Abstract
Purpose :
To assess the accuracy of deep learning models to predict glaucoma development from fundus photographs several years prior to disease onset.
Methods :
We retrospectively included 66,721 fundus photographs from 3,272 eyes of 1,636 subjects that had participated the ocular hypertension treatment study (OHTS). Fundus photographs and visual fields have been previously examined by two independent readers from the optic disc and visual field reading centers of the OHTS with further confirmation by an endpoint committee. We selected all fundus photographs from non-glaucoma eyes and only baseline and two follow up visits of eyes that eventually converted to glaucoma. Therefore, 41,298 fundus photographs that were selected belonged to eyes with elevated IOP without apparent glaucomatous optic neuropathy (GON) and normal visual field, as defined by the OHTS study. We then trained and validated a MobileNetV2 deep learning architecture using 85% of the fundus photographs and further re-tested the models using 15% held-out fundus photographs.
Results :
The selected fundus photographs from those patients who converted to glaucoma were 4-7 years prior to the glaucoma onset. The area under the receiver operating characteristic curve (AUC) of the deep learning model in predicting glaucoma development 4-7 years prior to glaucoma onset was 0.77 (95% confidence interval 0.75 - 0.79).
Conclusions :
From baseline fundus photographs, deep learning models can predict glaucoma development prior to disease onset with reasonable accuracy. Eyes with visual field abnormality but not GON had a higher tendency to be missed by the deep learning algorithms.
This is a 2020 ARVO Annual Meeting abstract.