In under a minute, an artificial intelligence program can take a picture of the back of a person’s eye and—by analyzing the strength of the blood vessels that feed the retina—find clues that may point to higher risks of a stroke or heart attack.
Using scans from more than 95,000 people collected through U.K. and European biobanks and studies, researchers at St George's Hospital Medical School, part of the University of London, were able to analyze the varying widths and the amount of twists and turns of both eyes’ tiny veins and arteries.
By combining that data with a person’s age, smoking history and previous medical conditions, a study found the AI model’s predictions could perform just as well or better than standard Framingham risk scores for cardiovascular-related deaths several years into the future, without the need for a blood draw or measuring a person’s blood pressure.
At the same time, the AI’s findings performed similarly to those established risk scores when it came to individual predictions for myocardial infarction and stroke, though their additions did not improve the overall performance of the Framingham model. The results were published in the British Journal of Ophthalmology.
The researchers described their AI-enabled risk prediction program, dubbed Quartz, as a potentially low-cost and non-invasive method for screening larger populations of people for heart disease, as retinal imaging is already established in the U.S. and the U.K.
The eyes have long served as a window into health, being the only place in the human body where a clinician can directly observe the circulatory system at work from the outside.
The first FDA clearance for an AI-powered diagnostic went to a program for spotting the signs of diabetic retinopathy in the eye’s blood vessels. That automated test was incorporated earlier this year into Baxter’s Welch Allyn retinal cameras.
Meanwhile, researchers at the University of Leeds this past January published study results showing an eye scan could help estimate the size and strength of the heart’s left ventricle, the workhorse that pumps oxygenated blood to the body.
That AI predicted a patient’s chances of having a heart attack in the next 12 months with an accuracy between 70% and 80%, as an enlarged ventricle can be a major sign of impending cardiac disease. Those researchers said their system could help refer patients during routine eye exams for more in-depth screenings with ultrasound or MRI.