One of the fastest growing causes of blindness around the world is an eye disease that’s almost entirely preventable.
Diabetic retinopathy is a condition that occurs among diabetics when high blood sugar levels damage the retinal blood vessels, leading to complete vision impairment over time. The disease is a threat to those who have lived with diabetes for years, but it can be detected early and treated if patients are regularly screened.
Unfortunately, in India, which is home to over 69 million diabetics (as of 2015), regular eye examinations aren’t easy to come by, particularly outside big cities. While access to even basic healthcare is difficult, the problem is compounded by a serious shortage of trained ophthalmologists. So, some 45% of patients suffer from vision loss before they’re even diagnosed with diabetic retinopathy.
And that’s where artificial intelligence comes in.
Specialist doctors are trained to diagnose the disease by analysing retinal photographs and looking for different types of lesions, such as microaneurysms or haemorrhages, that can indicate its severity. Last year, Google announced that it had taught an image-recognition algorithm how to detect signs of diabetic retinopathy using a dataset of 128,000 retinal photographs. In subsequent tests with other images, the algorithm managed to perform on par with a panel of ophthalmologists.
Since then, the project team has been working to validate the results with two hospital chains in India, Aravind Eye Care and Sankara Eye Hospital, and has recently completed initial clinical trials. Aravind even found the algorithm performing slightly better than its average ophthalmologist, Lily Peng, product manager at Google Research and a former nanoscientist and bioengineer, said during a talk earlier this year.
Now, Google is in the early stages of figuring out pilot deployments for the technology in India.
“Machine learning has the capability of helping extend the reach of healthcare providers and bringing high quality care to everyone, especially rural and under-served communities where there is a shortage of experts,” Peng told Quartz in an email.
That’s particularly important in India, where the condition is a major cause of preventable blindness. In 2014, a study by the All India Ophthalmological Society found that diabetic retinopathy was detected in nearly 22% of its sample of over 6,200 diabetic patients across the country. More importantly, signs of the condition were detected even in patients who hadn’t yet experienced any vision impairment, suggesting that early screening is the one thing that could make a big difference in avoiding blindness.
As advanced as the diagnostic algorithm is, though, we’re still a long way from technology replacing doctors, even as AI and virtual reality are being increasingly incorporated into India’s healthcare sector. Peng notes that the successful adoption of the Google algorithm depends on healthcare providers who will need to adapt to handle an increase in patients as more people are diagnosed with diabetic retinopathy.
“Machine learning’s true potential will only be realised when deployed in partnership with healthcare providers,” Peng said.
But as diabetes spreads, notably among low-income communities who can hardly afford medical care, the first order of business is still the matter of getting ordinary Indians through the door.