Evaluation of Artificial Intelligence-Based Grading of Diabetic Retinopathy in Primary Care

• Published in: JAMA network open 2018-09, Vol.1 (5), p.e182665-e182665
• Main Authors: Kanagasingam, Yogesan, Xiao, Di, Vignarajan, Janardhan, Preetham, Amita, Tay-Kearney, Mei-Ling, Mehrotra, Ateev
• Format: Article
• Language: English
• Subjects: Adolescent; Adult; Aged; Aged, 80 and over; Artificial intelligence; Artificial Intelligence - standards; Artificial Intelligence - trends; Clinical medicine; Diabetes; Diabetes and Endocrinology; Diabetic retinopathy; Diabetic Retinopathy - classification; Diabetic Retinopathy - diagnosis; Female; Humans; Male; Middle Aged; Online Only; Original Investigation; Primary care; Primary Health Care - methods; Primary Health Care - standards; Primary Health Care - statistics & numerical data; Research Design - standards; Research Design - statistics & numerical data; Sensitivity and Specificity; Western Australia
• ISSN: 2574-3805; 2574-3805
• DOI: 10.1001/jamanetworkopen.2018.2665

There has been wide interest in using artificial intelligence (AI)-based grading of retinal images to identify diabetic retinopathy, but such a system has never been deployed and evaluated in clinical practice. To describe the performance of an AI system for diabetic retinopathy deployed in a primary care practice. Diagnostic study of patients with diabetes seen at a primary care practice with 4 physicians in Western Australia between December 1, 2016, and May 31, 2017. A total of 193 patients consented for the study and had retinal photographs taken of their eyes. Three hundred eighty-six images were evaluated by both the AI-based system and an ophthalmologist. Sensitivity and specificity of the AI system compared with the gold standard of ophthalmologist evaluation. Of the 193 patients (93 [48%] female; mean [SD] age, 55 [17] years [range, 18-87 years]), the AI system judged 17 as having diabetic retinopathy of sufficient severity to require referral. The system correctly identified 2 patients with true disease and misclassified 15 as having disease (false-positives). The resulting specificity was 92% (95% CI, 87%-96%), and the positive predictive value was 12% (95% CI, 8%-18%). Many false-positives were driven by inadequate image quality (eg, dirty lens) and sheen reflections. The results demonstrate both the potential and the challenges of using AI systems to identify diabetic retinopathy in clinical practice. Key challenges include the low incidence rate of disease and the related high false-positive rate as well as poor image quality. Further evaluations of AI systems in primary care are needed.