DermaAI is an end-to-end medical AI application capable of classifying 12 different skin diseases. It uses a custom-trained EfficientNetB1 model and features a mobile-responsive UI with real-time camera integration.
Live Demo:
- start the backend first - https://dermaai-0h90.onrender.com
- frontend - https://derma-ai-ten.vercel.app
- Multi-Class Diagnosis: Detects 12 conditions including Acne, Melanoma, Eczema, and Psoriasis.
- Advanced AI Model: Powered by EfficientNetB1 using Transfer Learning, trained on a merged dataset of ~16,000 clinical and dermoscopic images.
- Imbalance Handling: Implements automated Class Weights to accurately detect rare diseases (e.g., Cancer) despite dataset imbalance.
- Mobile-First Design: Built-in camera support allows users to analyze skin lesions directly from their phone.
- Smart Analysis: Provides a top prediction along with "Second Best Guesses" and confidence scores.
- Frontend: HTML5, Tailwind CSS, Vanilla JavaScript (Hosted on Vercel)
- Backend: Python 3.10, FastAPI, Uvicorn (Hosted on Render)
- Machine Learning: TensorFlow/Keras, NumPy, Pillow, EfficientNetB1
The model is trained to recognize the following 12 classes:
- Acne
- Actinic Keratoses (Pre-cancerous)
- Basal Cell Carcinoma (Cancer)
- Benign Keratosis
- Dermatofibroma
- Eczema
- Melanocytic Nevi (Mole)
- Melanoma (Cancer)
- Nail Fungus
- Psoriasis
- Ringworm (Fungal)
- Vascular Lesions
Model Performance Validation Accuracy: ~80% Architecture: EfficientNetB1 (Pre-trained on ImageNet, fine-tuned on custom dataset) Loss Function: Categorical Crossentropy with Weighted Classes
Medical Disclaimer This tool is for educational and screening purposes only. The results generated by this AI are probabilistic and do not constitute a medical diagnosis. Users should always consult a certified dermatologist for professional medical advice