Google expects AI to predict heart disease by observing retinas

    Abdulaziz Sobh


    That's what Google researchers did when applying artificial intelligence to predict something deadly serious: the likelihood of a patient suffering a heart attack or stroke. The researchers took these determinations when examining images of the patient's retina.

    Google, which presents its findings Monday in Nature Biomedical Engineering, an online medical journal, says the method is as accurate as predicting cardiovascular disease through more invasive measures that involve inserting a needle into a patient's arm.
    At the same time, Google warns that more research is needed.

    According to the company, medical researchers have previously shown some correlation between retinal vessels and the risk of a major cardiovascular event. Using the image of the retina, Google says it was able to quantify this association and 70% of the time accurately predict which patient within five years would experience a heart attack or other major cardiovascular event, and which patient would not. Those results were in line with the test methods that require a blood draw to measure a patient's cholesterol.

    Google used data-based models of 284,335 patients and was validated in two independent datasets of 12,026 and 999 patients.

    "The warning of this is that it's early, (and) we've trained this in a small data set," says Lily Peng of Google, a doctor and lead researcher on the project. "We believe that the accuracy of this prediction will increase a bit more as we get more complete data, and find out that we can do this is a good first step, but we have to validate."

    Peng says Google was a little surprised by the results. His team had been working on the prediction of eye disease, then expanded the exercise by asking the model to predict from the image if the person was a smoker or what their blood pressure was. Beyond predicting the factors that put a person at risk for heart attack or stroke was a consequence of the original investigation.

    Google's technique generated a "heat map" or graphical representation of data that revealed which pixels of an image were most important in predicting a specific risk factor. For example, Google's algorithm paid more attention to blood vessels to make predictions about blood pressure.

    "Pattern recognition and the use of images is one of the best areas for AI at this time," says Harlan M. Krumholz, professor of medicine (cardiology) and director of Yale's Center for Research and Evaluation of Results, who considers that the investigation is proof of that. concept.

    It will help us to understand these processes and diagnoses in ways that we have not been able to do before, "he says." And this will come from photographs and sensors and a whole range of devices that will help us to essentially improve the physical examination and I think that with greater precision will sharpen our understanding of diseases and people and combine it with treatments. "

    In the event that more research is done over time, physicians, as part of routine health checks, could study such retinal images to help assess and manage risks to patients' health.

    How much time could I take?

    Peng says it's more in the "order of years" than something that will happen in the next few months. "It's not only when it's going to be used, but how it will be used," he says.

    But Peng is optimistic that artificial intelligence can be applied in other areas of scientific discovery, perhaps even in cancer research.

    Medical discoveries are usually made through what she says is a sophisticated way of "guessing and testing", which means developing hypotheses from observations and then designing and executing experiments to test them.

    But observing and quantifying associations with medical images can be a challenge, says Google, because of the wide variety of features, patterns, colors, values and shapes that are present in real images.

    "I'm very excited about what this means for the discovery," says Peng. "We hope that researchers from other places will take what we have and build on it."


    Commenting only available for logged in users