Clinical Positioning and Implementation of a Deep-Learning Retinal Biomarker (Reti-CVD) for Cardiovascular Risk Stratification: A Narrative Review.

Cardiovascular disease (CVD) prevention depends on accurate risk stratification before symptoms develop. Standard tools such as the Pooled Cohort Equations, QRISK3, and SCORE2 require laboratory data and are less informative in borderline-risk individuals, creating a role for accessible adjuncts. Retinal imaging directly visualizes the systemic microvasculature, and deep-learning oculomics may provide complementary risk information. Reti-CVD generates a three-tier classification from a retinal photograph and is among the more extensively validated retinal-AI tools. This narrative review evaluates its clinical positioning and implementation as an exemplar rather than a product endorsement, organizing evidence by cohort, comparing the approach with established scores and subclinical atherosclerosis markers, and considering implementation, regulation, and equity. RetiCAC was trained using coronary artery calcium as a surrogate label; subsequent Reti-CVD studies included UK Biobank, Singapore SEED, and CMERC-HI. Reported discrimination was approximately 0.75 by the Harrell C-index, with modest reclassification improvement, particularly in borderline-risk groups. As the commercial product DrNoon for CVD, the tool holds marketing authorization from Korea's Ministry of Food and Drug Safety (MFDS) and, according to the manufacturer, CE certification under the EU Medical Device Regulation (MDR); in Korea it entered outpatient practice through a time-limited non-covered (out-of-pocket) assessment-deferral pathway, and it has not yet received US FDA authorization. Most evidence originates from one research group and one commercial algorithm, and no randomized or outcome-based study has shown that Reti-CVD-guided care improves clinical outcomes. These observational findings remain hypothesis-generating rather than evidence of established clinical utility. Reti-CVD is therefore best regarded as a non-invasive risk enhancer for borderline/intermediate-risk reclassification, not as a tool of established clinical utility; independent validation, intervention trials, and cost-effectiveness and reimbursement evidence are needed before broad integration.