Clinical Computer Vision for Sickle Cell Retinopathy Screening Under Variable Fundus Camera Calibration: A Test-Time Adaptation Evaluation of Lesion Detection Sensitivity Across Nigerian Eye Clinics

Fundus cameras deployed across Nigerian eye clinics differ in sensor response, illumination geometry and colour calibration, so a lesion detector trained on one device may lose sensitivity when transferred to another. the present analysis treats that calibration shift as a domain-generalisation problem and evaluates test-time adaptation as a mitigation strategy for sickle cell retinopathy screening. Because no multi-site image corpus was available for direct measurement, the study characterises the operating characteristics of a before-after-control-impact evaluation design under a declared variance structure, using a replication-based power analysis with between-cluster SD = 6.0, residual SD = 4.5 and intraclass correlation = 0.64 across 2000 replications per design cell. The design grid comprised 8, 16, 24 and 32 clusters and effect sizes of 3, 5 and 8 percentage points. The analysis indicates that 50 clusters are required to detect a 5-percentage-point sensitivity gain at 80% power, and that the minimum detectable effect falls steeply as cluster count rises from 8 to 32. Calibration heterogeneity is not a nuisance to be averaged away but a structured source of variance that dictates the feasible size of any adaptation trial, and it specifies the design conditions under which test-time adaptation can be credibly evaluated in Nigerian screening settings.