Artificial intelligence-based quantification of retinal microvascular biomarkers from fundus photography of chronic kidney disease: a case-control study.

Background: Chronic kidney disease (CKD) is frequently asymptomatic in its early stages and remains substantially underdiagnosed, largely due to the lack of accessible and non-invasive risk identification tools. Retinal microvascular alterations may reflect systemic microvascular changes associated with CKD, offering a potential non-invasive window for early disease-related microvascular assessment. Objective: To identify retinal microvascular parameters associated with CKD and evaluate their discriminatory ability using AI-based fundus image analysis. Methods: In this single-center case-control study, fundus photographs from healthy controls and patients with CKD were analyzed using an AI-based platform to quantify retinal vascular features. To avoid inter-eye dependency, one eye per participant was included. A total of 322 participants were analyzed, including 110 controls, 142 with CKD stages 1-2, and 70 with CKD stages 3-5. Feature selection was performed using LASSO regression, followed by multivariable logistic regression adjusted for age, sex, and body mass index (BMI). Model performance was evaluated using receiver operating characteristic (ROC) analysis. Results: In multivariable analysis adjusted for age, sex, and BMI, lower arteriovenous ratio (AVR; OR = 0.617, Conclusion: A limited set of AI-derived retinal microvascular parameters was associated with CKD and demonstrated moderate discriminatory ability. Notably, several retinal parameters were already altered in early-stage CKD, with consistent directional changes observed across disease stages. These findings suggest that retinal microvascular alterations may be detectable in the early stages of CKD, highlighting their potential as non-invasive indicators of early disease-related microvascular changes. Further validation in larger and more diverse populations is warranted.