Interpretability Audits of Computer Vision Tuberculosis Screeners Under Digital X-Ray Detector Gain Drift: A Healthcare Calibration Robustness Evaluation

Detector gain drift in digital radiography silently rescales pixel intensities, and computer vision tuberculosis screeners that appear well calibrated on stable acquisitions can lose both discrimination and interpretability when gain drifts between sites or across service intervals. the present analysis treats interpretability as a measurable property of the screener-detector system rather than of the network weights alone, and it characterises how a before-after-control-impact audit design performs when the audit must detect a drift-induced degradation in calibration. No observational data are used or claimed. The analysis is a replication-based evaluation of the estimator under a declared variance structure with between-cluster SD = 6.0, residual SD = 4.5 and intraclass correlation = 0.64, with power estimated across 2000 design cells over cluster counts of 8, 16, 24 and 32 and effects of 3, 5 and 8 percentage points. The design attains 80% power at 54 clusters for a 5-percentage-point degradation. The results indicate that the audit is feasible at realistic cluster counts only for the larger effects, that the minimum detectable effect falls steeply as clusters are added, and that the intraclass correlation of 0.64 dominates the precision of the interaction estimate.