MelanoAI is a deep learning-powered web application designed for preliminary classification of skin lesion images into three categories: benign, malignant, and invalid (non-skin images). It features a clean Streamlit UI, model explainability via Grad-CAM, and an integrated logging system.
⚠️ Disclaimer
This is the first version of MelanoAI. The current model is not clinically validated and may produce unreliable predictions. It should not be used for any medical diagnosis or decision-making.
Use for research, demonstration, and educational purposes only.
- 🖼️ Upload and classify skin lesion images (
benign,malignant,invalid) - 📈 Model confidence display
- 🔥 Grad-CAM heatmap overlay for interpretability
- 🧾 Prediction logging system (CSV)
- 🔐 Optional user authentication with SQLite
- 📄 Downloadable prediction report (PDF)
- 📦 Modular architecture, ready for production-scale improvements
- Model: Custom CNN with optional MobileNetV2 backbone
- Input Shape:
(224, 224, 3) - Classes:
benign,malignant,invalid - Trained on: HAM10000 dataset + 1000 curated invalid images
- Training strategy:
- Data Augmentation
- Class Balancing
- EarlyStopping
- Grad-CAM support via final convolutional layers
- High false positive rate for
malignanton invalid images - Misclassification of real benign cases as malignant
- Overfitting risk due to limited data diversity
- Grad-CAM may not always align with clinically relevant regions
- Currently lacks robustness for real-world generalization
This version is primarily for demonstration and portfolio use — real medical deployment would require extensive retraining, augmentation, testing, and regulatory compliance.
git clone https://github.com/your-username/MelanoAI.git cd MelanoAI
- Create a virtual environment:
python -m venv .venv
.venv\Scripts\activate
source .venv/bin/activate
pip install -r requirements.txt
streamlit run app/main.py