With help from AI, eye images offer window into cardiovascular risk: Study

A new system that uses artificial intelligence (AI) to assess cardiovascular risk based on images of the eye captured during eye exams demonstrated strong correlation with a standard cardiovascular risk assessment, according to a study presented at the American College of Cardiology’s Annual Scientific Session (ACC.26). Researchers said using AI to screen for heart disease risk during routine eye exams could help more people become aware of their risk and facilitate referrals for preventative care.

Heart disease is the leading cause of death worldwide. Primary care providers and cardiologists use risk calculators to estimate a patient’s risk of atherosclerosis-a buildup of plaque in the arteries that can lead to heart attacks, stroke and premature death—and to guide recommendations for lifestyle modifications and medications, if needed. However, not all people see a primary care provider regularly and may not be aware of their risk or steps they can take to reduce it.

“The awareness that someone might be at risk is really one of the key missing pieces,” said Michael V. McConnell, MD, clinical professor of medicine at Stanford University in Stanford, California, and the study’s lead author. “The image of the back of your eye has a wealth of health information. We can analyze these images with AI to help people become aware of their risk and have the opportunity to get guideline-based evaluation and preventive therapy.”

McConnell serves as chief health officer at Toku, the company that created the AI system used in the study. The system, called CLAiR, received Breakthrough Device designation from the U.S. Food and Drug Administration (FDA). The results of this first prospective evaluation of CLAiR in the U.S. will support FDA submission.

In this study, 1 in 4 participants were found to have an elevated risk of heart disease based on standard cardiovascular risk assessments including blood pressure and cholesterol screening. The AI-based method that analyzed blood vessels at the back of the eye using retinal images taken during the visit largely matched this determination, identifying at-risk participants with a sensitivity of 91.1% and a specificity of 86.2%.

“Even just a standard retinal photo provides high resolution imaging of your blood vessels—it’s a literal window into vascular tissue,” McConnell said.

Previous studies have shown that eye images can be used to assess conditions such as diabetes, but most methods have relied on interpretation by human experts. With CLAiR, developers sought to demonstrate how this approach can be scaled up for clinical implementation by using AI to automate image analysis. The AI system was trained to recognize patterns in blood vessel appearance that are associated with the development of heart disease.

Researchers enrolled 874 participants between 40-75 years of age who were not taking lipid-lowering medications and did not have known atherosclerosis. Participants were recruited at 10 eye care and primary care sites across the United States; half were female, 19% were Black or African American and 26% were Hispanic.

Each participant underwent retinal imaging with standard cameras that are used at most eye clinics. The CLAiR system was then used to analyze the images and identify participants whose likelihood of experiencing heart disease or stroke in the next 10 years was 7.5% or higher, which is the threshold commonly used to identify patients who would likely benefit from taking statins.

The standard ASCVD risk estimator was also used to identify participants in the 7.5% 10-year risk category to compare the CLAiR result. For this, data were collected on participants’ age, sex, smoking status, blood pressure and cholesterol during the same clinic visit.

In total, 26% of participants were found to have a 10-year ASCVD risk score of 7.5% or greater using the standard risk estimator. The CLAiR system showed agreement with these results, correctly identifying positive cases 91.1% of the time (sensitivity) and correctly identifying negative cases 86.2% of the time (specificity).

Based on the results, researchers said the AI system shows promise as a noninvasive screening method that should be feasible to implement in most eye care settings, although more work is needed to facilitate referral of at-risk patients for cardiovascular evaluation and treatment in primary care following retinal image screening.

“This approach would not replace the standard cardiovascular risk evaluation, but it’s a potential way to bring greater awareness, especially for people who should be on preventive care, but who have not yet had a thorough evaluation,” McConnell said. “For patients to benefit, we need to implement clear pathways to connect your elevated risk from your eye exam to help you see your clinician and ultimately get guideline-based preventive therapy.”

Overall, 94% of the images acquired in the study were able to be used by the AI system, offering evidence that the approach works well across cameras used in different clinics. Retinal imaging takes about five minutes, and the CLAiR algorithm returns results in about 30 seconds, suggesting that implementing this approach would not add much time to clinical workflows. McConnell said that the CLAiR system is not designed for use in people who are pregnant or have advanced eye disease, which can affect the condition of blood vessels in the eye.

Although retinal imaging is available in most U.S. eye clinics, it is not covered by all vision insurance plans as part of a standard visit, and patients may be charged an additional fee for the imaging.

Reference:

With help from AI, eye images offer window into cardiovascular risk, American College of Cardiology, Meeting: American College of Cardiology's Annual Scientific Session