Cardiovascular risk scores are widely used for risk stratification but may fail to identify a substantial proportion of individuals with subclinical atherosclerosis who are at increased risk of future cardiovascular events. Vascular ultrasound can directly detect carotid and femoral atherosclerotic plaques but its implementation is limited by the need for trained operators and expert interpretation. The AI-THEROSCOPE study aims to develop and validate an artificial intelligence-based tool capable of detecting subclinical atherosclerosis through the analysis of non-mydriatic retinal fundus images. Participants undergo clinical assessment, laboratory testing, carotid and femoral ultrasound, and retinal fundus photography. The performance of the AI model will be evaluated against vascular ultrasound findings as the reference standard for the presence of subclinical atherosclerosis.

Cardiovascular disease remains the leading cause of mortality worldwide. Current cardiovascular risk prediction models are useful for population-level risk estimation but may underestimate risk in a substantial proportion of individuals who already have subclinical atherosclerosis. Vascular ultrasound of the carotid and femoral arteries allows direct visualization of atherosclerotic plaques and improves cardiovascular risk stratification, but its widespread use is limited by the requirement for specialized equipment and trained personnel.

Retinal fundus imaging provides a non-invasive assessment of the microvasculature and has emerged as a promising tool for cardiovascular risk evaluation. Recent advances in artificial intelligence and deep learning have demonstrated the ability of retinal image analysis to identify cardiovascular risk factors and predict cardiovascular outcomes.

The AI-THEROSCOPE study is a prospective observational study designed to develop and validate an artificial intelligence model for the detection of subclinical atherosclerosis using non-mydriatic retinal fundus photographs. Adult participants without previous cardiovascular disease undergo standardized clinical evaluation, laboratory testing, carotid and femoral vascular ultrasound, and bilateral retinal fundus photography.

The presence of carotid and/or femoral atherosclerotic plaque assessed by vascular ultrasound serves as the reference standard. Deep learning techniques will be used to train and validate predictive models based on retinal images. Model performance will be evaluated using discrimination metrics including the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value, and negative predictive value.

The ultimate objective of the project is to develop a scalable, non-invasive, and easily deployable tool that may facilitate early detection of subclinical atherosclerosis and improve cardiovascular risk stratification in clinical practice and population screening programs.

Inclusion Criteria:

• Adults aged 18 years or older.
• No previous established cardiovascular disease.
• Undergoing cardiovascular risk assessment and carotid and femoral vascular ultrasound.
• Ability to provide written informed consent.

Exclusion Criteria:

• Previous acute coronary syndrome, stroke, or peripheral arterial disease.
• Previous carotid or femoral vascular surgery or stenting.
• Previous ophthalmologic surgery.
• Retinal or ocular diseases that significantly affect retinal vasculature or image quality, including moderate or severe diabetic retinopathy, retinal vascular occlusion, advanced hypertensive retinopathy, exudative age-related macular degeneration, or macular edema.
• Inability or unwillingness to provide informed consent.

Adults undergoing cardiovascular risk assessment at a university hospital cardiovascular risk unit and healthcare workers or relatives participating in cardiovascular screening programs. All participants undergo clinical evaluation, laboratory testing, carotid and femoral vascular ultrasound, and non-mydriatic retinal fundus photography.

Individual participant data (IPD) will not be publicly shared. De-identified data may be made available upon reasonable request to the principal investigator, subject to institutional policies, ethical approval, data protection regulations, and applicable legal requirements.