Article(id=1217789889345667072, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1217789884081820362, articleNumber=null, orderNo=null, doi=10.12404/j.issn.1671-1815.2406155, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=research-article, receivedDate=1723824000000, receivedDateStr=2024-08-17, revisedDate=1744214400000, revisedDateStr=2025-04-10, acceptedDate=null, acceptedDateStr=null, onlineDate=1768273335061, onlineDateStr=2026-01-13, pubDate=1753632000000, pubDateStr=2025-07-28, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1768273335061, onlineIssueDateStr=2026-01-13, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1768273335061, creator=13701087609, updateTime=1768273335061, updator=13701087609, issue=Issue{id=1217789884081820362, tenantId=1146029695717560320, journalId=1146123166801305609, year='2025', volume='25', issue='21', pageStart='8761', pageEnd='9209', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1768273333807, creator=13701087609, updateTime=1768273602927, updator=13701087609, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1217791012932604619, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1217789884081820362, language=EN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1217791012932604620, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1217789884081820362, language=CN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=8993, endPage=9001, ext={EN=ArticleExt(id=1217789889911898150, articleId=1217789889345667072, tenantId=1146029695717560320, journalId=1146123166801305609, language=EN, title=Retinal Vessel Segmentation Network Based on Self-attention Mechanism and Lateral Output Loss Function, columnId=1156262729162810294, journalTitle=Science Technology and Engineering, columnName=Papers·Automation and Computational Technology, runingTitle=null, highlight=null, articleAbstract=

Image segmentation is a fundamental problem in medical image analysis, the typical deep learning based UNet architecture (UNet) and its variants are widely used in retinal vessel segmentation. However, the UNet network extracts feature information from images through local convolution modules, which makes the global information of the images difficult to be correlated and the long-distance dependencies between pixels difficult to be effectively captured. Considering the problems with the UNet network model and the characteristics of retinal vascular images, an attention module was added to the skip connections of UNet to capture long-distance dependencies between blood vessels. In addition, to enhance the segmentation ability of the network, the group normalization(GN) was used instead of the original batch normalization (BN) of the UNet network model, and the corresponding groups were selected for different channels. To update parameters and optimize the network, the final cross entropy loss function was designed using the side output layer and the final output layer. Experiments are implemented on the DRIVE dataset and CHASEDB1 dataset, and the experimental results show that the proposed model has better image segmentation performance.

, correspAuthors=Ying-mei WANG, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, authorCompany=null, fund=null, authors=null, authorsList=Zhen-hua YU, Ben-cong YAN, Ying-mei WANG), CN=ArticleExt(id=1217789891891609816, articleId=1217789889345667072, tenantId=1146029695717560320, journalId=1146123166801305609, language=CN, title=基于自注意力机制和侧输出损失函数的视网膜血管分割网络, columnId=1156262729783567290, journalTitle=科学技术与工程, columnName=论文·自动化技术、计算机技术, runingTitle=null, highlight=null, articleAbstract=

图像分割是医学图像分析中的一个基本问题,基于深度学习的典型UNet架构(UNet architecture)分割网络模型及其变式被广泛应用于视网膜血管分割之中。但是UNet网络通过局部卷积模块提取图像的特征信息,难以关联图像的全局信息,无法有效捕捉像素之间的长距离依赖关系。考虑到UNet网络模型存在的问题和视网膜血管图像的特点,在UNet跳跃连接中加入注意力模块,可以捕捉血管之间的长距离依赖关系。此外,为增强网络的分割能力,使用群归一化(group normalization,GN)代替UNet网络模型原始的批归一化(batch normalization,BN),对不同的通道选择对应的分组。为更新参数和优化网络,利用侧输出层和最后的输出层设计了交叉熵损失函数。在DRIVE数据集和CHASEDB1数据集上进行了实验,实验结果表明所提出的模型有更好的图像分割效果。

