AI uses eye scans to predict heart disease

By using deep-learning neural networks, the researchers were also able to predict other cardiovascular risk factors not previously thought to be present or quantifiable in retinal images, such as age, gender, smoking status, and blood pressure.While the algorithm did not outperform existing medical approaches such as blood tests, the use of eye scans is appealing as once further developed it could provide a tool that people could one day use to screen thselves for health risks related to heart disease.According to the paper, which was published recently in Nature Biomedical Engineering, the researchers were able to use the scans to train the machine because markers of cardiovascular disease, such as hypertensive retinopathy and cholesterol boli, can often manifest in the eye.{{quote-A:R-W:450-Q: The algorithms were not taught what to look for, instead the syst “learned” by reviewing the data and detecting patterns in people’s eyes, }}“Because blood vessels can be non-invasively visualised from retinal fundus images, various features in the retina, such as vessel calibre, bifurcation or tortuosity, microvascular changes and vascular fractal dimensions, may reflect the systic health of the cardiovascular syst as well as future risk,” the researchers wrote.“The clinical utility of such features still requires further study. In this work, we donstrate the extraction and quantification of multiple cardiovascular risk factors from retinal images using deep learning.”The pattern-recognising algorithms, known as neural networks, were trained on scanned retinal images of more than 280,000 people from the US and the UK.One aspect of the study was that the algorithms were not taught what to look for, instead the syst “learned” by reviewing the data and detecting patterns in people’s eyes by pinpointing the patterns often found in the eyes of people at risk..The chair of the American Optometric Association (AOA)’s New Technologies Committee, Dr Paul Barney, labelled the study impressive, but baulked at the researchers’ suggestion that the technology might eventually be made available in outpatient settings.“We need to recognise good technology that will be around and has the potential for providing better care to our patients, and how we can incorporate that into optometry,” he said.“It could create a gateway for people to see their doctor of optometry for a comprehensive eye examination, explain the risk factors for cardiovascular disease and orchestrate the care that they need. The earlier the symptoms are detected, the better.”However, despite the potential applications for the technology, Dr Maulik Majmudar, associate director of the Healthcare Transformation Lab at Massachusetts General Hospital, told The Washington Post that the results donstrated the difficulty of making significant improvents in cardiovascular risk prediction. He gave the example of age and gender as powerful predictors of risk that were already available to doctors without the need for any additional testing.Another vulnerability that has been pointed out by critics is the fact that it is often unclear how or why many neural networks, such as the one used to analyse the retinal scans, reach their conclusions.The research was conducted by Google and Verily Life Sciences, a subsidiary of Google’s parent Alphabet.