Automatic Optic Disk and Cup Segmentation of Fundus Images Using Deep Learning

Abstract

Automatic segmentation of optic disk (OD) and cup regions in fundus images is essential in deriving clinical parameters, such as, cup-to-disk ratio (CDR), to assist glaucoma diagnosis. This paper presents a deep learning system using fully convolutional neural networks (FCN) to perform such segmentation, discusses various strategies on how to leverage multiple doctor annotations and prioritize pixels belonging to different regions while training the neural network. Experimental evaluations on Drishti-GS dataset demonstrate that the presented method achieves comparable and superior F-score to prior work on optic disk and cup segmentation, respectively.