, correspAuthors=王迎美, authorNote=null, correspAuthorsNote=
* 王迎美(1987—),女,汉族,山东新泰人,博士,副教授。研究方向:医学图像处理与重建。E-mail:
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=KDnacJsG4yxf+ngLx7QdJA==, magXml=eD/wedAE2/YYfPM1jdb0lg==, pdfUrl=null, pdf=84lcrcOjsZTQIiZX/rDwwA==, pdfFileSize=11966888, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=qeOen5uEPUhKGUHkBi+5Dw==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=xwwYgwL8i6HqZAFLoJscrg==, mapNumber=null, authorCompany=null, fund=null, authors=

于振华(2000—),男,汉族,山东滨州人,硕士研究生。研究方向:医学图像处理。E-mail:

, authorsList=于振华, 闫本聪, 王迎美)}, authors=[Author(id=1217860109099778188, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=18766638125@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1217860109263356058, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, authorId=1217860109099778188, language=EN, stringName=Zhen-hua YU, firstName=Zhen-hua, middleName=null, lastName=YU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Mathematics and Statistics, Shandong University of Technology, Zibo 255000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1217860109435322541, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, authorId=1217860109099778188, language=CN, stringName=于振华, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=山东理工大学数学与统计学院, 淄博 255000, bio={"content":"

于振华(2000—),男,汉族,山东滨州人,硕士研究生。研究方向:医学图像处理。E-mail:

