Clinical Computer Vision for Sickle Cell Leg Ulcer Staging Under Variable Smartphone Lighting: A Domain-Adaptation Evaluation of Wound-Classification Sensitivity Loss

Smartphone-based wound photography could extend leg-ulcer staging for sickle cell disease beyond specialist clinics, but the sensitivity of convolutional classifiers degrades when illumination shifts between capture sessions. the present analysis formalises that degradation as a domain-adaptation problem and characterises the operating characteristics of a before-after-control-impact design for evaluating candidate adaptation strategies. The analysis treats the between-session illumination shift as a domain shift with a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64) and estimates power 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 achieves 80% power at a 5-percentage-point sensitivity gain with 52 clusters. Minimum detectable effect curves and coefficient plots show that the interaction term is estimable at clinically meaningful magnitudes only when cluster counts exceed 24. The contribution is a design-theoretic account of how illumination-driven sensitivity loss can be detected and attributed, together with the boundary conditions under which smartphone-based staging evaluations are informative.