Diabetic retinopathy is one of the devastating and most feared results of diabetes. According to the CDC, almost one-third of adults over age 40 with diabetes have diabetic retinopathy. If not detected and treated early, the patient’s vision can suffer permanent damage. We’ve written previously about AI in retinopathy screening, including IDx-DR, Eyenuk’s EyeArt, and Sweetch.
Researchers at the University of Michigan Kellog Eye Center recently reported their success pairing smartphone images of the retina with artificial intelligence for early detection of diabetic retinopathy. The Michigan researchers used a smartphone-mounted device to take high-quality retinal images and then determine in real time with AI whether a patient needed a referral to an ophthalmologist for further investigation.
The project involved two stages. One phase of the project was designing the RetinaScope, a portable, smartphone-based handheld platform that is inexpensive and easy to use. The second phase used deep neural network software — a type of AI — to enhance and review images with automated lesion grading to determine the need for follow-up. For the AI solution, the Michigan team used Eyenuk’s EyeArt, mentioned above.
In test comparisons with two trained human image graders, the EyeArt software working with RetinaScope images from 53 subjects showed an 86.8% sensitivity to the disease (higher than the 80% required for screening) and greater specificity than the human image graders in confirming the absence of diabetic retinopathy.
The next steps for the Michigan team include hardware and software improvements, developing a version that does not require dilated pupils, and obtaining FDA clearance. If they are successful in creating a low-cost and accessible way for individuals to screen for diabetic retinopathy, the quality of life for millions of people, while saving enormous amounts in healthcare costs.
