Artificial intelligence in fundus photography for type 2 diabetes: a scoping review of systemic biomarkers and multi-organ risk prediction

Type 2 diabetes mellitus (T2DM) is associated with multi-organ complications, including cardiovascular and renal disease. Fundus photography provides a non-invasive window into systemic microvascular health, and artificial intelligence (AI) has enabled extraction of retinal biomarkers for systemic risk prediction beyond diabetic retinopathy detection. We conducted a methodologically structured scoping review following PRISMA-ScR guidance to map AI applications in retinal imaging for multi-organ risk stratification in T2DM. Studies using machine learning or deep learning models to predict cardiovascular, renal, or cerebrovascular outcomes were identified and characterized. Rather than quantitative pooling, we examined modeling strategies, validation approaches, performance reporting, and translational readiness across heterogeneous study designs. AI models frequently demonstrated promising discrimination; however, substantial heterogeneity was observed in cohort size, outcome definitions, imaging modalities, and validation strategies. External validation was limited, calibration was inconsistently assessed, and subgroup analyses addressing fairness and device-related domain shift were rarely reported. Most studies emphasized discrimination metrics without comprehensive evaluation of clinical utility.Retinal AI shows potential for scalable systemic risk surveillance in T2DM, but rigorous external validation, standardized reporting, and prospective implementation studies are required to enable safe and equitable clinical translation.