Optic Cup Segmentation Method by a Modified VGG-16 Network

Glaucoma is a chronic eye disease in which the optic nerve is progressively damaged. As it cannot be cured, the best way to prevent visual damage is early detection and subsequent treatment. The optic nerve head examination, which involves measurement of cup-to-disc ratio, is considered one of the most valuable methods of structural diagnosis of the disease. Segmentation of optic cup on retinal fundus images can be used to estimate cup-to-disc ratio. This paper presents a novel approach for automatic optic cup segmentation, which is based on deep learning, namely, modified VGG-16 network and transfer learning technique. The modified network combines the residual, squeeze-and-excitation and multiscale feature. Our proposed method is tested on publicly available databases GlaucomaRepo and Drishti-GS. The evaluation of proposed method contains comparison with the original VGG-16 network and other state-of-the-art methods on above two fundus datasets which are captured from different devices. Experimental results show that our method outperforms the existing methods in robustness and accuracy.