Enhancing Tuberculosis Diagnosis with DenseNet121 and Grad-CAM: A Deep Learning Approach for Accurate and Interpretable Chest X-ray Analysis

In this paper, GradCAM has been used with a DenseNet121 architecture for classifying Tuberculosis. The model was trained and evaluated over the dataset of 420 chest X-ray images for the classification of TB-positive versus TB-negative cases. Performance metrics denote an overall accuracy of 90.48%. This is with a precision of 90% and a recall of 100% for the TB negative class, and a precision of 100<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> with a recall of 43 % for the TB positive class. The corresponding f1-scores for these classes are 0.95 and 0.60, respectively. The macro average f1-score stands at 0.77, thereby showing that on the whole, there is a fair amount of balance in performance across both classes. The weighted average f1-score Metropolitan: 0.89, indicating good overall performance but disparity in recall between classes. GradCAM visualizations enable a look at the areas of interest that the model focuses on while making predictions, providing insight into the decision-making process of DenseNet121. The study epitomizes both the potential and limitation of DenseNet121 on the classification of TB, pointing out a future road map for improvements in sensitivity toward TB-positive cases.