The Comparison of Retinal Vessel Segmentation Methods in Fundus Images

Abstract

Due to complexity of retinal fundus images, they are usually affected with noise and lighting during image acquisition. It brings difficulty to segment retinal vessels accurately, so accurate retinal vessel segmentation is still a challenging task in fundus images analysis. Five kinds of typical retinal vessel segmentation methods are briefly introduced in this paper, which are based on thresholding, matched filtering, mathematical morphology, tracking, and deep learning. Each method has its own characteristics. The experiment results show that the method based on deep learning is the best segmentation method among them, which can effectively assist doctors to detect and diagnose cardiovascular and ophthalmic diseases in early stage, and provide the decision support for ophthalmic disease computer-aided diagnosis and establishment of large-scale screening system.