基于改进U-Net 的视网膜血管分割算法  

Retinal blood vessel segmentation algorithm based onimproved U-Net

作　　者：刘远 李柏承[1] 吴春波 LIU Yuan;LI Baicheng;WU Chunbo(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《光学仪器》2024年第5期9-16,共8页Optical Instruments

基　　金：国家自然科学基金(62005165)。

摘　　要：在面对视网膜图像中细小血管时,现有算法存在分割精度低的问题。通过在U-Net中引入残差模块与细节增强注意力机制模块,提出了一种改进的U-Net分割算法。在编解码阶段,用残差模块取代传统的卷积模块,解决了网络随深度增加而退化的问题;同时在编码器和解码器间增加细节增强注意力机制,减少编码器输出中的无用信息,从而提高网络抓取有效特征信息的敏感度。此外,基于标准图像集DRIVE的实验结果表明,所提算法的分割准确率、灵敏度与F1值较U-Net算法分别提高了0.46%,2.14%,1.56%,优于传统分割算法。The existing algorithms have the problem of low segmentation accuracy when facing small vessels in retinal images.In this paper,an improved U-Net segmentation algorithm is proposed by introducing residual module and detail enhancement attention mechanism module into U-Net network.In the coding and decoding stages,the residual module is used to replace the traditional convolutional module,which solves the problem of network degradation with increasing depth.Meanwhile,a detail enhancement attention mechanism is added between the encoder and the decoder to reduce the useless information in the output of the encoder,so that the sensitivity of the network to capture valid feature information is improved.In addition,the experimental results based on the standard image set DRIVE reveal that the segmentation accuracy,sensitivity and F1 score of the proposed algorithm are improved by 0.46%,2.14%and 1.56%,respectively compared to the U-Net,which is superior to the traditional segmentation algorithms.

关 键 词：图像分割 视网膜血管 细节增强 残差模块 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]