Methods and applications of artificial intelligence in tuberculosis diagnostics: advances, opportunities, and challenges

Abstract

Tuberculosis (TB) remains a leading global health challenge, with more than ten million cases annually and persistent diagnostic gaps that delay treatment and drive transmission. Despite advances in molecular assays and imaging technologies, conventional diagnostic tools, such as sputum microscopy, culture, and nucleic acid amplification tests, continue to face limitations related to suboptimal sensitivity, slow turnaround times, infrastructure requirements, and dependence on highly trained personnel. These constraints disproportionately affect high-burden, resource-limited settings, where delays in case detection significantly hinder TB control. Artificial intelligence (AI) techniques represented by machine learning (ML) and deep learning (DL) algorithms are reshaping the TB diagnostic landscape by enabling rapid, automated, and scalable analysis of complex clinical, radiological, and genomic datasets. AI-enhanced chest radiography achieves near-expert performance, with models reporting AUC values approaching 0.95–0.99. In parallel, AI-driven genomic prediction systems leverage whole-genome sequencing (WGS) data to identify drug-resistance patterns with F1-scores exceeding 90%, offering a faster and more comprehensive alternative to traditional drug-susceptibility testing (DST). Integrative multimodal frameworks combining imaging, genomics, and clinical parameters further support earlier and more precise diagnosis, personalized therapy selection, and real-time risk stratification. Beyond high diagnostic performance, a successful translation of AI into routine TB care requires careful attention to ethical, regulatory, and implementation challenges. Data heterogeneity, algorithmic bias, limited model interpretability, and insufficient regulatory harmonization remain major barriers to equitable deployment. Sustainable adoption will depend on transparent validation, representative datasets, federated learning strategies, and collaboration among clinicians, microbiologists, data scientists, engineers, and policymakers. With thoughtful governance and context-sensitive implementation, AI has the potential to transform TB diagnosis into a more accurate, efficient, unbiased and patient-centered continuum, advancing global efforts toward TB elimination by 2030.