Decoding the Shadows: A Deep Learning and Explainable AI Approach for Enhanced Tuberculosis Detection from Chest X-Rays

Tuberculosis (TB), a persistent global health challenge, disproportionately affects vulnerable populations and necessitates rapid and accurate diagnostic tools. Chest X-rays (CXRs) remain a cornerstone of TB screening and diagnosis, but their interpretation is subjective and prone to inter-observer variability. Deep learning (DL) has emerged as a powerful tool for automated image analysis, demonstrating remarkable potential in identifying subtle pathological patterns indicative of TB. However, the “black box” nature of many DL models hinders clinical adoption due to a lack of trust and transparency. This paper presents a novel approach for TB detection from CXRs that integrates cutting-edge deep learning architectures with explainable AI (XAI) techniques. We argue that by providing interpretable insights into the model’s decision-making process, we can significantly enhance clinician confidence and facilitate more informed diagnostic pathways. Our methodology involves training a Convolutional Neural Network (CNN) on a diverse dataset of CXRs and then employing XAI methods such as Grad-CAM and SHAP to highlight the specific regions of the X-ray that contribute to the TB classification. The paper details our data collection strategy, patient demographics, the chosen DL architecture, and the implementation of XAI. We present a comprehensive evaluation of the model’s performance in terms of accuracy, sensitivity, and specificity, alongside qualitative assessments of the generated explanations. Our findings demonstrate that the proposed DL-XAI framework not only achieves high diagnostic accuracy but also offers valuable visual justifications, paving the way for a more robust and trustworthy automated TB detection system in clinical practice.