We will investigate whether retinal fundus photographs can predict carotid intima-media thickness (cIMT), an ultrasound-derived marker of subclinical atherosclerosis and cardiovascular risk. Using UK Biobank data, we will link left/right fundus images with carotid ultrasound-derived cIMT measurements and relevant demographic/clinical factors (e.g., age, sex, BMI, blood pressure, diabetes, lipids, smoking). We will develop and validate machine learning models (deep learning for images plus statistical/ML models for combined data) to predict continuous cIMT and explore risk group classification. We will assess performance using appropriate train/validation/test splits and report results in aggregate only. Model interpretability will be explored using saliency/attribution methods to identify retinal features associated with cIMT. The goal is to support non-invasive, scalable approaches for early cardiovascular risk assessment and prevention.
