Artificial intelligence for pulmonary tuberculosis detection on chest X-ray images: A systematic review and diagnostic accuracy meta-analysis
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Artificial intelligence (AI)-based computer-aided detection (CAD) of tuberculosis (TB) on chest X-ray (CXR) may standardize screening and triage in settings with few radiologists but its accuracy in clinical cohorts remains uncertain. We searched five databases (PubMed, ScienceDirect, IEEE Xplore, OpenAlex, Lens.org) on 15 March 2026 for studies published between January 2020 and March 2026, and we appraised these studies using QUADAS-2. Evaluations from clinical cohorts with complete 2 × 2 data and a microbiological or composite reference standard for all participants were pooled using a bivariate random-effects model, with 95% prediction intervals. We accounted for overlapping participants by performing a post hoc analysis with one evaluation per cohort and used GRADE to assess the certainty of the evidence. No protocol was prepared or registered. We included 21 studies (26 evaluations of AI models). Nine evaluations from three independent cohorts were pooled. Pooled sensitivity was 0.896 (95% confidence interval [CI] 0.874 to 0.915, prediction interval 0.798 to 0.950) and pooled specificity was 0.689 (95% CI 0.642 to 0.732; prediction interval 0.499 to 0.831). Both confidence intervals include the World Health Organization (WHO) triage thresholds (sensitivity 90%, specificity 70%). When a screening study with partial verification was added, pooled specificity increased to 0.735. Eighteen of 24 assessed evaluations were at high risk of bias and only 1 of 21 studies was prospective, fully verified, externally validated and met the WHO thresholds. Certainty of evidence was low for sensitivity and very low for specificity. A subsequent PubMed search found additional eligible clinical evaluations that were not screened, so the review is incomplete for those studies. In clinical cohorts, AI-CXR sensitivity is close to the WHO triage target, and specificity is close to, but not clearly above, its target. The evidence is pooled across only three cohorts, and specificity varies by product and setting. Thresholds need to be calibrated locally, and prospective, externally validated evaluations are needed before widespread deployment.