基于改进U⁃Net的视盘视杯联合分割方法  

Optic disk and cup joint segmentation network based on improved U⁃Net

作　　者：周利涛 王志超 施璜浩 常珊 Zhou Litao;Wang Zhichao;Shi Huanghao;Chang Shan(Institute of Bioinformatics and Medical Engineering,School of Electrical and Information Engineering,Jiangsu University of Technology,Changzhou 213001,China)

机构地区：[1]江苏理工学院电气信息工程学院生物信息与医药工程研究所,常州213001

出　　处：《现代计算机》2024年第3期48-53,60,共7页Modern Computer

基　　金：国家自然科学基金青年项目(81603152);常州市科技支撑计划(社会发展)(CE20205033)。

摘　　要：青光眼是一种不可逆的致盲性眼疾,疾病早期症状不明显使得许多患者错失治疗的最佳时机。眼底照相作为最常见的青光眼筛查手段,眼底杯盘比值是诊断青光眼的重要指标之一。针对图像中视盘视杯分割精度不高的问题,构建了一种改进U⁃Net的视盘视杯联合分割模型CASSP⁃Net,引入CBAM注意力机制和空洞空间金字塔结构,进一步提升视盘视杯联合分割的精确度,在Drishti⁃GS和REFUGE数据集中进行测试,在Dice和IoU上分别获得92.03%和85.23%的较好表现。Glaucoma is an irreversible blinding eye disease.The early symptoms of the disease are not obvious,causing many patients to miss the best opportunity for treatment.Fundus photography is a common glaucoma screening method,and the fundus cup⁃to⁃disc ratio is one of the important indicators for diagnosing glaucoma.In order to solve the problem of low optic disc and optic cup segmentation accuracy in images,an improved U⁃Net optic disc and optic cup joint segmentation model CASSP⁃Net was con⁃structed.The CBAM attention mechanism and hole space pyramid structure were introduced to further improve the optic disc and optic cup joint segmentation.The accuracy was tested in the Drishti⁃GS and REFUGE data sets,and achieved good performance of 92.03%and 85.23%on Dice and IoU respectively.

关 键 词：青光眼 视盘 视杯 眼底图像分割 深度学习 

分 类 号：R775[医药卫生—眼科] TP391.41[医药卫生—临床医学]