Deep learning AI IDs diabetic retinopathy, eye diseases using retinal images

Researchers have developed a deep learning system (DLS) using artificial intelligence (AI) capable of identifying diabetic retinopathy and related eye diseases using retinal images, according to a study published in JAMA. The system's performance was comparable to human graders.

The DLS system was trained to identify diabetic retinopathy using 76,370 images, possible glaucoma with 125,189 images and age-related macular degeneration (AMD) with 72,610 images. Performance of the system was measured from identification of diabetic retinopathy using 112,648 images, possible glaucoma with 71,896 images and AMD with 35,948 images.

To evaluate the system, researchers tested the sensitivity and specificity of DLS using 494,661 retinal images. Results showed the DLS system achieved 90.5 percent sensitivity and specificity of 91.6 percent for detecting referable diabetic retinopathy; 100 percent sensitivity and 91.1 percent specificity for vision-threatening diabetic retinopathy; 96.4 percent sensitivity and 87.2 percent specificity for possible glaucoma; and 93.2 percent sensitivity and 88.7 percent specificity for age-related macular degeneration when compared to a professional grader.

“In this evaluation of retinal images from multiethnic cohorts of patients with diabetes, the DLS had high sensitivity and specificity for identifying diabetic retinopathy and related eye diseases,” concluded first author Daniel Shu Wei Ting, MD, PhD and colleagues. “Further research is necessary to evaluate the applicability of the DLS in health care settings and the utility of the DLS to improve vision outcomes.”