Joint Optic Nerve Head and Cup Segmentation Based on Deep Residual Network

Glaucoma is an irreversible eye disease that affects human vision adversely. It affects the structure of optic nerve head. To predict glaucoma at an initial stage, it requires accurate screening. For accurate screening, analysis of the subtle variations in the structural features of the optic disk (OD) and optic cup (OC) is required and for that an accurate segmentation of OC and OD is a pre-requisite. In this work, we have introduced a novel architecture for the joint segmentation of OC and OD and refer to that as JS-Net. Unlike most of the state-of-the-art techniques that rely on separate segmentation of OC and OD, in this work, the introduced architecture jointly segments both OD and OC in a single step. The proposed architecture is an extension of the conventional U-Net with extra residual blocks introduced into the encoder and decoder stages. After the segmentation, cup to disk ratio (CDR) is calculated using the joint segmentation map in order to determine whether it is a case of healthy or glaucomatous eye. We have tested the proposed model on two benchmark datasets namely ORIGA and REFUGE. Extensive quantitative as well as qualitative analyses are performed to show that the proposed JS-Net outperforms other existing algorithms. The model achieves an average segmentation accuracy of 98%.