Clinical Validation of a Program for Diagnosing Vision-Threatening Diabetic Retinopathy Based on Automatic Segmentation Algorithms

Abstract

Background. Diabetic retinopathy is a very common, debilitating disease that requires early diagnosis and treatment. The development of new screening methods is a priority area of medicine in recent years. Purpose: Approbation of the software (SW) based on algorithms for automatic segmentation of signs of DR “Retina AI” in clinical practice, the study of the capabilities of the software “Retina AI” in the diagnosis of vision-threatening DR.Methods. Analysis of clinical data obtained from patients undergoing diagnostics and treatment at the Federal Research and Clinical Center for Specialized Types of Medical Care and Medical Technologies of the Federal Medical and Biological Agency of Russia and the Moscow Regional Scientific Research Clinical Institute. 123 photographs of the fundus were analyzed using the RETINA.AI platform. Fundus photography was carried out using a fundus camera VISUCAM 500 (Zeiss).Results. In the process of analyzing fundus photographs with Retina AI algorithms, a high efficiency of automatic detection of vision-threatening retinopathy was revealed. The accuracy of the method in diagnosing vision-threatening DR was 95 %, sensitivity — 96.59 %, specificity — 91.4 %, ROC AUC — 0.94.Conclusion. The DR diagnostic program based on adaptive segmentation algorithms is a tool for solving the DR screening problem.