Google AI tool can predict cardiovascular risk

Can a look inside your eye reveal the condition of your heart? An algorithm being developed by Google is really set on proving that your eyes are the window to your soul. A deep learning model based on the retinal images of nearly 285,000 people can predict risk factors such as age, gender, blood pressure and smoking status — some of the big predictors of heart diseases.

The technology could reveal the heart’s health condition after matching the eye scans with a matrix for cardiovascular risks. The algorithm has proved to be correct in 70 per cent of the cases where it has been tested so far. Google’s new AI technique could eliminate the need to have a long series of conventional tests and scans usually required to detect cardiovascular risks.

So, how did the research to develop a device to detect diabetic retinopathy help in determining the risk for cardiovascular diseases? Google already had access to the retinal images from an American and a UK dataset that it used to train its deep learning model. It studied the anatomical features in the eyes such as the optic disc or blood vessel to generate the predictions about age, gender, smoking status and blood pressure. This model was validated using another 13,000 records from these datasets.

Last year, the University of Leeds had done a similar study on the deep learning process of AI, where it analysed retinal scans and cardiac scans from more than 5,000 people. The AI system identified associations between pathology in the retina and changes in the patient’s heart. Once the image patterns were learnt, the AI system could estimate the size and pumping efficiency of the left ventricle of the heart from retinal scans alone. An enlarged ventricle is linked with an increased risk of heart disease. Combined with basic demographic data about the patients, their age and sex, the AI system could predict their risk of a heart attack over the subsequent 12 months.

How did the idea come about? Google was working with its partner hospitals in India to develop an AI tool that could screen for diabetic retinopathy without trained personnel. Diabetic retinopathy is a complication where uncontrolled blood sugar over several years damages the blood vessels of the retina, sometimes resulting in blindness. “The problem is that people do not perceive any changes to their vision until it is very late. Laser treatment can halt the progress of the disease but it cannot restore vision,” said Dr R Kim from Aravind Eye Hospital that was one of the partners of Google. The simple solution was to screen those coming to the diabetic clinics. “We have been diagnosing diabetes retinopathy remotely, with diabetes clinics sending us the photographs online. What the Google AI did was to ensure that an eye doctor did not have to go through each and every photograph to screen for the condition,” said Dr Kim. This AI-enabled device has already received the CE certification for use in Europe and has been used by Aravind Hospital as well for picking up cases in remote areas. The hospital alone has used it to screen 200,000 people so far.

Dr Rajiv Raman from Sankara Netralaya, which also collaborated for the initial study in 2016, is now studying whether these images can predict the occurrence of diabetic retinopathy. The challenge is to see whether it can work with real-world data. “The company has demonstrated that the retinal fundus images that we take to check for diabetes retinopathy may be used to check for the risk of cardiovascular diseases. Now, we need to show for sure that it happens,” he said.

“We have to collect data from thousands of patients. Along with retinal images, we have to collect data on parameters like their blood pressure, blood glucose levels and cholesterol levels,” said Dr Kim. The patients would then need to be followed for two years to see whether the algorithm can accurately predict the risk of cardiovascular diseases. The Indian Express