Diagnostic Accuracy of Chest X-ray Computer-Aided Detection Software for Detection of Prevalent and Incident Tuberculosis in Household Contacts

This prospective cohort study in South Africa followed 483 household contacts of rifampicin-resistant TB for 4.6 years, evaluating the diagnostic and prognostic accuracy of three CAD software—qXRv3.0.0, CAD4TBv7.0, and Lunit INSIGHT v3.1.4.111. Among participants, 23 had prevalent TB (7 detected routinely), and 38 developed incident TB. qXR outperformed the other CAD solutions, achieving an AUC of 0.91 (95% CI: 0.83–0.99) for detecting prevalent TB, compared to 0.88 for Lunit INSIGHT and 0.87 for CAD4TB. Notably, over 30% of individuals with high qXR scores but negative routine sputum were later diagnosed with TB through enhanced sputum testing or follow-up, underscoring qXR’s superior ability to flag true TB cases missed by routine diagnostics. While the AUC for predicting incident TB ranged between 0.60–0.65 across all software, the study reinforces qXR’s effectiveness as the leading CAD tool for TB screening and triage, highlighting its potential to enhance early detection and intervention in high-burden settings.