Artificial Intelligence Benchmarking of Automated Diabetic Foot Ulcer Severity Staging Under Unmeasured Smartphone Flash Falloff and Perfusion Confounding: A Healthcare Calibration-Transfer Evaluation

Automated diabetic foot ulcer (DFU) severity staging from smartphone photographs is attractive for triage in resource-limited settings, but its benchmark performance is routinely reported without accounting for two acquisition artefacts: radial illumination falloff from the camera flash and perfusion-dependent colour shifts in the wound bed. The framework formalises flash falloff as a radial attenuation field and perfusion confounding as a multiplicative shift in tissue chromophore concentration, then derives the bias in estimated staging accuracy when either is ignored. A design evaluation was conducted across a grid of 8, 16, 24 and 32 clusters and effect sizes of 3, 5 and 8 percentage points, with 2000 replications per cell and a declared variance structure of between-cluster SD = 6.0, residual SD = 4.5 and intraclass correlation = 0.64. Power reached 80% at 52 clusters for a 5-percentage-point effect. The analysis indicates that calibration-transfer benchmarks must report the acquisition and perfusion covariates they condition on, and that cluster counts below 52 yield inadequate power for the effect sizes typical of DFU staging studies.