Results of screening for diabetic retinopathy with artificial intelligence-based RetCAD software

Abstract

Background: Diabetic retinopathy is one of top leading causes of blindness and low vision Worldwide. It is considered avoidable blindness, and early detection is the most important management strategy in these conditions. In our country, artificial intelligence has not been used in ophthalmology, especially DR screening. Aim: To assess the accuracy of an artificial intelligence-based RetCAD software screening in diabetic retinopathy in Mongolian diabetic retinopathy patients. Materials and Methods: The research was conducted in a cross-sectional model among the patients who attended the outpatient clinic of the Mongolian-Japan hospital of MNUMS. A total of 400 participants, 18-92 age were randomly selected for the study, with a 1:3 ratio of diabetic retinopathy and relatively healthy controls. All participants underwent complete eye examinations. Fundus color images were obtained for all participants using a RetCAD fundus camera. A comparison of the conclusions made by artificial intelligence with the conclusions of the ophthalmologist’s examination was studied. Results: The average age of the participants was 51.45±9.40 years, and the gender ratio was not statistically different between the two groups. Our studies found that 59.5% of people with DR were diagnosed with mild NPDR, 23% with moderate NPDR, 9.5% with severe NPDR, and 8.0% with PDR. The agreement between the ophthalmologist’s diagnosis and the AI program was 91.4% for mild, 95.6% for moderate, 94.7% for severe, and 93.7% for PDR. The Kappa Measure of Agreement was for no DR 0.73 and DR 0.71. RetCAD artificial intelligence software had a sensitivity of 92.4%, a specificity of 94.7%, and 0.93 AUC for the evaluation of fundus images. Adjudication was provided by a second expert examiner in case of disagreement between the AI software and the reference examiner. Conclusion: RetCAD offers better sensitivity and specificity compared to ophthalmologist examination and a lower cost and shorter time, less misdiagnosis for early detection of DR.