AI Based Medical Diagnosis for Early Detection of Tuberculosis Using Chest X Ray Images

Abstract

Tuberculosis (TB) has recently reclaimed its status as the world’s leading infectious killer, surpassing COVID-19 mortality rates. A primary obstacle to eradication is the "missing millions" millions of undiagnosed cases that drive community transmission. This research explores the transformative role of Artificial Intelligence (AI) and Deep Learning (DL) in automating the interpretation of chest X-ray (CXR) images to bridge this diagnostic gap. We analyze the evolution of neural network architectures, from Convolutional Neural Networks (CNNs) like ResNet and EfficientNet to Vision Transformers (ViTs) and hybrid models, which have achieved diagnostic accuracies exceeding 99%. The paper further investigates the critical role of precision lung segmentation using U-Net variants and the emergence of multimodal data fusion combining CXR with acoustic cough analysis and clinical data to enhance diagnostic robustness. Clinical evidence, such as the Yichang study, demonstrates that AI can increase diagnostic yields by over 230% compared to manual reviews, particularly in resource-constrained primary care settings. Finally, we address deployment challenges, including "black-box" skepticism mitigated by Explainable AI (XAI), domain shift problems, and the potential of Edge AI for real-time inference in remote areas