SRM INSTITUTE OF SCIENCE AND TECHNOLOGY
SCHOOL OF COMPUTING
DEPARTMENT OF DATA SCIENCE AND BUSINESS
SYSTEMS
18CSP107L - MAJOR PROJECT
SKIN DISEASE DETECTION
AND REMEDIAL SYSTEM
RA2011027010066
HIMANI KALRA
RA2011027010082
VISHAL SUGUR
ABSTRACT
• Traditional skin disease diagnosis relies on visual examination by
dermatologists, but a CNN model shows high accuracy in
distinguishing conditions, revolutionizing dermatology.
• Extensive validation against diverse real-world cases ensures the
CNN system's clinical utility, promising reliable and timely
diagnosis for effective treatment.
• Deep learning's potential in dermatology shines through the
remarkable accuracy of the trained model, offering hope for
improved diagnosis and treatment outcomes.
2
PROBLEM STATEMENT
• Skin diseases pose a global health challenge, demanding
enhanced diagnostic accuracy and accessibility. Current
methods relying on subjective visual assessments lead to
errors and long wait times for dermatologist appointments.
To mitigate these issues, a robust skin diseases detection
system leveraging machine learning and computer vision is
needed. It should prioritize accuracy, user-friendliness, and
data security to revolutionize dermatological care.
3
LITERATURE REVIEW
Title Author Year Journal Algorithm Key Findings
used
Skin Disease Isahu Ram 2023 International Random It addresses accurate classification
Classification Sahu, Conference forest,Support across multiple skin conditions,
using Machine Akhilesh for Emerging vector utilizing a dataset of 750 images
Learning Kumar Technology machine,naive with preprocessing techniques for
based Shrivas, (INCET) s bayes,KNN enhanced performance. The
Proposed Abhinav proposed model , signals a
Ensemble Shukla potential advancements in
Model dermatological diagnosis and
treatment via machine learning.
Artificial Esteva,et al 2023 Dept. of CNN This research paper classifies skin
Neural Computer cancers with proficiency
Networks for Science and comparable to dermatologists,
Dermatologist- Engineering, leveraging diverse datasets for
level Skin Bangladesh meticulous training. The findings
Cancer University of highlight AI-driven tools' potential
Classification Business and to improve diagnostic accuracy in
Technology dermatology
4
LITERATURE REVIEW
Title Author Year Journal Algorithm Key Findings
used
Skin Disease Kumar Ojha, 2022 2022 IEEE CNN and The paper is achieving a
Detection and M., Karakattil, India Council MobileNet validation accuracy of 73.8%.
Classification D. R., & International The system, encompassing 6 skin
Bency, S. M Subsections conditions with augmented data,
Conference provides a user-friendly approach
(INDISCON) for self-testing and proactive
disease management.
Multi- Athina, F. 2022 2022 ResNetV2, The study uses LIME(Local
classification H., et al International CNN Interpretable Model-Agnostic
Network for Conference Explanations) for explaining
Detecting on model predictions, enhancing
Skin Diseases Innovation, transparency. Additionally, a
using Deep Intelligence custom 10-layered CNN
Learning and for architecture and a web
XAI Informatics application using ResNet50V2
and are developed for accurate and
Computing. user-friendly skin disease
diagnosis.
5
LITERATURE REVIEW
Title Author Year Journal Algorithm used Key Findings
"Deep Haenssle, H.A. 2021 Int J Mol Sci. CNN,LBP This study proposes a
Learning et al. 2021 collaborative learning
Models approach merging CNN and
Competing LBP for precise skin disease
with detection, achieving 93.60%
Dermatologists accuracy and 77.32%
in validation accuracy on a well-
Dermoscopic known dataset. Its findings
Melanoma emphasize the significance of
Recognition" integrated methods in
enhancing early and accurate
diagnosis amid increasing
data volumes in dermatology.
Human Skin S Ahmed, et al 2021 EAI Endorsed CNN The paper addresses data
Diseases Transactions scarcity and sample imbalance
Detection and on challenges. Leveraging deep
Classification Pervasive learning, it shows promise in
using CNN Health and improving diagnosis accuracy,
Technology emphasizing the importance
2021 of early detection for
preventing severe outcomes. .
6
SYSTEM
REQUIREMENTS
FUNCTIONAL REQUIREMENTS:
• Data collection and pre-processing
• Model Architecture
• Feature Extraction and Representation
• Classification and Prediction
NON-FUNCTIONAL REQUIREMENTS:
• User-friendly interface
• Accuracy and reliability
• Performance Metrics
7
EXISTING SOLUTION
• AI aids in skin disease detection by accurately identifying and
categorizing lesions, facilitating early diagnosis and classification of
dermatological conditions.
• Precise lesion segmentation separates affected areas, improving
diagnostic accuracy and treatment planning.
• AI's preprocessing procedures refine data, reducing noise and
enhancing images for more reliable skin disease detection outcomes.
8
PROPOSED
METHODOLOGY
• Utilize EfficientNet for skin disease detection, leveraging its
superior performance in image classification, ensuring accurate
identification of various dermatological conditions.
• Gather a diverse skin disease dataset for effective model training,
encompassing a wide range of conditions to enhance classification
accuracy.
• Preprocess data to ensure uniform image quality, format, and
resolution, optimizing model training, and evaluation for reliable
skin disease classification.
9
MODULE DESCRIPTION
• Data Preprocessing: Resize, normalize, and augment images.
Remove backgrounds, convert color spaces, and reduce noise to
enhance image quality for accurate detection.
• Data Loading & Augmentation: Efficiently load datasets and apply
augmentation techniques like rotation, flipping, zooming, and
shifting to diversify training data.
• Model Architecture Definition: Define CNN architecture,
experiment with layers (convolutional, pooling, fully connected),
and activation functions for optimal performance.
10
MODULE DESCRIPTION
• Training: Feed batches of preprocessed data, compute loss,
and update model parameters using optimization algorithms.
Monitor progress and save checkpoints.
• Evaluation: Assess model performance on validation/test data,
compute metrics (accuracy, precision, recall, F1-score), and
generate reports and visualizations for analysis.
• User Interface: Develop GUI or web app for users to upload
images, display predictions, confidence scores, and information
about detected skin conditions for seamless interaction.
11
ARCHITECTURE
DIAGRAM
EfficientNet CNN Architecture
IMAGE CLASSIFICATION
Melanoma Seborrheic keratosis Psoriasis
12
EfficientNetB3 Architecture
13
RESULT ANALYSIS
14
RESULT ANALYSIS
15
RESULT ANALYSIS
16
JOURNAL PUBLICATION
We published our paper in the International Journal Of Innovative Science and Research Technology on
10th may, 2024.
17
THANK YOU!
18
