基于眼底图像的视网膜血管分割方法综述

Literature Review on Fundus Image Based Retinal Vascular Segmentation Method

视网膜血管分割是眼底图像研究的基础。探讨视网膜血管分割方法的现状,总结不同分割方法的优缺点,综述基于眼底图像的血管分割方法。根据视网膜图像特点,分析基于窗口、基于分类和基于跟踪三类方法;根据是否采用图像的特征数据规则,研讨监督方法和非监督方法。研究对比发现,基于窗口的方法大多结合滤波器实施分割,因此计算量较大;基于分类的方法需人工提取特征,因此算法效率低;基于跟踪的方法过于依赖初始种子点的选取;非监督方法需自行制定判断规则,受制于规则的适用性;监督方法需要大量先验数据进行建模,技术上存在瓶颈。后续应借助深度学习技术,结合神经网络算法,训练得到更加精准的分类模型,进一步提升视网膜血管分割精度和效率。

Retinal vascular segmentation is the basis of fundus image research.Discussed is the development of ret inal vascular segmentat ion technology,summar ized are the advantages and disadvantages of different algorithms,and reviewed are the vascular segmentation methods based on fundus image.According to the characteristics of retinal image,window based,classification based and tracking based methods are analyzed.From whether to adopt the feature data rules provided by images,supervised methods and unsupervised methods are summarized.Through study and comparison,it is found that the window based method needs to be combined with a filter,so the calculation amount is large;the classification based method needs to manually extract features,so the algorithm efficiency is low;the tracking based method relies much on the selection of initial seed points;the unsupervised method needs to make its own judgment rules,which is subject to the applicability of the rules;the supervised method needs a large number of prior data for modeling,which makes it difficult to advance technologically.Deep learning technology and neural network algorithm should be used to train more accurate classification models and further improve the accuracy and efficiency of retinal vascular segmentation in the future.