Enhanced tuberculosis detection using Vision Transformers and explainable AI with a Grad-CAM approach on chest X-rays

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Pathology

Background:

• Tuberculosis (TB) diagnosis from chest X-rays is crucial but challenging due to subtle disease manifestations.
• Traditional methods using Convolutional Neural Networks (CNNs) require extensive pre-processing and lack generalizability.
• Accurate and interpretable AI tools are needed for early TB detection, especially in resource-limited settings.

Purpose of the Study:

• To develop and evaluate a novel Vision Transformer (ViT) model integrated with Gradient-weighted Class Activation Mapping (Grad-CAM) for enhanced TB diagnosis from chest X-rays.
• To improve diagnostic accuracy and model interpretability compared to existing computational methods.
• To facilitate clinical application and assist radiologists in automated TB detection.

Main Methods:

• A Vision Transformer (ViT) model with a Conv2D stem and transformer encoder blocks was developed to process raw X-ray pixels directly.
• Self-attention mechanisms in the ViT model were utilized to capture long-range dependencies and complex patterns.
• Gradient-weighted Class Activation Mapping (Grad-CAM) was incorporated to provide visual explanations of the model's diagnostic decisions.

Main Results:

• The ViT-Grad-CAM model achieved high performance on validation and test sets, with accuracy, recall, and F1-scores consistently above 0.97.
• The model demonstrated superior performance compared to existing methods for TB detection in chest X-rays.
• Grad-CAM provided visual heatmaps highlighting significant regions, enhancing model transparency and aiding radiologist verification.

Conclusions:

• The novel ViT-Grad-CAM model offers a significant improvement in automated tuberculosis detection from chest X-rays.
• The model's high accuracy and interpretability show strong potential for clinical application in real-world settings.
• This approach enhances diagnostic precision and supports radiologists in timely and accurate TB diagnosis.