Clinical Computer Vision for Buruli Ulcer Lesion Boundary Estimation Under Variable Smartphone Camera Focus: A Test-Time Adaptation Evaluation of Surgical Referral Concordance

Smartphone-based clinical photography is increasingly used to document Buruli ulcer lesions in endemic districts where surgical referral decisions must be made without specialist review. Lesion boundary estimation from such images is sensitive to camera focus variation, yet the operating characteristics of test-time adaptation under realistic focus degradation have not been characterised against surgical referral thresholds. Power was estimated across 2000 design cells spanning 8, 16, 24 and 32 clusters and effect sizes of 3, 5 and 8 percentage points. The analysis found that 52 clusters were required to detect a 5 percentage point improvement in boundary concordance at 80% power. The intraclass correlation of 0.64 substantially inflated the cluster requirement relative to naive individual-level calculations. The contribution is a design-level characterisation of how focus-induced boundary error propagates into surgical referral concordance, and a specification of the cluster counts at which test-time adaptation effects become detectable. The results imply that single-site evaluations of adaptation protocols are underpowered for the effect sizes that matter clinically, and that referral concordance should be treated as a cluster-level rather than image-level endpoint.