A teenage girl has created a 3D-printed lens that, when fitted onto a smartphone and used with an app, can give a preliminary diagnosis for people with diabetic retinopathy, reports IEEE Spectrum. Diabetic retinopathy is the most common cause of vision loss for those who suffer from diabetes, and diagnosis of the disease is usually done via a two-hour exam, which also requires a specialized camera to take a photo of the retina. Sixteen-year-old Kavya Kopparapu, along with her team, trained an AI system to identify diabetic retinopathy in photos instead, sharply cutting down the time it takes for a diagnosis.
Kopparapu used her computer science skills to create the diagnostic system after her grandfather, who lives in a small city in East India, experienced symptoms of diabetic retinopathy. “The lack of diagnosis is the biggest challenge,” Kopparapu told IEEE. “In India, there are programs that send doctors into villages and slums, but there are a lot of patients and only so many ophthalmologists.”
In developing the app and lens, Kopparapu spoke to everyone she could, including ophthalmologists, neuroscientists, and experts in machine learning. Her team, which includes her brother and a high school classmate, then chose to use a convolutional neural network — a type of machine learning that has been successfully applied to analyzing images — for the diagnostic AI. The team decided to use a network model developed by Microsoft researchers, and used 34,000 retinal scans from the National Institutes of Health’s database to train the system to identify eye diseases.
Kopparapu said the scans were often too dark or out of focus, but were “very representative of the real-world conditions you’d get with using a smartphone.” AI systems learn by analyzing and spotting patterns in sets of data — in this case, the retinal scans.
The system can spot diabetic retinopathy with the same accuracy of a human doctor. In October, Kopparapu partnered with Aditya Jyot Eye Hospital in Mumbai, India, to test her app Eyeagnosis. By that time, the app could not only detect diabetic retinopathy in images, but also blood vessels and other details that would usually require a fluorescent dye injection. Kopparapu has since shipped the first prototype of her smartphone lens to the hospital, which works by focusing the smartphone’s flash to illuminate the retina. A photo is then snapped before AI searches for any abnormalities. IEEE reports, “Eyeagnosis has already been tried on five patients at the hospital, and in each case it made an accurate diagnosis.” Experts say the app and lens system has proven to be reliable in a number of different situations, is extremely affordable, and has commercial potential. While there are no clear expansion plans, Kopparapu recently presented the Eyeagnosis system at the O’Reilly Artificial Intelligence conference in New York in June.