Artificial intelligence for diagnosis and surveillance of tuberculosis in young adults: systematic review

Abstract

Introduction: Tuberculosis remains a public health priority due to its high global incidence. Diagnostic gaps persist in rural primary care, which justifies evaluating novel decision-support tools. Artificial intelligence, particularly computer-aided detection on chest radiographs, could optimize triage and shorten the time to microbiological confirmation.

Objectives: To assess computer-aided detection performance for tuberculosis in adults and appraise risk/incidence prediction models for use in rural settings such as Vinces.

Methods: Systematic review (2015–2025) of bibliographic databases and technical sources. Population: adults (≥15 years). Diagnostic accuracy studies required a microbiological reference standard. Dual screening, standardized data extraction, and methodological appraisal were performed. A qualitative synthesis summarized eligible studies.

Results: Computer-aided detection offers expert-comparable performance for triage, though it varies by threshold, prevalence, and software version. Before adoption, the threshold should be calibrated with local data. Evidence specific to adults aged 20–40 years is limited. Prediction models are feasible, but their transfer to rural settings requires high-quality data, external validation, and impact evaluation.

Conclusions: When embedded in diagnostic pathways with microbiological confirmation, computer-aided detection can shorten time to confirmatory testing and standardize chest radiograph interpretation in primary care, provided prior calibration and ongoing monitoring are in place. Prediction models offer potential value for planning, contingent on data robustness and validation.