Optic Disc and Cup Segmentation Based on Deep Learning

Abstract

Glaucoma is a disease that damages eye’s optic nerve, and it is the leading cause of global irreversible blindness. Optic nerve head (ONH) assessment is a convenient way to detect glaucoma early and cup to disc ratio (CDR) is an important index for ONH evaluation. Thus, it is a fundamental task to segment OD and OC from the fundus images automatically and accurately. Most existing method segment them separately, and rely on hand-crafted visual feature from fundus image. This paper presents universal approach for automatic optic disc and cup segmentation, which is based on deep learning, namely, modification of fully convolutional network (FCN) and the Inception building blocks in GoogleNet. For improving the segmentation performance further, we also introduce new optic disc localization and pre-processing method. Our experiments show that our method achieves quality comparable to current state-of-the-art methods.