An artificial intelligence (AI)-based grading system successfully diagnosed two patients with diabetic retinopathy, according to a study published by JAMA.
A research team—led by Yogesan Kanagasingam, PhD; Di Xiao, PhD; and Janardhan Vignarajan—sought to evaluate the performance of an AI-based grading system for diabetic retinopathy in a real-world clinical setting.
Diabetic retinopathy is a complication of diabetes that can cause damage to the retinas and is the most common cause of vision loss among people with diabetes. Though there’s been interest in using AI-based grading to help to identify the disease, researchers said no system has ever been used or evaluated in clinical practices.
For their study, the team took retinal photographs of 193 patients with diabetes at a primary care practice in western Australia between December 2016 and May 2017. They collected 386 images and had them evaluated by both the AI-based system and an ophthalmologist.
After evaluating the images, the ophthalmologist found that 183 patients had no signs of the disease, eight had mild, nonproliferative diabetic retinopathy and two patients had clinically significant diabetic retinopathy (one moderate and one severe) that required referral to a physician.
The AI system was able to accurately identify the two patients with moderate and severe diabetic retinopathy. The system classified 17 patients overall as having clinically significant diabetic retinopathy but 15 of those cases were false positives.
“Though there was a limited sample size, the AI system was effective in ruling out disease,” the study said. “However, the system had a high rate of false-positives with a specificity of 92 percent and positive predictive value of just 12 percent.”
Based on the results, the authors said they believe an AI-based grading system has potential for improving the efficiency in screening for diabetic retinopathy in primary care. They also pushed for more research.
“The ability to provide real-time eye screening at familiar primary care physician practices has many practical advantages, including comprehensive chronic disease management at a single location for patients with diabetes. There is also the potential for the AI system to be improved,” the authors said. “Further training of the AI system to differentiate drusen, sheen reflections, and exudates can improve the specificity.”