Google scientists use eye scans to predict heart disease risk

Eyes may open a window to a person’s vulnerability to heart problems.

Eye scans using deep learning technique are all that one may soon need to gauge whether he or she is prone to heart ailments in near future, according to a study by scientists at Google Research, Verily Life Sciences and Stanford School of Medicine.

The authors of the study, which appeared in the journal Nature Biomedical Engineering on Monday, said that they would be able to identify new indicators of heart disease risk present in pictures of retinas by analysing them with artificial intelligence.

“This is an early but promising start. If future research pans out, we do hope that the simpler technique of retinal fundus photography could also give additional information about cardiovascular risk non-invasively. But a lot of scientific work remains,” said Lily Peng of Google Research, the corresponding author of the study.

Traditionally, medical discoveries are often made through a sophisticated form of guess and test— making hypotheses from observations and then designing and running experiments to test the hypotheses. However, with medical images, observing and quantifying associations can be difficult because of the wide variety of features, patterns, colours, values and shapes that are present in real images, she said.

Advantages of deep learning

Deep learning, on the other hand, can generate very accurate models without us having to tell the system what to look for.

“Our approach uses deep learning to directly predict the outcome, and then have the system tell us what it was using to make the prediction through heatmaps. This allows us to draw connections between changes in the human anatomy and disease, akin to how doctors learn to associate signs and symptoms with the diagnosis of a new disease. This could help scientists generate more targeted hypotheses and drive a wide range of future research,” Peng told BusinessLine .

Earlier, the team had developed algorithm based on deep learning to diagnose blindness associated with diabetes.

The current screening of cardiovascular risk relies on a variety of variables derived from a patient’s history and blood samples, such as age, gender, smoking status, blood pressure, body mass index, glucose and cholesterol levels.

“Most cardiovascular risk calculators use some combination of these parameters to identify patients at risk of experiencing either a major cardiovascular event or cardiac-related mortality within a pre-specified time period, say, 10 years,” the scientists said.

More often than not, some of these parameters were not readily available. Citing information available from an electronic-health-record-based cardiovascular registry, maintained by the American College of Cardiology, the scientists said, the data required to calculate the 10-year- risk scores were available for less than 30 of the patients. “This is largely due to missing cholesterol values, which is not surprising given that a fasting blood draw is required to obtain these data,” they said.

Brings down prediction timeframe

Another significant aspect associated with the proposed screening is that it brings down the prediction time-frame to 5 years as against 10 years typically associated with the clinical risk predictors currently in use.

For the study, the scientists developed deep learning models using retinal fundus images of nearly 3,00,000 people available from two countries – the UK and the US – and validated them using those from another 13,000 patients. The authors however, cautioned that while the system achieved good results – that matched the accuracy of a standard blood test-based technique for estimating the risk of heart failure commonly used in Europe called SCORE -- a dataset of less than 300,000 patients is still small for AI, and should be tested further.

COMMENT NOW