Research On Retinal Vessel Segmentation In Color Fundus Images

Many cardiac and cerebral vascular diseases can cause the changes of human retinal vessel.Retinal vessel segmentation is the basis of the ophthalmic disease computer-aided diagnosis and large-scale screening system.Retinal vessel segmentation aims to generate an accurate vessel binary image from a given color fundus image.Although researchers have done a lot of works,retinal vessel segmentation still can be improved in accuracy.There are many difficulties to be resolved in retinal vessel segmentation,such as the segmentation of microvascular,vessels on the optic disc and intraretinal microvascular abnormalities.This article briefly introduces the significance of retinal blood vessels and the research status at home and abroad,and then some key technical problems of retinal vessel segmentation have been studied.At the same time,some new ideas and algorithms are proposed.The main works and contributions of this thesis are as follows:1、A new method for retinal vessel segmentation based on BP neural network.In view of the characteristics of retinal vessels,we present a new method for vessel segmentation based on BP neural network.This method consists of four steps:histogram equalization of green channel,morphological processing,Gaussian matched filters and Hessian matrix.The fundus vessels are segmented by BP neural network.We conduct the experiments on DRIVE and STARE database.The experimental results show that our method has a good effect on the segmentation of fundus retinal vessels.The average accuracy of DRIVE is 94.65%,and the average accuracy on STARE is 95.13%.2、Robust retinal vessel segmentation via Clustering-Based patch mapping functions.We propose a novel method that uses the clustering based patch mapping functions to segment the retinal vessels.First,all the training gradients and their corresponding ground truths are divided into small patches with a limited overlap.To better analyze the massive patches,a clustering algorithm is employed to cluster these patches.Next,the mapping function between the patches of each cluster and their corresponding patches in ground truths is solved based on the fact that the patches in one cluster have the similar structure and share the same mapping function.The mapping functions are simple and fast,and can act as a bridge between the vessel in color fundus image and its ground truth.With these mapping functions,we can quickly and easily segment the blood vessels in fundus images.Although the clustering in large scale training data is time consuming,it can be computed offline and only once for the application.In the test phase,the computational load is relatively low,as this kind of method only needs to figure out which cluster the local data should belong to.The method is tested on the public fundus database.The average accuracy is 95.38%on the DRIVE database,and the average accuracy on the STARE database is 95.32%.