Tuberculosis Detection Using Hybrid CNN-ViT Architecture
A hybrid deep learning framework combining ResNet-based feature extraction with Vision Transformer attention mechanisms for automated tuberculosis screening from chest X-ray images.
Problem Statement
Standard CNN models struggle to capture long-range dependencies in chest X-ray images relevant to tuberculosis detection. Vision Transformers require large datasets but outperform CNNs when sufficient data is available.
Method & Approach
We design a two-branch architecture: a CNN backbone (ResNet-50) for local texture and edge features, and a ViT module for global relational reasoning. Features are fused via a learned attention gate before classification.
Dataset / Data Source
Montgomery County X-ray Dataset, Shenzhen Hospital Chest X-ray Dataset (publicly available from NIH).
Results
Achieved competitive sensitivity and specificity compared to standalone CNN and ViT baselines on the test split. Grad-CAM visualizations show attention focused on lung fields consistent with TB pathology.
My Contribution
Full architecture design, training pipeline, hyperparameter tuning, XAI evaluation, and manuscript drafting.