README.md

metadata

title: Multi-Class Chest X-Ray Detection
emoji: 🫁
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: true
license: mit

🫁 Multi-Class Chest X-Ray Detection with AST

AI-powered detection of 4 respiratory diseases from chest X-rays

🌟 Features

• ✅ 4 Disease Classes: Normal, Tuberculosis, Pneumonia, COVID-19
• ✅ 87.29% Validation Accuracy
• ✅ 100% Pneumonia Specificity (no TB confusion!)
• ✅ 90% Energy Savings with Adaptive Sparse Training
• ✅ Fast Inference: <2 seconds per X-ray
• ✅ Explainable AI: Clear probability distributions

🎯 Key Achievement

Problem Solved: Previous binary models misclassified pneumonia as TB (30% false positive rate).

Our Solution: Multi-class training distinguishes between all 4 diseases with <5% false positive rate.

Disease	Test Accuracy	Notes
Normal	60%	Some COVID confusion
TB	80%	Strong performance
Pneumonia	100%	Perfect - no TB confusion!
COVID-19	80%	Good detection

🔬 Technology

• Model: EfficientNet-B2
• Training: Adaptive Sparse Training (AST)
• Dataset: COVID-QU-Ex (~33,920 chest X-rays)
• Sparsity: 90% (only 10% neurons active)
• Energy Savings: 90% vs traditional training

⚠️ Important Medical Disclaimer

This is a screening tool for research purposes only, NOT a diagnostic device.

Limitations:

• ❌ NOT FDA-approved for clinical diagnosis
• ❌ Cannot replace professional radiologist review
• ❌ All positive results require laboratory confirmation:
  • TB: Sputum AFB smear, GeneXpert MTB/RIF
  • Pneumonia: Sputum culture, blood tests
  • COVID-19: RT-PCR, rapid antigen test

Proper Use:

• ✅ Preliminary screening only
• ✅ Always consult healthcare professionals
• ✅ Confirm with clinical correlation and lab tests

Do not make medical decisions based solely on this tool.

📊 Performance Metrics

Metric	Value
Overall Accuracy	87.29%
Energy Savings	90%
Activation Rate	10%
Training Epochs	50
Inference Time	<2 seconds

🚀 How It Works

1. Upload a chest X-ray image (PNG, JPG)
2. Analyze - AI processes in <2 seconds
3. Review probability distribution for all 4 diseases
4. Confirm with professional medical evaluation

📈 Model Evolution

• v1.0 (Beta): Initial model with EfficientNet-B0 - 87.29% accuracy
• v2.0 (Current): Improved model with EfficientNet-B2 targeting 92-95% accuracy

🔗 Links

• GitHub: oluwafemidiakhoa/Tuberculosis
• Training Notebook: TB_MultiClass_Complete_Fixed.ipynb
• Documentation: Full README

👨‍💻 Developer

Oluwafemi Idiakhoa

• GitHub: @oluwafemidiakhoa
• Hugging Face: @mgbam

📄 License

MIT License - Free for research and educational use

---

Powered by Adaptive Sparse Training - Energy-efficient AI for accessible healthcare 🌍