AI-based retinal analysis demonstrated good agreement with actual brachial-ankle pulse wave velocity (ICC = 0.762, P<0.001) and moderate agreement with ankle-brachial index (ICC = 0.410, P<0.001).
Observational
No
Does AI-based retinal analysis accurately predict brachial-ankle pulse wave velocity and ankle-brachial index in patients undergoing cardiovascular risk assessment?
AI-derived retinal features show significant associations with baPWV and ABI, providing a non-invasive and scalable telemedicine strategy for arterial stiffness and cardiovascular risk assessment.
Effect estimate: ICC 0.762 for baPWV; ICC 0.410 for ABI
p-value: p=<0.001
Background Pulse wave velocity (PWV), known as the gold standard for evaluating arterial stiffness, is limited by device dependence. This study aims to explore the feasibility of retinal parameters to assess PWV based on artificial intelligence (AI), providing a new approach for remote and convenient cardiovascular risk assessment.Methods This retrospective study leveraged both cross-sectional and longitudinal data, enrolling patients from Peking University Third Hospital who underwent fundus photography and PWV testing simultaneously over five consecutive years. An optimised AI system automatically quantified retinal vascular parameters, optic morphological features, and lesion counts. Correlation analyses and multivariate linear regression models were employed using the training set to assess associations between retinal features and brachial-ankle PWV (baPWV) or ankle-brachial index (ABI), with validation performed in the test set.Results In the training set of 3,088 visits, baPWV and ABI showed significant correlations with retinal vascular morphology (including mean vessel density, arteriolar-to-venular ratio, regional curvature, fractal dimension), optic disk/cup parameters (disk area, circularity), and lesion counts (exudative and drusen) (all p < 0.001). Elevated baPWV was independently associated with older age, higher blood pressure, lower arteriolar-to-venular ratio, larger disk area, and increased soft exudates and drusen (F = 503.864, p < 0.001). Higher ABI was linked to older age, higher diastolic pressure, lower arteriolar-to-venular ratio, decreased vessel density, and fewer hard exudates (F = 71.368, p < 0.001). Validation demonstrated good agreement between predicted and actual baPWV (ICC = 0.762, p < 0.001), while agreement for ABI was moderate (ICC = 0.410, p < 0.001). Longitudinal analysis further revealed that baPWV changes correlated with fractal dimension, vessel density, arteriolar/venular ratio, and disk circularity, whereas ABI changes were associated with fractal dimension, vessel diameter, vessel density, arteriolar diameter, and drusen count (all p < 0.001).Conclusion This large-scale study demonstrates significant associations between AI-derived retinal features and baPWV/ABI, identifying robust predictive markers. The proposed AI-based retinal analysis offers a non-invasive, scalable, and remote screening paradigm for arterial stiffness and cardiovascular risk assessment.
Wang et al. (Fri,) conducted a observational in Arterial stiffness and cardiovascular risk. AI-based retinal analysis was evaluated on Agreement between AI-predicted and actual brachial-ankle PWV (baPWV) and ankle-brachial index (ABI) (ICC 0.762 for baPWV; ICC 0.410 for ABI, p=<0.001). AI-based retinal analysis demonstrated good agreement with actual brachial-ankle pulse wave velocity (ICC = 0.762, P<0.001) and moderate agreement with ankle-brachial index (ICC = 0.410, P<0.001).