Suraj Kumar, 271 Suryawanshi, Pranali, 225
Neogi, Biswarup, 215
T
P Tibarewala, D.N., 125, 185, 235, 249, 271
Pal, Akash, 149
Pal, Kunal, 185, 271, 315 V
Pal, Nitai, 101 Venkateswara Rao, M., 307
Pal, Palash, 101 Vidhya, S., 59
Panda, Ipsita, 271