Home

TB Detection using Segmentation, Classification, and Explainable AI

This project presents an AI-based tuberculosis (TB) detection system that integrates lung segmentation, disease classification, and explainable AI for interpretable medical imaging.

The pipeline is composed of three main stages:

1. Segmentation – isolates lung regions from chest X-rays using the ResUNet++ architecture.
2. Classification – classifies the cropped lung regions into Normal or TB-positive using EfficientNet-B0.
3. Explainable AI (XAI) – uses Grad-CAM to visualize model attention regions, helping clinicians interpret AI decisions.

The solution is designed to assist radiologists, improve diagnostic accuracy, and enable faster TB screening in under-resourced areas.

Key Results

Metric	Segmentation (ResUNet++)	Classification (EfficientNet-B0)
Dice Coefficient	0.96	—
Jaccard Index	0.93	—
Accuracy	0.98	0.79
Precision	0.97	0.80
Recall	0.96	0.79

Live Demo: TB Detection Web App
GitHub Repository: View on GitHub