As any ophthalmologist will tell you, an examination of the eyes can reveal signs of any number of diseases, including diabetes and high blood pressure. Google has taken that reality and combined it with deep learning algorithms to take the diagnostic potential to a new level. With its system, Google’s deep learning tech is able to predict cardiovascular risk in any given individual simply using images of their retina.
The new system was recently detailed by Google Research, as well as in a newly published study titled “Prediction of Cardiovascular Risk Factors from Retinal Fundus Photographs via Deep Learning.” As part of this, researchers trained deep learning algorithms using data on 284,335 patients.
Using that data, the deep learning system was trained to identify certain health issues and risk factors, such as very high blood pressure or whether the patient is a smoker or non-smoker. The latter distinction was performed with a 71-percent accuracy, while the system took blood pressure identification up a notch by predicting the patient’s systolic pressure within an average of 11mmHg.
That exceeds the abilities of human doctors, who are usually able to tell the difference between someone with normal or high blood pressure, but not able to estimate the systolic pressure. The system can also use retinal images to predict risk factors, which Google says includes things like the patient’s gender and age.
In addition to identifying risk factors, Google’s deep learning algorithms learned to predict the likelihood of a cardiovascular event, such as a heart attack or stroke, with “fairly” high accuracy. One test showed a 70-percent accuracy, for example, in determining which patient among two had experienced a major cardiovascular event following when the retinal image was taken.
Google indicates that its researchers will train the algorithms using larger datasets in the future.
SOURCE: Google Blog