"}, bioImg=null, bioContent=

于振华(2000—),男,汉族,山东滨州人,硕士研究生。研究方向:医学图像处理。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1217860108902645877, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, xref=null, ext=[AuthorCompanyExt(id=1217860108915228791, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, companyId=1217860108902645877, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Mathematics and Statistics, Shandong University of Technology, Zibo 255000, China), AuthorCompanyExt(id=1217860108923617400, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, companyId=1217860108902645877, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=山东理工大学数学与统计学院, 淄博 255000)])]), Author(id=1217860109548568764, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1217860109762478289, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, authorId=1217860109548568764, language=EN, stringName=Ben-cong YAN, firstName=Ben-cong, middleName=null, lastName=YAN, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Mathematics and Statistics, Shandong University of Technology, Zibo 255000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1217860109900890335, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, authorId=1217860109548568764, language=CN, stringName=闫本聪, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=山东理工大学数学与统计学院, 淄博 255000, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1217860108902645877, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, xref=null, ext=[AuthorCompanyExt(id=1217860108915228791, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, companyId=1217860108902645877, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Mathematics and Statistics, Shandong University of Technology, Zibo 255000, China), AuthorCompanyExt(id=1217860108923617400, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, companyId=1217860108902645877, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=山东理工大学数学与统计学院, 淄博 255000)])]), Author(id=1217860110014136560, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=yingmeiwang@sdut.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1217860110127382781, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, authorId=1217860110014136560, language=EN, stringName=Ying-mei WANG, firstName=Ying-mei, middleName=null, lastName=WANG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=School of Mathematics and Statistics, Shandong University of Technology, Zibo 255000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1217860111477948683, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, authorId=1217860110014136560, language=CN, stringName=王迎美, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=山东理工大学数学与统计学院, 淄博 255000, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1217860108902645877, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, xref=null, ext=[AuthorCompanyExt(id=1217860108915228791, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, companyId=1217860108902645877, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Mathematics and Statistics, Shandong University of Technology, Zibo 255000, China), AuthorCompanyExt(id=1217860108923617400, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, companyId=1217860108902645877, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=山东理工大学数学与统计学院, 淄博 255000)])])], keywords=[Keyword(id=1217860111834464551, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, orderNo=1, keyword=UNet), Keyword(id=1217860111964487990, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, orderNo=2, keyword=Swin-UNet), Keyword(id=1217860112287449418, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, orderNo=3, keyword=group normalization), Keyword(id=1217860112417472857, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, orderNo=4, keyword=lateral output layers), Keyword(id=1217860112585245033, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, orderNo=5, keyword=retinal blood vessel segmentation), Keyword(id=1217860112706879864, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, orderNo=1, keyword=UNet), Keyword(id=1217860112895623574, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, orderNo=2, keyword=Swin-UNet), Keyword(id=1217860113004675489, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, orderNo=3, keyword=群归一化), Keyword(id=1217860113117921714, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, orderNo=4, keyword=侧输出层), Keyword(id=1217860113252139460, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, orderNo=5, keyword=视网膜血管分割)], refs=[Reference(id=1217860117303837442, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2023, volume=261, issue=11, pageStart=3067, pageEnd=3082, url=null, language=null, rfNumber=[1], rfOrder=0, authorNames=Soleimani M, Cheraqpour K, Koganti R, journalName=Graefe's Archive for Clinical and Experimental Ophthalmology, refType=null, unstructuredReference=Soleimani M, Cheraqpour K, Koganti R, et al. Cellular senescence and ophthalmic diseases: narrative review[J]. Graefe's Archive for Clinical and Experimental Ophthalmology, 2023, 261(11): 3067-3082., articleTitle=Cellular senescence and ophthalmic diseases: narrative review, refAbstract=null), Reference(id=1217860117412889359, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2020, volume=392, issue=null, pageStart=221, pageEnd=232, url=null, language=null, rfNumber=[2], rfOrder=1, authorNames=Yu L, Qin Z, Zhuang T, journalName=Neurocomputing, refType=null, unstructuredReference=Yu L, Qin Z, Zhuang T, et al. A framework for hierarchical division of retinal vascular networks[J]. Neurocomputing, 2020, 392: 221-232., articleTitle=A framework for hierarchical