🫁 Tuberculosis Detection with Explainable AI (XAI)

📌 Overview

This project implements a deep learning-based Tuberculosis (TB) detection system using chest X-ray images from the TBX11K dataset.
To enhance trust and transparency in medical AI, the project integrates Explainable AI (XAI) techniques (such as Grad-CAM) to visualize which regions of the X-ray images influenced the model's predictions.

📊 Dataset

• Name: TBX11K Chest X-ray Dataset
• Content: X-ray images labeled as TB-positive or TB-negative
• Source: Public medical imaging dataset
• Preprocessing:
  • Image resizing and normalization
  • Data augmentation (rotation, flipping, zooming, shifting)
  • Train-validation-test split

🛠 Tech Stack

• Programming Language: Python
• Libraries:
  • Deep Learning: TensorFlow, Keras
  • Data Processing: pandas, numpy, cv2
  • Visualization: matplotlib
  • Evaluation: scikit-learn
  • XAI: Grad-CAM implementation

⚙ Methodology

1. Data Loading & Preprocessing
   • Extract and organize TBX11K dataset
   • Resize and normalize chest X-ray images
2. Data Augmentation
   • Applied transformations to reduce overfitting
3. Model Architecture
   • Custom CNN / Transfer Learning backbone
   • Classification head for binary TB detection
4. Training & Evaluation
   • Binary classification (TB-positive / TB-negative)
   • Metrics: Accuracy, Precision, Recall, F1-score
5. Explainable AI
   • Grad-CAM heatmaps to highlight decision-making regions
6. Visualization
   • Training history plots
   • Confusion matrix
   • Grad-CAM overlays

📈 Results

• High Recall for TB-positive cases to minimize false negatives
• Grad-CAM visualizations showing the lung regions most relevant to predictions