This article is more than 1 year old

AI eye-scanner can tell whether you'll croak it from a heart attack

If and when this hits the mainstream, who's going to trust their retinas to random models?

AI algorithms can predict whether a patient is at risk of suffering a stroke, heart attack, or dying from heart disease just by studying images of their retinas, according to research out of England.

The thin layers of tissue at the back of the eyeball sensitive to light can reveal a surprising amount of information on their own. Previous studies have found that retinal images can accurately predict a person's age, sex, if they are smoker, or if they are diabetic or not. The width and network of blood vessels in the retina are affected by blood pressure, a factor important in measuring cardiovascular health risks, too.

A team of academics led by St George's, University of London have conducted what they believe is the largest AI retinal study yet. Retinal scans from over 70,000 people across two biomedical datasets, the UK Biobank and the European Prospective Investigation into Cancer in Norfolk, were used to train and test QUARTZ – QUantitative Analysis of Retinal vessels Topology and siZe – an AI-based system using supervised learning. "Sensitivity analyses restricted model development and validation to white ethnicity," the authors noted. 

By studying images of eyeballs and the medical histories of each participant, QUARTZ learns how common visual features in retina scans correlate to coronary heart disease, heart attacks, myocardial infarction, and strokes. The system can thus predict, from a given retina photo, whether someone is at risk of dying from heart disease over the next five years.

QUARTZ's performance was compared to the Framingham Risk Score (FRS), an algorithm used to predict the chances of a person developing cardiovascular disease within the next ten years. QUARTZ's predictions were generally as accurate or better than FRS, according to the study's results, which were published in the British Journal of Ophthalmology.

The researchers believe AI retinal scans could be used in clinical settings to assess a patient's cardiovascular health without the need to take blood pressure measurements or blood tests one day. 

"Retinal imaging is established within clinic and hospital eye care and in optometric practices in the US and UK," they wrote in the paper. "AI-enabled vasculometry risk prediction is fully automated, low cost, non- invasive and has the potential for reaching a higher proportion of the population in the community because of 'high street' availability and because blood sampling or sphygmomanometry are not needed."

Don't expect optometrists to start offering heart check-ups any time soon, however. There are numerous technical and regulatory hurdles in deploying AI-based retinal scans in clinical settings.

More ophthalmologists would have to be hired and trained in interpreting the results, and what happens next? Who would patients be referred to receive therapeutic treatment, a pair of researchers from the University of Dundee in Scotland asked. 

"A large, randomized clinical trial—is surely required before the CVD risk prevention guidelines can be changed to incorporate retinal measurements as part of our routine risk prediction assessment," they wrote in an editorial critiquing the research. The team agreed and said more "experimental evidence would be needed to formally assess the effectiveness on [cardiovascular disease] prevention before advocating implementation."

Ify Mordi, a clinical senior lecturer, at the University of Dundee, told The Register the trial needs to show that AI systems are better than current methods used to screen for diseases. "The measure of 'better' in this type of randomized trial would usually be something related to what you hope to improve [such as] reduction in deaths, heart attacks and strokes," he said.

"It could be that if a trial showed retinal scanning was at least as good then theoretically why would you do [blood pressure] readings and blood tests, particularly, for example, in diabetic patients who will be getting retinal photos anyway? [It] could help reduce burden on already stretched healthcare systems." ®

More about

TIP US OFF

Send us news


Other stories you might like