Google Uses Deep Learning To Help Predict Heart Disease Risk

Tech giant Google is not one to shy away from revolutionary new uses of artificial intelligence, and the newest out of the company's wheelhouse is an AI program that's able to predict a patient's risk of heart disease by looking for signs of a number of key risk factors in their eyes. This is a feat mostly unique to AI, to be sure; while some human doctors may have dealt with enough heart disease patients and potentials to be able to spot some telltale signs, it took Google's machine 284,335 sets of eyes' worth of training to attain its expertise, essentially making it better than any human at this kind of prediction.

Key risk factors for heart disease include both obvious things like your age and gender, and things that may be a bit harder to detect, such as whether you smoke, your systolic blood pressure, and whether you've had major cardiovascular events in the past. Eye factors like blood vessels and lines in the ocular disc can give away things like that to the very well-trained eye, and this AI was very well-trained to look for discrepancies of just those sorts. Google's AI was checked for accuracy against two independently collected data sets, consisting of eye imagery from 12,026 and 999 patients. Across the two sets, the AI was able to classify eye photos based on most metrics with an accuracy rate ranging between 70% and 97%.

This is an example of AI technology opening almost entirely new doors, and finding new ways of doing things. It shows the potential of AI to investigate, quantify, and put to concrete evidence what even the most knowledgeable human may chalk up to a hunch. While the link between the eyes and the heart may be somewhat well-known, no human could ever hope to study enough pairs of peepers to actually figure out what all of the different properties of the eye meant as far as heart disease risk. Through mass aggregation and deep learning, an AI program was able to accomplish this feat, and will only continue to improve as it is further trained, tested, and perhaps eventually used in the real world.

You May Like These
More Like This:
Android Headlines We Are Hiring Apply Now