Computer Vision for Sickle Cell Retinopathy Screening Under Variable Fundus Camera Chromatic Aberration: A Domain-Invariant Representation Evaluation of Artificial Intelligence Referral Equity

Sickle cell retinopathy screening programmes that deploy fundus cameras across multiple manufacturers face a representational problem that is rarely quantified: chromatic aberration varies systematically between devices, and a convolutional network trained on images from one camera family may encode device-specific colour profiles as if they were retinopathy features. the present analysis treats that risk as a design property of the screening pipeline rather than a nuisance to be corrected post hoc. We formalise chromatic aberration as a bounded, device-indexed perturbation operator on the retinal image manifold and define a domain-invariant representation as one whose referral decision is Lipschitz-stable under that operator. Using a before-after-control-impact design with a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64), we characterise the operating characteristics of the evaluation across 2000 replications per design cell at cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points.54 clusters are required to detect a 5-percentage-point referral-equity effect at 80% power under this variance structure, and that the minimum detectable effect falls steeply as cluster count rises from 8 to 32. We argue that referral equity in artificial-intelligence retinopathy screening is not a property of a model alone but of the joint distribution of camera chromatic profiles and the representation's invariance class, and that evaluation designs which ignore this coupling will systematically overstate equity gains.