改进U-net++的青光眼视盘视杯分割方法  

The improved U-net++method for glaucoma optic disc and optic cup segmentation

作　　者：刘然 刘建霞 王海翼 LIU Ran;LIU Jianxia;WANG Haiyi(College of Information and Computer,Taiyuan University of Technology,Jinzhong 030600,China)

机构地区：[1]太原理工大学信息与计算机学院,山西晋中030600

出　　处：《电子设计工程》2023年第1期27-33,共7页Electronic Design Engineering

基　　金：山西省回国留学人员科研教研资助项目(HGKY2019040)。

摘　　要：青光眼是当前世界范围内致盲的一种主要病因,其发病过程并没有明显的特征。视杯盘比是青光眼诊断中最主要的评估指标之一。由于眼底图像具有一定的复杂性,视盘视杯分割很容易受到眼底血管和病变区域等的影响,因此传统方法并不能精确地分割出视盘视杯。针对该问题,提出了一种改进U-net++的网络模型算法,数据预处理中引入极坐标变换,在网络的部分关键层引入可变形卷积核代替传统卷积核,在编码器部分引入注意力机制。该算法在Drishti-GS1数据集上的视杯和视盘的Dice系数达到了0.9253和0.9850,oe误差降低到0.06158,较现有的先进算法有一定的提升。Glaucoma is currently a major cause of blindness worldwide,and its pathogenesis has no obvious characteristics.The cup⁃to⁃disc ratio is one of the most important evaluation indicators in the diagnosis of glaucoma.Due to the complexity of the fundus image,the segmentation of the optic disc and optic cup is easily affected by fundus blood vessels and diseased areas,so traditional methods cannot accurately segment the optic disc and optic cup.In response to this problem,an improved U-net++network model algorithm is proposed.Polar coordinate transformation is introduced in data preprocessing,deformable convolution kernels are introduced in some key layers of the network to replace traditional convolution kernels,and attention is introduced in the encoder part.mechanism.The Dice coefficient of the cup optic Disc on the Drishti-GS1 dataset of this algorithm reaches 0.9253 and 0.9850,and the oe error is reduced to 0.06158,which is a certain improvement over the existing advanced algorithms.

关 键 词：青光眼 视杯视盘 U-net++ 极坐标 可变形卷积 注意力机制 

分 类 号：TN957.52[电子电信—信号与信息处理]