DermaScan AI

A web app that classifies 7 skin conditions including Melanoma using EfficientNet-B0 + Flask.

Not a medical diagnosis. Always consult a dermatologist.

What it does

• Classifies 7 skin conditions with confidence score
• Rejects poor quality images before inference
• Returns severity, symptoms & precautions per condition
• Drag & drop UI with dark/light mode

Tech Stack

Python, Flask, PyTorch, EfficientNet-B0 (timm), Vanilla JS

Run Locally

git clone https://github.com/yourusername/dermascan-ai
cd dermascan-ai
pip install -r requirements.txt
python app.py

Model

EfficientNet-B0 fine-tuned on HAM10000 dataset (10,015 dermoscopic images, 7 classes) achieving 82% validation accuracy. AUC-ROC used over accuracy due to class imbalance.

Roadmap

• REST API + image quality validation
• 7-class disease info system
• Responsive frontend
• Fine-tuned on HAM10000 — 82% val accuracy
• Grad-CAM heatmap
• Docker + deployment