Artificial Intelligence Tuberculosis Screening on Chest Radiographs Under Variable Portable X-Ray Generator Ageing: A Test-Time Calibration Evaluation of Healthcare Notification Yield

Portable chest radiography is the imaging backbone of tuberculosis screening in settings where fixed digital units are unavailable, but the generators that produce those images age unevenly, and the resulting drift in beam quality and detector response is rarely represented in the data used to train and validate artificial intelligence readers. the present analysis treats generator ageing as a distribution-shift problem and asks what happens to notification yield when a deployed screening model is recalibrated at test time rather than retrained. The analysis is a design study, not an empirical evaluation: it characterises the operating characteristics of a before-after-control-impact estimator under a declared variance structure with an intraclass correlation of 0.64, a between-cluster standard deviation of 6.0 and a residual standard deviation of 4.5, across cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points, with 2000 replications per design cell. Under those assumptions the design attains 80 per cent power at 52 clusters for a 5 percentage point effect. The contribution is a calibration-aware framing of notification yield that separates the accuracy loss caused by generator ageing from the yield loss caused by threshold placement, and a demonstration that test-time recalibration is estimable within a cluster-randomised screening programme without additional imaging. The paper concludes that generator age should be recorded as a routine covariate in screening registries, because without it the calibration signal cannot be separated from the case-mix signal.