Artificial Intelligence Benchmarking of Low-Dose Computed Tomography Denoising for Paediatric Tuberculosis Screening Under Unmeasured Scanner Contrast-Resolution Coupling: A Healthcare Diagnostic-Robustness Evaluation

Low-dose computed tomography denoising algorithms for paediatric tuberculosis screening are benchmarked almost exclusively on image-similarity metrics that assume scanner contrast and resolution properties are known and stable, yet in district-level deployment these properties are neither measured nor reported. the present analysis characterises the operating characteristics of a before-after-control-impact benchmarking design for denoising robustness when scanner contrast-resolution coupling is unmeasured. 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 standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64, estimated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The design required 54 clusters to detect a 5-percentage-point effect at 80% power. Estimated power rose with cluster count and effect size, and the minimum detectable effect at 80% power fell as clusters increased. The intraclass correlation of 0.64 dominated the variance budget, meaning that scanner-level coupling between contrast and resolution behaves as a cluster-level nuisance that no within-scanner denoising metric can absorb. The contribution is a formal statement of the conditions under which denoising benchmarks for paediatric tuberculosis screening can support diagnostic-robustness claims, and a demonstration that those conditions are not met by current similarity-based protocols.