Effectiveness of a computer-assisted algorithm for onsite screening of diabetic retinopathy from retinal photographs at diabetic outpatient clinics

Abstract

Purpose: To examine the effectiveness of a computer-assisted algorithm for onsite screening for diabetic retinopathy (DR) at diabetic outpatient clinics. Methods: 1263 patients were examined over two years. Undilated fundus photographs were acquired at the clinic. Photographs were independently assessed by an ophthalmologist and optometrist in a darkened room in a masked fashion and also processed through the algorithm. DR was defined per the International Clinical Diabetic Retinopathy Disease Severity Scale and severity of diabetic retinopathy.

Results: 2526 eyes of 1263 patients were assessed. The algorithm successfully graded 2153 (85%) images with 63.04% sensitivity and 79.63% specificity compared to an ophthalmologist; in comparison to an optometrist, sensitivity and specificity were 60.87% and 79.05%, respectively. The agreement between ophthalmologist and optometrist was kappa=0.835 for presence of DR, 0.835 for severity of DR. Conclusion: This algorithm may be a utilized in a diabetic clinic for a quick screening with only the retinal photographs.