division of retinal vascular networks, refAbstract=null), Reference(id=1217860117542912792, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2024, volume=128, issue=null, pageStart=107454, pageEnd=null, url=null, language=null, rfNumber=[3], rfOrder=2, authorNames=Qin Q, Chen Y, journalName=Engineering Applications of Artificial Intelligence, refType=null, unstructuredReference=Qin Q, Chen Y. A review of retinal vessel segmentation for fundus image analysis[J]. Engineering Applications of Artificial Intelligence, 2024, 128: 107454., articleTitle=A review of retinal vessel segmentation for fundus image analysis, refAbstract=null), Reference(id=1217860117735850799, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2015, volume=null, issue=null, pageStart=234, pageEnd=241, url=null, language=null, rfNumber=[4], rfOrder=3, authorNames=Ronneberger O, Fischer P, Brox T, journalName=International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin:Springer International Publishing, refType=null, unstructuredReference=Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation[C]// International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin:Springer International Publishing, 2015: 234-241., articleTitle=U-Net: convolutional networks for biomedical image segmentation, refAbstract=null), Reference(id=1217860117874262843, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2024, volume=24, issue=24, pageStart=10362, pageEnd=10369, url=null, language=null, rfNumber=[5], rfOrder=4, authorNames=汪南洋, 沈疆海, journalName=科学技术与工程, refType=null, unstructuredReference=汪南洋, 沈疆海. 基于MSHAM-UNet的岩心孔洞图像分割方法[J]. 科学技术与工程, 2024, 24(24): 10362-10369., articleTitle=基于MSHAM-UNet的岩心孔洞图像分割方法, refAbstract=null), Reference(id=1217860118125921098, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2024, volume=24, issue=24, pageStart=10362, pageEnd=10369, url=null, language=null, rfNumber=[5], rfOrder=5, authorNames=Wang Nanyang, Shen Jianghai, journalName=Science Technology and Engineering, refType=null, unstructuredReference=Wang Nanyang, Shen Jianghai. Image segmentation method of rock core hole based on MSHAM-UNet[J]. Science Technology and Engineering, 2024, 24(24): 10362-10369., articleTitle=Image segmentation method of rock core hole based on MSHAM-UNet, refAbstract=null), Reference(id=1217860118302081876, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2017, volume=28, issue=6, pageStart=1303, pageEnd=1314, url=null, language=null, rfNumber=[6], rfOrder=6, authorNames=Zhang K, Sun M, Han T X, journalName=IEEE Transactions on Circuits and Systems for Video Technology, refType=null, unstructuredReference=Zhang K, Sun M, Han T X, et al. Residual networks of residual networks: multilevel residual networks[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2017, 28(6): 1303-1314., articleTitle=Residual networks of residual networks: multilevel residual networks, refAbstract=null), Reference(id=1217860118570517347, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=2108, pageEnd=null, url=null, language=null, rfNumber=[7], rfOrder=7, authorNames=Baozhou Z, Hofstee P, Lee J, journalName=ArXiv, refType=null, unstructuredReference=Baozhou Z, Hofstee P, Lee J, et al. An attention module for convolutional neural networks[J]. ArXiv, 2021: 2108.08205., articleTitle=An attention module for convolutional neural networks, refAbstract=null), Reference(id=1217860118679569264, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=2818, pageEnd=2826, url=null, language=null, rfNumber=[8], rfOrder=8, authorNames=Szegedy C, Vanhoucke V, Ioffe S, journalName=IEEE Conference on Computer Vision and Pattern Recognition (CVPR), refType=null, unstructuredReference=Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 2818-2826., articleTitle=Rethinking the inception architecture for computer vision, refAbstract=null), Reference(id=1217860118797009784, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2022, volume=51, issue=2, pageStart=241, pageEnd=249, url=null, language=null, rfNumber=[9], rfOrder=9, authorNames=田会娟, 翟佳豪, 柳建新, journalName=光子学报, refType=null, unstructuredReference=田会娟, 翟佳豪, 柳建新, 等. 基于SRN-UNet 的低质量虹膜分割算法[J]. 光子学报, 2022, 51(2): 241-249., articleTitle=基于SRN-UNet 的低质量虹膜分割算法, refAbstract=null), Reference(id=1217860118956393348, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2022, volume=51, issue=2, pageStart=241, pageEnd=249, url=null, language=null, rfNumber=[9], rfOrder=10, authorNames=Tian Huijuan, Zhai Jiahao, Liu Jianxin, journalName=Acta Photonica Sinica, refType=null, unstructuredReference=Tian Huijuan, Zhai Jiahao, Liu Jianxin, et al. A low-quality iris image segmentation algorithm based on SRN-UNet[J]. Acta Photonica Sinica, 2022, 51(2): 241-249., articleTitle=A low-quality iris image segmentation algorithm based on SRN-UNet, refAbstract=null), Reference(id=1217860119119971215, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2021, volume=26, issue=9, pageStart=2146, pageEnd=2155, url=null, language=null, rfNumber=[10], rfOrder=11, authorNames=郭宁, 柏正尧, journalName=中国图象图形学报, refType=null, unstructuredReference=郭宁, 柏正尧. 注意力机制下密集空洞卷积的肺部图像分割[J]. 