Abstract
Purpose :
Current manual diabetic retinopathy (DR) screening setups cannot scale to triage the ever-increasing population of diabetic patients at risk of vision loss. EyeArt meets this growing need with a fully-automated, cloud-based screening system. Diagnostic efficacy of EyeArt is demonstrated on a large dataset of both mydriatic and non-mydriatic color retinal fundus images
Methods :
EyeArt’s hybrid approach combines novel morphological image analysis with state-of-the-art deep learning techniques to create an automated DR screening system engineered for large scale deployment on the cloud. EyeArt automatically analyzes multiple fundus images of a patient to generate patient-level DR screening recommendation, while flagging and excluding external eye images and poor quality images.
850,908 color retinal fundus images of 107,001 diabetic patient visits or encounters from the EyePACS screening program were automatically analyzed by EyeArt. EyePACS human graders provided the ICDR severity level and Clinically Significant Macular Edema (CSME) surrogate markers present/absent for use as reference standard to evaluate EyeArt’s DR screening recommendations. A patient was deemed non-referable if there was mild or no signs of DR and no CSME surrogate markers in both eyes. 5291 encounters did not have grading due to image quality and were excluded from this analysis. 54,481 (53.6%) cases were non-mydriatic, 46,580 (45.8%) cases were mydriatic, and dilation status was missing for 649 cases.
Results :
In the 101,710 encounters, prevalence of encounters with moderate non-proliferative DR (NPDR) or higher or with surrogate markers for CSME was 19.3% and prevalence of encounters with potentially treatable DR (severe NPDR or PDR) was 5.1%. EyeArt’s screening sensitivity was 91.3% and specificity was 91.1% with area under the receiver operating characteristic curve (AUROC) of 0.965. EyeArt’s sensitivity for detecting potentially treatable DR was 98.5%. The performance split as per dilation status during imaging is listed in Table 1.
Conclusions :
EyeArt automated system has high screening sensitivity and specificity on both mydriatic and non-mydriatic retinal images as demonstrated on a large real world dataset making it safe and effective for DR screening.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.