Retinal vessel segmentation using simple SPCNN model and line connector

Abstract

The effective segmentation of retinal blood vessels is essential for the medical diagnosis of ophthalmology diseases. In this paper, a novel approach is presented to segment retinal vessels accurately and efficiently. Firstly, we propose a simple simplified pulse coupled neural network utilizing the similarity of adjacent neurons to acquire the basic structure of blood vessels. Then we apply a line connector to solve the problem of broken vessels occurring in the segmentation, in order to present a complete structure of the blood vessels and improve the accuracy of vessel identification. Experimental analyses on two publicly available databases show that the proposed methods with or without the line connector outperform the most existing methods in terms of average accuracy and have a fast response time. It is of great importance for medical diagnosis with high accuracy and short time consumption. Our methods are practicable either for retinal vessel segmentation, or for other applications of clinical research.