Upload a skin photo and the AI will classify it into one of 7 lesion types.

🚀 Try the Live Demo

SkinDx is an AI-powered web app that uses a fine-tuned ResNet-50 CNN model trained on the HAM10000 dataset to classify skin lesions into 7 categories.

• Upload skin lesion images → get predicted class and confidence.
• Supports 7 skin lesion types (akiec, bcc, bkl, df, mel, nv, vasc).
• Confidence breakdown visualization with urgency indicators.

1. 📸 Upload a skin lesion photo
2. 🔍 Click Analyze
3. 🤖 AI model (ResNet-50) predicts lesion type + confidence
4. 📊 Results shown in the app (class, risk, confidence, chart)

⚠️ Disclaimer: This tool is not a medical device. Results are for educational purposes only – always consult a dermatologist.

• Python 3.11.x (not 3.12+)
• uv package manager
• Docker Desktop
• Make: winget install GnuWin32.Make (Windows) | brew install make (macOS) | sudo apt install make (Linux)
• GPU – model trained on RTX 5080

Runs with Docker Desktop (WSL2 on Windows recommended). This ensures consistent environments across systems.

Components:

• Streamlit → Web UI
• FastAPI → Model inference API
• Jupyter/MLflow → Notebooks & experiment tracking

Everything runs together via Docker Compose (already included in Docker Desktop).

Start with:

docker compose up --build

Or, if you prefer Makefile shortcuts:

make up

• Streamlit UI → http://localhost:7860
• FastAPI Docs (Swagger UI) → http://localhost:8000/api-docs
• Jupyter/MLflow (notebooks) → http://localhost:8888

  ⚠️ The first time you start notebooks, Jupyter requires an access token. Copy the full URL with ?token=... from the container logs (shown in the terminal).

• main.py → FastAPI inference service
• src/ui/ → Streamlit frontend
• src/model_loader.py → HuggingFace model loader
• notebooks/ → Data exploration, training & evaluation
• tests/ → Pytest unit/integration tests
• Model hosted on HF: https://huggingface.co/lisekarimi/resnet50-ham10000

1. Data exploration (HAM10000 distribution)
2. Class imbalance handling (smart augmentation)
3. ResNet-50 training & tracking with MLflow
4. External validation on ISIC 2019 dataset

This project includes a full CI/CD pipeline (tests, linting, security scans, deployment in Hugging Face).

Additional details are available here.