Artificial Intelligence Counterfactual Calibration of Low-Dose Chest Radiograph Tuberculosis Screening Under Variable Digital Detector Gain Drift: A Domain-Adaptation Evaluation of Healthcare Diagnostic Safety

Digital detector gain drift in low-dose chest radiography displaces the pixel-intensity distribution on which artificial-intelligence tuberculosis classifiers depend, and the resulting calibration error is a patient-safety problem rather than a mere nuisance. the present analysis treats counterfactual calibration as a design problem for domain adaptation and characterises, by formal analysis and a replication-based evaluation of the before-after-control-impact estimator, how much detector heterogeneity a screening programme can absorb before the safety margin collapses. The analysis is theoretical: no observational data are used or claimed, and the operating characteristics of the design are characterised under a declared variance structure with between-cluster SD = 6.0, residual SD = 4.5 and intraclass correlation = 0.64, across 2000 design cells. Power was estimated for cluster counts of 8, 16, 24 and 32 and effects of 3, 5 and 8 percentage points; 54 clusters are required for 5 percentage points at 80 per cent power. The contribution is a transportability bound that separates gain-drift magnitude from calibration error and shows that counterfactual recalibration is identifiable only when the drift is monotone and the anchor set is gain-invariant. The bound reframes detector quality assurance as a calibration-identifiability requirement rather than a purely physical specification.