Identifying Causes of Ungradable Fundus Photos in an Artificial Intelligence Assisted Screening Program for Diabetic Retinopathy

Abstract

Purpose : The purpose of this study is to identify the causes of ungradable fundus scans in an artificial intelligence (AI) assisted diabetic retinopathy screening program. By examining the prevalence and characteristics of ocular pathologies in patients with ungradable scans, we aim to identify potential contributing factors to the quality of fundus imaging and to inform the development of strategies for improving the accuracy and effectiveness of diabetic retinopathy screening.

Methods : A retrospective chart review of patients screened for diabetic retinopathy with EyeNuk's (Los Angeles, CA) EyeArt software at eight Philadelphia locations between 7/1/2021–6/30/22. Patients included in this study were 18 and older, diagnosed with diabetes, had an undilated fundus photo taken that was “ungradable” for diabetic retinopathy by the EyeArt AI program, and completed a slit lamp exam with the Temple Ophthalmology Department. Records were reviewed for diagnoses at the follow up visit. Statistical analysis was done in Microsoft Excel.

Results : 60 patients were diagnosed as referable by the AI program and completed a face to face follow up exam. 58/60 (97%) of patients were afflicted with at least one ocular pathology when examined by slit lamp. 46/60 (77%) of patients had cataracts in at least one eye. 23/60 (38%) had dry eye disease, 14/60 (23%) had diabetic retinopathy, 11/60 (17%) had glaucoma, glaucoma suspect or ocular hypertension, 5/60 (8%) had hypertensive retinopathy, 3/60 (5%) had epiretinal membrane, 2/60 (3%) had lattice degeneration. One case of each of the following was reported: macular hole, retinal hole, pituitary adenoma, bitemporal hemianopsia, posterior embryotoxon, congenital hypertrophy of retinal pigment, posterior vitreous detachment, iris nevus, Meibomian gland dysfunction, diplopia, pterygium, corneal scar, and posterior capsule opacification.

Conclusions : A diabetic screening program actually screens for multiple eye diseases all of which are important to uncover and treat. In this study population cataracts are a major contributing factor to the quality of fundus imaging, and subsequently the referrable result of “ungradable”. Therefore, the population of patients with ungradable images are an important cohort to examine because there is frequently a treatable eye disease which may be contributing to vision loss unrelated to diabetes.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.