Accuracy of Artificial Intelligence in Detecting Tuberculosis from Chest X-ray: A Systematic Review and Meta-Analysis of Diagnostic Performance

Tuberculosis (TB) remains a major global public health challenge, particularly in low-resource countries where access to trained radiologists is limited, making Chest X-ray (CXR) screening difficult to scale. The advancement of Artificial Intelligence (AI) and Computer-Aided Detection (CAD) technology offers a potential solution by providing automated TB detection and supporting diagnostic workflows. To assess their clinical readiness, this systematic review and meta-analysis was conducted using the PRISMA 2020 protocol and included studies from PubMed, Scopus, and Semantic Scholar that evaluated AI-CAD systems (Index Test) against microbiological or extended reference standards (Reference Standard). The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2 and QUADAS-C) tools were applied to measure risk of bias, and a random-effects model was used to estimate pooled Diagnostic Odds Ratio (DOR). Six studies with approximately 38,940 participants were eligible for analysis. Results showed a pooled DOR of 0.133 (95% CI: 0.047–0.377), indicating a significantly lower diagnostic error rate (P=0.000). Although sensitivity was consistently high (83.3%–100%), specificity varied widely (26.8%–98.9%), resulting in notable heterogeneity and a wide prediction interval (0.003–6.411). These findings conclude that AI-CAD tools demonstrate strong potential for TB screening but should undergo local validation, threshold calibration, and operational evaluation before broad clinical implementation, especially where specificity remains below the WHO Target Product Profile.