Artificial Intelligence Optimal Transport Domain Adaptation for Diabetic Retinopathy Grading Under Variable Fundus Camera Colour Calibration: A Fairness Evaluation of Healthcare Diagnostic Equity

Colour calibration differences between fundus cameras shift the pixel distributions on which diabetic retinopathy graders are trained, and a model that performs well on one camera may silently misgrade images from another. The framework treats each camera as a source or target measure on a colour-appearance space, defines a calibrated transport plan under a colour-constancy constraint, and derives a bound on the target-domain grading risk that separates covariate shift from label shift. A before-after-control-impact design is characterised under a declared variance structure with between-cluster SD 6.0, residual SD 4.5 and intraclass correlation 0.64; power was estimated across 2000 design cells for cluster counts of 8, 16, 24 and 32 and effects of 3, 5 and 8 percentage points. The design analysis indicates that 54 clusters are required for 80% power at a 5-percentage-point effect. The framework proposes that fairness-aware transport regularisation, not unconstrained distribution matching, is the condition under which diagnostic equity across cameras can be assessed rather than assumed.