Googles AI detects diabetic retinopathy at similar rate to doctors

Google researchers have developed an artificial intelligence (AI) that can detect diabetic retinopathy in patients with diabetes.
The AI detects diabetic retinopathy, the leading cause of blindness among adults, by examining retinal photos.
A study published in the Journal of the American Medical Association reports that the AI detected retinopathy at roughly the same rate as human ophthalmologists.
Google now hopes that this technology could be used to screen more people than doctors are able to, particularly in countries where screening is limited.
Deep learning
Google’s AI uses deep learning to detect diabetic retinopathy, which involves the identification of faces, animals, and objects in pictures uploaded to Google’s online services.
The concept of using the AI to identify retinopathy began when a Google researcher noticed Indian doctors were struggling with the amount of screening required.
The Google Brain team, a team inside Google that provides AI software, requested over four dozen doctors in India and the US to identify conditions in photos, such as diabetic retinopathy, and then fed around 128,000 of these images into Google’s neural network.
The system consistently identified diabetic retinopathy more than the doctors, avoiding both false positives and false negatives more than 90 per cent of the time – the National Institute of Health recommend a standard of at least 80 per cent accuracy in diabetic retinopathy screenings.
AI won’t replace doctors
Lily Peng, a physician and biomedical engineer at Google, insists the idea behind this AI is not to replace doctors, rather to help them.
Her team are currently conducting additional trials to train Google’s diagnostic AI, with preliminary results showing the system continues to perform as well as trained doctors.
The findings have been met positive feedback by scientists. David McColloch, a clinical professor of medicine at the University of Washington who specialises in diabetes, said: “This is a well validated technology that can bring screening services to remote locations where diabetic retinal eye screening is less available.”