DERMAI: SKIN DISORDER DETECTION AND

TREATMENT ADVISOR POWERED BY RESNET50

WITH GRAD-CAM VISUALIZATION

PRESENTED BY: NIVETHA. R(210421121033)

PURNAA. A (210421121036)
SARANYA. A (210421121043)

GUIDED BY: MR. K. PRADEEP

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 PROBLEM STATEMENT

Skin diseases are a significant global health concern, affecting millions

annually and contributing to reduced quality of life, psychosocial distress, and,
in severe cases, mortality. Early and accurate diagnosis of skin diseases is
crucial for effective treatment and prevention of complications. However,
challenges like overlapping visual symptoms, misdiagnosis due to limited
access to specialists, and lack of explainable diagnostic systems persist .

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 ABSTRACT

DermAI is an advanced system designed to detect skin disorders and recommend

treatments using the ResNet50 deep learning model. The system analyzes skin
images to identify various skin conditions, such as acne, melanoma, and psoriasis,
with high accuracy. To enhance interpretability,Grad-CAM visualization is
integrated, providing insights into the areas of the skin image that contribute to
the model’s predictions. DermAI offers a reliable, user-friendly platformfor both
early detection and treatment advice, improving healthcare accessibility and
aiding in timely interventions for skin-related issues.

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 LITERATURE REVIEW
S.No Paper Title/Year Author Name Discussion
1. Skin Lesion Classification Ethiopian Artificial Skin lesions are essential for the early detection
and Detection Using Intelligence Institute, Addis and management of a number of dermatological
Machine Learning Ababa 40782, Ethiopia; disorders. Learning-based methods for skin lesion
Techniques: A Systematic [email protected] analysis have drawn much attention lately. A
Review review of the most-recent methods for skin lesion
classification, segmentation, and detection is
presented in this survey paper.
2. An AI-Assisted Skincare Gowravi Malalur This paper presents an AI-powered skincare
Routine Recommendation Rajegowda1 , Yannis recommendation system integrated into an
System in XR Spyridis2 , Barbara Villarini3 XR platform. It uses a CNN to analyze skin
and Vasileios Argyriou1 type and recommend personalized products
based on facial image data. The system
achieves 93% accuracy in identifying skin
issues. This innovative approach enhances
the beauty industry by offering immersive
and personalized skincare experiences.

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 LITERATURE REVIEW
S.No Paper Title/Year Author Name Discussion
3. Deep learning models Kaushik P. Venkatesh, This paper explores the application of deep
across the range of skin Marium M. Raza, learning models for diagnosing a range of
disease Grace Nickel, skin diseases. Using large datasets of skin
Serena Wang & images, the models are trained to accurately
Joseph C. Kvedar
identify various conditions. The study
demonstrates high accuracy in disease
classification, improving diagnostic
efficiency.

4. Deep learning-based skin care Jinhee Lee MS, Huisu This article aims to propose a method for
product recommendation: A Yoon PhD, Semin Kim PhD estimating the efficacy of cosmetics based
focus on cosmetic ingredient , Chanhyeok Lee MS, on their ingredients and introduces a system
analysis and facial skin that recommends personalized products for
Jongha Lee MS, Sangwook
conditions
Yoo PhD consumers, combined with AI skin analysis.

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 LITERATURE REVIEW
S.No Paper Title/Year Author Name Discussion
5. Ramya G.; This research uses image processing techniques
Skin Disease Detection based on Saanthosh Guru A.; to diagnose skin diseases by classifying images
Machine Learning Techniques Kabilan S.; Ramkaran R. into normal and abnormal conditions. After
deblurring and noise reduction, the system
focuses on identifying conditions like
melanoma and acne as abnormal. Image
segmentation aids in the early detection of
various skin disorders.
6. Cosmetics suggestion system Ian Goodfellow, Yoshua This paper presents a cosmetics suggestion
based on skin conditions Bengio, and Aaron system using deep learning to recommend
Courville products based on individual skin types. It
leverages AI algorithms to analyze data and
predict the best cosmetics for various skin
conditions, such as oily, dry, or neutral. This
approach simplifies the process of selecting
suitable products in the growing beauty
industry.
DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 EXISTING METHODOLOGY
1.Deep Convolutional Neural Networks (CNNs) for Skin Disease Detection:
Methodology: CNNs are commonly used in image classification tasks, including the detection of skin
diseases. Deep CNNs like ResNet50 are used to automatically extract features from skin images and classify
them into different categories (e.g., melanoma, acne, psoriasis).testing, no real-time feedback.

2. Transfer Learning for Skin Disorder Classification

Methodology: Transfer learning is an effective technique where models trained on large datasets like
ImageNet are fine-tuned to classify skin diseases. By using a pre-trained network (e.g., ResNet50), transfer
learning enables faster and more efficient training on smaller database.

3.Grad-CAM (Gradient-weighted Class Activation Mapping) for Model Interpretability

Methodology: Grad-CAM is a visualization technique that helps to interpret and understand the decision-
making process of CNNs. It generates heatmaps highlighting the regions of an input image that are most
influential in the model's prediction.

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 OBJECTIVE
1.Develop a Robust Skin Disease Prediction Model:
Use machine learning techniques, specifically Convolutional Neural Networks (CNNs), to
classify skin diseases accurately. Focus on diseases with high morbidity and mortality (e.g.,
melanoma) and psychosocial impact (e.g., acne).
2.Incorporate Explainable AI (XAI):
Enable the model to provide insights into its predictions (e.g., highlighting features like scaling
or redness), ensuring greater trust and interpretability for clinicians.
3.Address Dataset Challenges:
Mitigate class imbalance and limited diversity in existing datasets through data augmentation and
hierarchical classification strategies.
4.Enable Broader Accessibility:
Design the model for deployment in web-based applications to provide diagnostic support in
underserved regions.

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
USE CASE DIAGRAM

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
WORK IN PROGRESS

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 FUTURE WORKS

1.Expansion of Skin Disorder Database

2.Integration with Telemedicine Platforms
3.Real-time Skin Monitoring with Mobile Integration
4.Personalized Skincare Product Recommendations
5.Incorporation of Multi-modal Data
6.AI-Driven Continuous Learning and Adaptation
7.Enhanced Explainability with Deep Learning Models
8.Collaboration with Dermatologists for Real-world Validation

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 NOVELTY
1. Integration of Advanced Deep Learning and Visualization
DermAI combines ResNet50 for skin disorder detection with Grad-CAM for interpretability, providing both accurate
diagnosis and visual insights. This dual approach enhances model transparency and trust, making it accessible for both
users and healthcare professionals.
2. Real-time, Personalized Treatment Recommendations
Beyond detection, DermAI offers personalized treatment suggestions based on the diagnosed skin condition. This
feature helps users access customized solutions, improving their skin health management and encouraging timely
interventions.
3. Comprehensive
. Multi-Condition Detection
DermAI utilizes a single model to detect a wide range of skin conditions, such as melanoma, acne, and psoriasis. This
multi-class classification ensures users receive a thorough diagnosis, simplifying the process of identifying various skin
disorders.
4. Utilization of Transfer Learning for Efficient Training
By fine-tuning the pre-trained ResNet50 model, DermAI leverages transfer learning, which reduces the need for large
skin disease datasets while still achieving high accuracy. This makes the technology accessible even in data-limited
settings.

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)
 THANK YOU

DEPARTMENT OF BIOMEDICAL
CHENNAI INSTITUTE OF TECHNOLOGY(Autonomous)