Association of optic disc parameters and glaucoma incidence based on automated segmentation, evidence from the UK Biobank

Abstract

AbstractObjectiveTo assess the correlation between glaucoma incidence and optic disc parameters obtained through an automated deep learning (DL) algorithm segmentation.Methods and AnalysisWe obtained eligible fundus photographs and corresponding participant data from the UK Biobank. To accurately assess the optic disc parameters and their relationship with glaucoma incidence using Cox proportional hazard regression models, we developed a DL algorithm that automatically segmented the optic disc and cup and calculated various parameters including the vertical cup-to-disc ratio (VCDR), ovality index, cup-to-disc area ratio, rim area, disc area, and disc rotation from the fundus photos. We performed two logistic regression models, with model A comprising sociodemographic and health covariates and model B including additional ophthalmic features. Receiver operating characteristic curves (ROC) and areas under the curve (AUC) were plotted and calculated for each model to evaluate their performance.ResultsA total of 44,376 subjects with fundus photos were included in our study. After a median follow-up of 10.1 years, 354 incident glaucoma were documented. Subjects with larger VCDR had a higher risk of incident glaucoma; the HR (95% CI) was 2.05 (1.57-2.66) in the multivariable-adjusted model (p<0.001). The results remain significant in the sensitivity analysis that excluded fundus photographs with “Reject” quality. After adding the optic disc parameters into the regression model A, the AUC increased by 4.2% to 78.6%.ConclusionThe VCDR calculated by automatic optic disc segmentation model shows potential as a biomarker for evaluating the risk of glaucoma.What is already known on this topicGlaucoma is a worldwide leading cause of irreversible vision loss, and its early diagnosis is of great necessity.What this study addsData from the UK Biobank shows the optic disc parameters and their relationship with glaucoma incidence.We develop a DL-based algorithm for optic disc segmentation in Color fundus photos and validate its efficacy in glaucoma prediction.How this study might affect research, practice or policyThe VCDR calculated using an automatic optic disc segmentation based on a DL model can serve as a biomarker to predict the incidence of glaucoma.