🏥 NailHealth AI - Mobile Disease Detection App

AI-powered mobile app that detects systemic diseases through nail photo analysis

Demo Video • Documentation • Report Issue

---

🎯 Project Overview

NailHealth AI uses Google's Health AI Developer Foundations (HAI-DEF) models to detect serious diseases through nail signs captured via smartphone camera. The app provides instant clinical explanations and disease predictions.

🔬 Medical Capabilities

Detects 7 nail conditions indicating systemic diseases:

• White Nails (Terry's Nails) → Liver disease, kidney disease, diabetes
• Blue Nails → Oxygen deficiency, lung disease, heart failure
• Clubbing → Lung cancer, COPD, heart disease
• Spoon Nails (Koilonychia) → Iron deficiency anemia
• Black Lines → Melanoma, trauma, endocarditis
• Psoriasis → Psoriatic arthritis, metabolic syndrome
• Onychogryphosis → Poor circulation, fungal infection

---

🏗️ Architecture

2-Model AI Pipeline

📱 User Photo
    ↓
☁️ Hugging Face Inference API
    ↓
🤖 Model 1: MedSigLIP (Fine-tuned)
   → Classifies nail sign type
    ↓
🤖 Model 2: MedGemma 4B (Fine-tuned)
   → Generates clinical explanation
   → Predicts diseases
   → Recommends actions
    ↓
📊 JSON Response → Mobile App

Tech Stack

Component	Technology
Mobile App	React Native + Expo
Backend API	Flask (Python)
Cloud Hosting	Hugging Face Inference Endpoints
ML Models	MedSigLIP + MedGemma 4B
Storage	Hugging Face Hub
Training	Kaggle GPU (Tesla T4)

---

📁 Repository Structure

NailHealth-AI-Mobile-App/
├── 📱 mobile-app/          # React Native Expo app
│   ├── App.js              # Main app component
│   ├── app.json            # Expo configuration
│   ├── package.json        # Dependencies
│   └── assets/             # Images, icons
│
├── 🔧 api-server/          # Flask API for model inference
│   ├── app.py              # API endpoints
│   ├── Dockerfile          # Container configuration
│   ├── requirements.txt    # Python dependencies
│   └── models/             # Fine-tuned model checkpoints
│
├── 📚 docs/                # Documentation
│   ├── SETUP_GUIDE.md      # Local setup instructions
│   ├── DEPLOYMENT_GUIDE.md # Cloud deployment steps
│   └── ARCHITECTURE.md     # Technical architecture
│
├── 📓 notebooks/           # Training notebooks
│   ├── medsiglip_training.ipynb
│   └── medgemma4b_training.ipynb
│
└── 📋 README.md            # This file

---

🚀 Quick Start

Prerequisites

• Node.js 18+ and npm
• Python 3.10+
• Hugging Face account
• Expo account (free)
• Smartphone (iOS/Android)

1. Clone Repository

git clone https://github.com/isumenuka/NailHealth-AI-Mobile-App.git
cd NailHealth-AI-Mobile-App

2. Setup Mobile App

cd mobile-app
npm install
npx expo start

Scan QR code with Expo Go app on your phone.

3. Setup API Server (Local Testing)

cd api-server
pip install -r requirements.txt
python app.py

API runs at http://localhost:8080

4. Deploy to Hugging Face

See HUGGING_FACE_SETUP.md for detailed steps.

1. Create a model repository on Hugging Face
2. Upload your model files
3. Create an Inference Endpoint
4. Use the provided URL in your app

---

📱 Mobile App Features

Current Features ✅

• 📸 Camera integration with photo capture
• 🔄 Real-time image upload to API
• 🤖 AI-powered nail disease classification
• 📊 Clinical explanation generation
• 🏥 Disease probability ranking
• 📋 Recommended medical tests
• 🎨 Professional medical UI design
• ⚡ Expo Go instant testing
• 📱 Cross-platform (iOS + Android)

Coming Soon 🔜

• 📜 Disease history tracking
• 📈 Progress monitoring over time
• 🌍 Multi-language support
• 🔔 Appointment reminders
• 📤 PDF report export
• 👨‍⚕️ Doctor consultation booking

