Skin Lesion AI Analysis System

Overview

A full-stack application that leverages advanced Deep Learning (CNNs), Explainable AI (GradCAM), and Image Segmentation to analyze skin lesions directly from uploaded images. The system predicts risks for various skin conditions, such as Melanoma, and provides actionable insights with visual explanations, empowering users with a reliable tool for preliminary dermatological screening.

🚀 Key Features

• Unified ML Pipeline: End-to-end processing integrating Image Classification (ResNet/CNNs), GradCAM heatmap generation, Lesion Segmentation, and a specialized Risk Scoring algorithm.
• Explainability (XAI): Visual heatmaps overlay exactly where the model focuses when making predictions, increasing trust and interpretability.
• Full-stack Integration: Secure user authentication (JWT), history tracking, and interactive clinical dashboards using React.
• Robust Machine Learning: Engineered using widely-recognized medical datasets (HAM10000) and evaluated on the PH2 dataset.

🏗 System Architecture

• Frontend: React.js 19, Vite, Tailwind CSS (Handles user interface, secure image uploading, and interactive charts).
• Backend: FastAPI & Uvicorn (Robust Python REST API to handle processing requests, run ML inference pipelines securely, and validate users).
• Machine Learning Layer: PyTorch & Torchvision pipelines designed for high-accuracy skin lesion classification and pixel-wise segmentation.
• Database: MongoDB & Motor (Stores scan histories securely with asynchronous I/O access).

🛠 Tech Stack

Component	Technologies Used
Frontend	React, Vite, Tailwind CSS, Recharts, Framer Motion
Backend	Python 3.9+, FastAPI, PyMongo, Motor, Passlib, Uvicorn
Data Sci / ML	PyTorch, Torchvision, OpenCV, Scikit-Learn, NumPy, Pillow
Database	MongoDB

⚙️ Local Setup & Installation

Prerequisites

• Python 3.9 or higher
• Node.js & npm
• MongoDB (Running locally or a URI for an Atlas cluster)

Backend Setup

1. Open a terminal and navigate to the project root.
2. Initialize a virtual environment and install dependencies:

   python -m venv venv
   source venv/bin/activate  # On Windows use: venv\Scripts\activate
   pip install -r requirements.txt

3. Set up the .env file in the backend/app/ (or root) directory pointing to your MongoDB instance (if required).
4. Run the FastAPI server:

   cd backend
   uvicorn app.main:app --reload --port 8000

   The backend should now run at http://localhost:8000.

Frontend Setup

1. Open another terminal in the frontend directory:

   cd frontend
   npm install

2. Start the Vite development server:

   npm run dev

   The application UI will be available at http://localhost:5173.

📁 Repository Structure

• backend/: FastApi application files, database initialization, routes, user authentication, and API controllers.
• frontend/: React components, views, layout styles, and client utilities.
• ml_pipeline/: Dedicated scripts for model training, HAM10000 processing, segmentation testing, and overall risk pipelines.
• PH2_Explainability/: Data outputs focusing on explainability layers and GradCAM inference derived from the PH2 dataset.

🤝 Contributors

• Developed collaboratively with Pranav-Dawange and ronitjain7.