Project Report: DermAssist - AI-Driven Diagnostic Support for Skin Lesion Evaluation

DOMAIN: Machine learning for biomedical image analysis and diagnosis

1. Project Topic:

DermAssist: AI-Driven Diagnostic Support for Skin Lesion Evaluation

Objective:
This project focuses on the development of an AI-based web application aimed at assisting
healthcare professionals in diagnosing skin lesions, such as melanoma, basal cell carcinoma, and
benign lesions, using dermoscopic images. The system applies image preprocessing techniques
like noise reduction, contrast enhancement, segmentation, and masking, followed by AI-based
image analysis to provide diagnostic results.

2. Team Members' Names:

● MURUGAN S (23I338)
● SANGAVI K (23I351)
● SWETHA T (23I366)

3. Faculty Details:
SANGEETHA B
DEPARTMENT OF INFORMATION TECHNOLOGY
PSG COLLEGE OF TECHNOLOGY
PHONE NO: 9894940664
EMAIL ID: bsg.it@psgtech.ac.in

4. Workflow & Current Progress:

Workflow:
The workflow involves several key stages:

1. Data Collection:
We selected appropriate datasets for skin disease detection, including:

○ ISIC (International Skin Imaging Collaboration) Dataset

○ HAM10000 Dataset
These datasets contain high-quality dermoscopic images labeled for various skin
conditions, including melanoma.
2. Image Preprocessing:
We implemented preprocessing techniques to enhance the quality of images before
passing them to the AI model:

○ Noise Reduction: Applied filters such as Gaussian and median filtering to

remove unwanted noise from images.
○ Contrast Enhancement: Utilized methods like histogram equalization and
adaptive histogram equalization (CLAHE) to enhance lesion visibility.
○ Segmentation: Used techniques to identify and isolate regions of interest
(lesions) in the image, such as thresholding and edge-based methods.
○ Masking: Applied binary masks to further isolate lesions for improved analysis.
3. AI Model Development:
We used Convolutional Neural Networks (CNNs) for classifying skin lesions. We
leveraged pre-trained models such as ResNet50 for transfer learning, fine-tuning it to
identify melanoma and other skin conditions from the processed images.

4. Website Development:
A web application was developed to allow users to upload images and receive diagnostic
results from the AI model.

Current Progress:

● Image Preprocessing: Successfully completed noise reduction, contrast enhancement,

segmentation, and masking on the datasets.
● Model Training: The AI model is currently being trained using pre-processed datasets
and transfer learning. The model is designed to classify images into different categories
of skin lesions, including melanoma.
● Website: The basic web interface is complete, with image upload functionality working
and diagnostic results being generated using the AI model.

5. Individual Contributions:

● Swetha T:
Led the image preprocessing stage, including noise reduction, contrast enhancement,
and segmentation. Played a key role in integrating the preprocessing pipeline with the AI
 model.

● Murugan S:
Developed the web interface, ensuring that users can easily upload images and receive
diagnostic results. Worked on integrating the front-end with the back-end AI model.

● Sangavi K:
Focused on AI model development, using pre-trained networks like ResNet50 for
transfer learning. Contributed to model fine-tuning and training on the dataset.

6. Strategies to Solve the Problem:

● Preprocessing:
Preprocessing techniques like noise reduction, contrast enhancement, and image
segmentation are essential for improving the quality of input images. These steps ensure
that the model can focus on the regions of interest, such as lesions, leading to more
accurate results.

● AI Model:
By using transfer learning with a pre-trained model like ResNet50, we leverage the
power of existing deep learning models, reducing training time and improving accuracy.
Fine-tuning the model specifically for skin lesion detection allows us to obtain high-
quality results tailored to our use case.

● Web Application:
Developing a user-friendly interface allows healthcare professionals to easily upload
dermoscopic images and receive AI-based diagnostic results, enhancing the usability and
accessibility of the tool.

7. Technologies Used:

● Programming Languages:
○ Python: Used for backend image processing, AI model development, and
integration.
○ HTML/CSS/JavaScript: Used for frontend web development.
● Frameworks/Tools:
○ TensorFlow/Keras: For AI model development using Convolutional Neural
Networks (CNNs) and transfer learning with models like ResNet50.
○ Flask/Django: For backend development and API creation to handle image
uploads and AI inference.
○ OpenCV: Used for image preprocessing tasks like noise reduction, contrast
enhancement, segmentation, and masking.
 ○ PIL (Pillow): For basic image processing tasks, including resizing, cropping, and
format conversion.
● Web Technologies:
○ Bootstrap: For responsive design of the web application.
○ JavaScript (AJAX): For asynchronous image upload and results display on the
web interface.
● Deployment:
○ Heroku or AWS: For deploying the web application.
○ Docker: Used to containerize the application for consistent deployment across
environments.

Conclusion:
The project is progressing well with completed preprocessing steps and an AI model under
development. The website is functional for uploading images and generating diagnostic results,
and the AI model is being trained with promising results. With the integration of image
preprocessing techniques, transfer learning for AI, and a user-friendly interface, DermAssist will
significantly aid healthcare professionals in skin disease diagnosis.

This AI-based system has the potential to assist in the early detection of skin diseases,
particularly melanoma, and will be a valuable tool for dermatologists worldwide.

This report gives a clear overview of the project, detailing the workflow, contributions, and
technologies used. You can fill in team members’ names and supervisor details as needed. Let
me know if you would like to make further adjustments!