Assessing the possibility of using portable and stationary non-mydriatic fundus cameras for diabetic retinopathy screening assisted by an artificial intelligence-based software platform in primary care

Abstract

Objectives: To assess the possibility of using portable and stationary non- mydriatic (N M) fundus cameras for diabetic retinopathy ( DR) screening as- sisted by the artificial intelligence ( AI)-based Retina- AI CheckEye© soft- ware platform in primary care. Methods: In this prospective, open-label study, 609 subjects (1218 eyes) with either diagnosed diabetes mellitus (D M) or risk factors for D M were divided into two groups depending on whether the fundus camera was stationary or portable. NM single-field fundus photography was per- formed with a stationary fundus camera in group 1 and a portable cam- era in group 2. The AI-b ased Retina- AI CheckEye© software platform was used for the analysis of digital color fundus photographs of subject eyes for signs of D R. The numbers of poor-quality fundus images and the pres- ence or absence of DR were noted, and the stage of DR was assessed. Results: In group 1 and group 2, there were 37 eyes and 339 eyes, respec- tively, whose images could not be processed by the neural network. D R was found in 15 subjects (5.17 %) in group 1 and 8 subjects (2.51 %) in group 2. Previously undiagnosed DM complicated by DR was discovered in 13 (4.5 %) of the subjects included in group 1 versus 7 (2 %) of the sub- jects included in group 2. Conclusions: Digital color fundus images taken with stationary and port- able NM fundus cameras through non-dilated pupils and analyzed by the AI-b ased Retina- AI CheckEye© software platform provided D R detection and grading by stages among subjects with diagnosed DM as well those with undiagnosed D M. The percentage of poor-quality photographs can be reduced and the effectiveness of DR screening with the use of the A I- based Retina- AI CheckEye© software platform can be improved through the involvement of an experienced operator and better adherence to pro- tocol for uploading fundus images to the cloud storage.