Oral Cancer Detection - Implementation Guide

Prerequisites

• Python 3.8 or higher
• pip package manager
• Virtual environment (recommended)

Installation

# 1. Create virtual environment
python -m venv venv

# 2. Activate virtual environment
# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

# 3. Install dependencies
pip install -r requirements.txt

Project Structure

oral-cancer-detection/
├── data/
│   ├── train/
│   │   ├── cancerous/
│   │   └── non_cancerous/
│   ├── validation/
│   │   ├── cancerous/
│   │   └── non_cancerous/
│   └── test/
│       ├── cancerous/
│       └── non_cancerous/
├── models/
│   ├── best_model.h5
│   └── unet_model.h5
├── uploads/
├── results/
├── logs/
├── templates/
│   └── index.html
├── improved_oral_cancer_detection.py
├── improved_flask_app.py
├── requirements.txt
└── README.md

---

📊 Training Models

1. Prepare Your Dataset

Organize your dataset in the following structure:

data/
├── train/
│   ├── cancerous/       # Images of oral cancer
│   └── non_cancerous/   # Images of healthy mouths
├── validation/
│   ├── cancerous/
│   └── non_cancerous/
└── test/
    ├── cancerous/
    └── non_cancerous/

Recommended Dataset Size:

• Minimum: 500 images per class
• Good: 1000+ images per class
• Excellent: 2000+ images per class

2. Train VGG16 Model

3. Train Alternative Models

4. Train U-Net for Segmentation

🧪 Evaluation (Comprehensive Evaluation)

Performance Targets

Metric	Minimum	Good	Excellent
Accuracy	85%	90%	95%
Sensitivity	90%	95%	98%
Specificity	85%	90%	95%
ROC AUC	0.90	0.95	0.98

---

🌐 Running the Web Application

1. Start the Flask Server

python improved_flask_app.py

The server will start at http://localhost:5000

2. API Endpoints

Health Check

curl http://localhost:5000/health

Make Prediction

curl -X POST -F "file=@path/to/image.jpg" http://localhost:5000/predict

Get Statistics

curl http://localhost:5000/stats

3. Web Interface

Open your browser and navigate to:

http://localhost:5000

Upload an image and get instant predictions with:

• Classification result (Cancerous/Non-Cancerous)
• Confidence score
• Medical recommendation
• Grad-CAM visualization (what the AI is looking at)
• Segmentation mask (if cancerous)

---

File Upload Constraints

• Maximum file size: 16MB
• Allowed formats: PNG, JPG, JPEG
• Secure filename handling
• File integrity checks (SHA256)

Data Privacy

• No data is stored permanently without consent
• All predictions are logged with anonymized IDs
• HIPAA compliance considerations built-in

For questions or issues:

1. Check this README
2. Review code comments
3. Check logs in logs/app.log
4. Create an issue on GitHub

---

⚖️ Legal & Compliance

IMPORTANT DISCLAIMERS:

1. Not Medical Advice: This system provides AI predictions only and is NOT a substitute for professional medical diagnosis.

2. Experimental: This is a research/educational project. Not FDA approved or clinically validated.

3. Liability: Users assume all risks. Consult healthcare professionals for actual diagnosis.

4. Data Privacy: Comply with HIPAA, GDPR, and local regulations when handling medical data.

5. Informed Consent: Users must understand AI limitations before using the system.

---

📄 License

This project is for educational and research purposes only.

---

🙏 Acknowledgments

• Inspired by clinical need for early cancer detection

---

Remember: The goal is to ASSIST healthcare professionals, not replace them. Always prioritize patient safety and ethical AI use.