Título

ARTIFICIAL INTELLIGENCE FOR THE SCREENING OF DIABETIC RETINOPATHY WITH ONE RETINAL IMAGE PER EYE.

Objetivo

Diabetic retinopathy (DR) is a major cause of blindness. Our objective was to assess the performance of an artificial intelligence (AI) system embedded in a handheld smartphone-based retinal camera for DR screening based on the evaluation of only one retinal image per eye.

Método

Images were collected from individuals with diabetes during a mass screening initiative of DR in Blumenau, Southern Brazil, by trained operators. Automatic analysis was performed with an AI system (EyerMaps ™, Phelcom Technologies LLC, Boston, USA) with one macula-centered, 45o field of view retinal image per eye. The output was compared to a retinal specialist reading, considered the ground truth. Patients with ungradable images were excluded from the analysis.

Resultado

Images from 686 individuals (average age 59.2 + 13.3 y. o., 56.7% women, diabetes duration 12.1 + 9.4 years) were analyzed. Rates of insulin use, daily glycemic monitoring and treatment for systemic hypertension were 68.4%, 70.2% and 70.2%, respectively. More than half of patients underwent their first retinal examination during the event, the majority (82.5%) relying exclusively on the public health system. Individuals who were illiterate or who had not completed elementary school were 43.4%. DR classification according to the ground truth was as follows: absent or nonproliferative mild DR 86.9 %, more than mild (mtm) DR 13.1 %. AI percentage rates of sensitivity, specificity, positive predictive value and negative predictive values for mtmDR were (95% IC) 93.6 (87.8-97.2), 71.7 (67.8-75,4), 42.7 (39.3-46.2) and 98.0 (96.2-98.9), respectively. Area under the ROC was 86.4%.

Conclusão

A high sensitivity was obtained for DR screening with a portable retinal camera and AI with only one image per eye, a simpler protocol as compared to the traditional, two images per eye protocol. Simplifying the process of DR screening could contribute to increased adherence rates and increased overall coverage of such programs.