The eye as a window on the heart: what an AI found in the retina

Why look for the heart in the eye?

Diseases of the heart and blood vessels, such as heart attacks and strokes, remain a leading cause of death. To find out who is most at risk, doctors use well-known factors: age, sex, blood pressure, smoking and cholesterol. That usually means a blood test and a doctor's appointment.

The back of the eye, the retina, photographed through the pupil, is the only place in the body where you can see small blood vessels directly without any cut. The photo is quick and painless, and many people with diabetes already have it to check their eyes.

The researchers asked a simple question: does a photo of the eye hold information about the heart that human eyes cannot see?

How was the AI trained?

The team used deep learning, a method where a program learns to recognize patterns by itself after seeing a huge number of examples. Here it looked at retinal photos from 284,335 patients, along with known details for each one, such as age, sex, blood pressure and smoking.

To make sure the AI had not simply memorized its examples, the researchers tested it on two other groups it had never seen, of 12,026 and 999 patients.

They also checked which parts of the image the AI relied on. It used familiar structures of the eye, such as the optic disc (where the nerve of the eye leaves for the brain) and the blood vessels.

What could it read from the retina?

From a single photo, the AI estimated patients' age with an average error of 3.26 years. It recognized sex with a score of 0.97, on a scale where 1 is a perfect result and 0.5 is no better than chance.

It also estimated systolic blood pressure (the top number) with an average error of 11.23 mmHg, and smoking status with a score of 0.71. Before this work, these were not thought to be visible in a retina at all.

Finally, to flag the people who later had a serious heart event, such as a heart attack or stroke, its score was 0.70. That is better than chance, but far from perfect.

What does this mean for patients?

For now, nothing changes in usual care: no eye doctor works out your heart risk with this AI. The study mainly shows that a photo of the eye holds more information about general health than anyone expected.

In the longer term, a retinal photo taken to check the eyes could also act as an early signal to suggest a heart check. That is especially interesting for people with diabetes, who often have this test.

It also opens a new way of doing research: let the AI find links in images, then check with classic studies whether those links are real.

What are the limits of this study?

The score of 0.70 for serious heart events is modest: the AI is often wrong about who will have a problem. It does not replace a blood pressure check, a blood test or a doctor's opinion.

The study is retrospective, meaning it analyzed data that had already been collected. No trial has shown that using it improves patients' health. It would also need checking in other populations and with other cameras.