The diagnostic performance of an autonomous artificial intelligence (AI) system to diagnose referable diabetic retinopathy was compared to manual grading performed by ophthalmologists in a recent study.
Patients with type 1 and type 2 diabetes took part in a diabetic retinopathy screening program between 2011 and 2012 in Valencia, Spain. Per standard protocol, two images were collected for each eye. All patients received mydriatic drops. Two deidentified retinal images were obtained per patient: one disc and one fovea centered. The AI system and manual adjudicated grading outputs were compared for sensitivity and specificity for diagnosis of referable diabetic retinopathy and vision-threatening diabetic retinopathy.
Final analysis included 2,680 patients. Per manual grading, prevalence of referable diabetic retinopathy was 4.14% (n=111/2,680 and of vision-threatening diabetic retinopathy was 2.57% (n=69/2,680). The AI system, compared with manual grading, had a sensitivity rate of 100% (95% confidence interval [CI], 97-100) and a specificity rate of 81.82% (95% CI, 80-83) for referable diabetic retinopathy; for vision-threatening diabetic retinopathy, sensitivity and specificity rates were 100% (95% CI, 95-100) and 94.64% (95% CI, 94-95), respectively.
“Compared with manual grading by ophthalmologists, the autonomous diagnostic AI system had high sensitivity and specificity for diagnosing referable diabetic retinopathy and macular edema in people with diabetes in a screening program,” the researchers concluded. “Because of its immediate, point of care diagnosis, autonomous diagnostic AI has the potential to increase the accessibility of referable diabetic retinopathy screening in primary care settings.”