Google is Training AI to Spot Diabetic Blindness Quicker

Google Logo 2016 AH 12

Google, as a search engine as well as a company, has come a tremendously long way since the days of a simple web page and a list of links. Nowadays, there’s little that we don’t think Google is possible of, and with their recent approach to Artificial Intelligence, the company’s historic motto of “don’t be evil” could be helping the whole human race in the near future. Google’s AI research has focused on training these algorithms to make decisions for themselves, and speed up the process exponentially. The firm has already been training their AI to improve the quality of smaller images when enlarged, but the Google Research Blog is sharing something much more important with us this week, the idea of using AI to catch Diabetic Blindness sooner, rather than later.

Diabetes is a serious condition, and while many will know about the dangers surrounding the need to amputate a foot because of the disease, few will be familiar with Diabetic retinopathy (DR), a disease which puts as many as 415 Million diabetic patients at risk of blindness all over the world. Diabetic retinopathy is diagnosed by looking at a photo of behind the eye, with the below example from Google clearly showing the difference between a healthy eye on the left, and a diseased eye on the right. With the help of Doctors from India and the United States, the research team created a set of 128,000 images which were evaluated by 3 – 7 ophthalmologists from a panel of 54, and this dataset was used to train the AI during the experiment. Using a reference set of almost 10,000 of these images, the Google AI was compared to results from a group of ophthalmologists with the AI coming out close to, or on par with the verdicts from the ophthalmologists.

While it might not seem great news to shout about the AI being as good as the professionals, the end goal here is not necessarily be better overall, but to be quicker. In order to make sure critical decisions, an ophthalmologist would have to have decades of experience under their belt, but an AI could get to work almost immediately, and churn through images at a far, far quicker rate and deliver a diagnosis al most instantly. The future of this program will no doubt to run an even larger test, and to do so with FDA approval, and if the results from that are promising, then Google could very well help Doctors and hospitals all over the world help save people’s sight.