---

🛠️ Development

Mobile App Development

cd mobile-app

# Start development server
npm start

# Run on specific platform
npm run android  # Android emulator
npm run ios      # iOS simulator
npm run web      # Web browser

API Server Development

cd api-server

# Run with auto-reload
export FLASK_ENV=development
python app.py

# Build Docker image locally
docker build -t nailhealth-api .
docker run -p 8080:8080 nailhealth-api

Model Training

Training notebooks are in notebooks/ folder:

1. MedSigLIP Fine-tuning: Nail image classification
2. MedGemma 4B Fine-tuning: Clinical explanation generation

Both trained on Kaggle with Tesla T4 GPU (free tier).

---

📊 Performance Metrics

Metric	Value
Nail Sign Classification Accuracy	91.2%
F1-Score (Weighted)	0.89
API Response Time	~2.1 seconds
App Load Time	<1 second
Supported Image Formats	JPG, PNG
Max Image Size	5 MB

Model Details

MedSigLIP Fine-tuned

• Base: google/medsiglip-448
• Parameters: 400M
• Training: LoRA (Low-Rank Adaptation)
• Dataset: 700+ nail images (custom collected)

MedGemma 4B Fine-tuned

• Base: google/medgemma-4b-it
• Parameters: 4B
• Training: Instruction fine-tuning
• Dataset: 250+ clinical text pairs

---

🌐 Deployment Options

Recommended: Hugging Face Inference Endpoints

We recommend using Hugging Face Inference Endpoints for the easiest setup without managing complex cloud infrastructure.

• ✅ No Google Cloud required
• ✅ specific hardware selection (CPU/GPU)
• ✅ Auto-scaling (including scale-to-zero to save money)
• ✅ secure & private

👉 Follow the Hugging Face Setup Guide to get started.

---

📖 Documentation

Document	Description
SETUP_GUIDE.md	Complete local development setup
HUGGING_FACE_SETUP.md	Cloud deployment instructions
ARCHITECTURE.md	Technical architecture details

---

🔑 Environment Variables

Mobile App (mobile-app/.env)

API_URL=https://your-api-url.run.app
ENVIRONMENT=production

API Server (api-server/.env)

MODEL_PATH=/app/models
PORT=8080

---

📈 Roadmap

Phase 1: MVP ✅ (Completed)

• MedSigLIP fine-tuning
• MedGemma 4B fine-tuning
• Flask API development
• React Native mobile app
• Hugging Face deployment
• Basic UI/UX

Phase 2: Enhancement 🔄 (In Progress)

• Improve classification accuracy to 95%+
• Add disease history tracking
• Implement user authentication
• Create admin dashboard
• Multi-language support

Phase 3: Scale 📅 (Planned)

• Clinical validation study
• Regulatory compliance (FDA/CE)
• Doctor consultation integration
• App Store publication

---

🤝 Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository
2. Create feature branch (git checkout -b feature/AmazingFeature)
3. Commit changes (git commit -m 'Add AmazingFeature')
4. Push to branch (git push origin feature/AmazingFeature)
5. Open Pull Request

---

📜 License

This project uses models from Google's Health AI Developer Foundations (HAI-DEF):

• MedSigLIP: Apache 2.0 License
• MedGemma 4B: Apache 2.0 License

Project License: Apache 2.0

---

⚠️ Medical Disclaimer

IMPORTANT: This application is for educational and research purposes only.

• ❌ NOT a substitute for professional medical advice
• ❌ NOT approved by FDA or medical authorities
• ❌ NOT for clinical diagnosis
• ✅ Use only as a screening tool
• ✅ Always consult licensed healthcare professionals

---

🙏 Acknowledgments

• Google Health AI: For HAI-DEF models (MedSigLIP, MedGemma)
• Kaggle: For free GPU resources
• Expo Team: For amazing mobile development framework
• Medical Community: For nail disease datasets and research

---

📞 Contact

Developer: K.G.I Enuka
GitHub: @isumenuka
Twitter: @ezsumm
Website: isumenuka.me

Project Link: https://github.com/isumenuka/NailHealth-AI-Mobile-App

---

⭐ Star this repo if you find it helpful!

Made with ❤️ by K.G.I Enuka