Our eyes could reveal more about our health than we realize. Google, in collaboration with its health-tech subsidiary Verily, has developed an AI algorithm to assess an individual’s risk of heart disease just by examining their retinas.

The rear interior wall of the eye, known as the fundus, contains blood vessels that reflect the body’s overall health. Unlike blood vessels throughout the rest of the human body that are hidden underneath skin and other tissues, this group of blood vessels are directly visible through the pupil.

By peering into the human eye, the algorithm could identify tell-tale signs of heart-health risks. For example, if the blood vessels were constricting from hypertension, or were clotted with cholesterol.

The deep-learning models were trained from data on 284,335 patients to predict which individuals were at high risk of suffering from a cardiovascular event, like a heart attack or stroke. The system was able to identify and analyze a patient’s age, blood pressure, whether they smoked and other data points from the retinal images to help assess risk level. It also generated a graphical representation of the data that showed which pixels in the retina images were the most important for predicting a specific risk factor. For example, the algorithm focused on how the blood vessels looked when making predictions about a patient’s blood pressure.

“Here, we show that deep learning can extract new knowledge from retinal fundus images,” wrote the study authors in Nature Biomedical Engineering. “We also show that the trained deep-learning models used anatomical features, such as the optic disc or blood vessels, to generate each prediction.”

The method was tested with two independent datasets of 12,026 and 999 patients, and had a 70 percent accuracy rate when determining which patients had a cardiovascular event in the past five years and who didn’t. In comparison, blood tests – which are the commonly used method of practice – have a 72 percent prediction rate.

The results validate the algorithm, and show that eye scans could be become part of regular health check-ups in the future. For now, the researchers say the technique needs to continue being tested on larger datasets.