中国图象图形学报, 2021: 26(9): 2146-2155., articleTitle=注意力机制下密集空洞卷积的肺部图像分割, refAbstract=null), Reference(id=1217860120063689626, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2021, volume=26, issue=9, pageStart=2146, pageEnd=2155, url=null, language=null, rfNumber=[10], rfOrder=12, authorNames=Guo Ning, Bai Zhengyao, journalName=Journal of Image and Graphics, refType=null, unstructuredReference=Guo Ning, Bai Zhengyao. The integration of attention mechanism and dense atrous convolution for lung image segmentation[J]. Journal of Image and Graphics, 2021, 26(9): 2146-2155., articleTitle=The integration of attention mechanism and dense atrous convolution for lung image segmentation, refAbstract=null), Reference(id=1217860120227267488, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=1706, pageEnd=null, url=null, language=null, rfNumber=[11], rfOrder=13, authorNames=Vaswani A, Shazeer N, Parmar N, journalName=ArXiv Preprint ArXiv, refType=null, unstructuredReference=Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. ArXiv Preprint ArXiv, 2017: 1706.03762., articleTitle=Attention is all you need, refAbstract=null), Reference(id=1217860120403428262, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2023, volume=23, issue=16, pageStart=6974, pageEnd=6982, url=null, language=null, rfNumber=[12], rfOrder=14, authorNames=汤翔中, 高丙朋, journalName=科学技术与工程, refType=null, unstructuredReference=汤翔中, 高丙朋. 融合注意力空洞卷积和Transformer的矿石图像分割[J]. 科学技术与工程, 2023, 23(16): 6974-6982., articleTitle=融合注意力空洞卷积和Transformer的矿石图像分割, refAbstract=null), Reference(id=1217860120604754868, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2023, volume=23, issue=16, pageStart=6974, pageEnd=6982, url=null, language=null, rfNumber=[12], rfOrder=15, authorNames=Tang Xiangzhong, Gao Bingpeng, journalName=Science Technology and Engineering, refType=null, unstructuredReference=Tang Xiangzhong, Gao Bingpeng. Ore image segmentation based on attention hole convolution and Transformer[J]. Science Technology and Engineering, 2023, 23(16): 6974-6982., articleTitle=Ore image segmentation based on attention hole convolution and Transformer, refAbstract=null), Reference(id=1217860120814470083, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=205, pageEnd=218, url=null, language=null, rfNumber=[13], rfOrder=16, authorNames=Cao H, Wang Y, Chen J, journalName=European Conference on Computer Vision, refType=null, unstructuredReference=Cao H, Wang Y, Chen J, et al. Swin-UNet: UNet-like pure Transformer for medical image segmentation[C]// European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 205-218., articleTitle=Swin-UNet: UNet-like pure Transformer for medical image segmentation, refAbstract=null), Reference(id=1217860120986436559, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=1055, pageEnd=1059, url=null, language=null, rfNumber=[14], rfOrder=17, authorNames=Huang H, Lin L, Tong R, journalName=ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), refType=null, unstructuredReference=Huang H, Lin L, Tong R, et al. UNet 3+: a full-scale connected UNet for medical image segmentation[C]// ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Barcelona: IEEE, 2020: 1055-1059., articleTitle=UNet 3+: a full-scale connected UNet for medical image segmentation, refAbstract=null), Reference(id=1217860121175180249, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=74, pageEnd=84, url=null, language=null, rfNumber=[15], rfOrder=18, authorNames=Xiang T, Zhang C, Liu D, journalName=Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin:Springer, refType=null, unstructuredReference=Xiang T, Zhang C, Liu D, et al. BiO-Net: learning recurrent bi-directional connections for encoder-decoder architecture[C]// Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin:Springer, 2020: 74-84., articleTitle=BiO-Net: learning recurrent bi-directional connections for encoder-decoder architecture, refAbstract=null), Reference(id=1217860121292620770, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2024, volume=53, issue=3, pageStart=2027, pageEnd=2037, url=null, language=null, rfNumber=[16], rfOrder=19, authorNames=Ibrahim H A, Mohamed A E N, Ammar R, journalName=Journal of Optics, refType=null, unstructuredReference=Ibrahim H A. Mohamed A E N, Ammar R, et al. Efficient color image enhancement using piecewise linear transformation and gamma correction[J]. Journal of Optics, 2024, 53(3): 2027-2037., articleTitle=Efficient color image enhancement using piecewise linear transformation and gamma correction, refAbstract=null), Reference(id=1217860121418449899, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=10033, pageEnd=10041, url=null, language=null, rfNumber=[17], rfOrder=20, authorNames=Wu K, Peng H, Chen M, journalName=Proceedings of the IEEE/CVF International Conference on Computer Vision, refType=null, unstructuredReference=Wu K, Peng H, Chen M, et al. Rethinking and improving relative position encoding for vision transformer[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Montreal: IEEE, 2021: 10033-10041., articleTitle=Rethinking and improving relative position encoding for vision transformer, refAbstract=null), Reference(id=1217860121573639155, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=8791, pageEnd=8800, url=null, language=null, rfNumber=[18], rfOrder=21, authorNames=Song Z, Yu J, Chen Y P P, journalName=Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, refType=null, unstructuredReference=Song Z, Yu J, Chen Y P P, et al. Transformer tracking with cyclic shifting window attention[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans: IEEE, 2022: 8791-8800., articleTitle=Transformer tracking with cyclic shifting window attention, refAbstract=null), Reference(id=1217860121691079679, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2019, volume=39, issue=6, pageStart=1856, pageEnd=1867, url=null, language=null, rfNumber=[19], rfOrder=22, authorNames=Zhou Z, Siddiquee M M R, Tajbakhsh N, journalName=IEEE Transactions on Medical Imaging, refType=null, unstructuredReference=Zhou Z, Siddiquee M M R, Tajbakhsh N, et al. UNet++: redesigning skip connections to exploit multiscale features in image segmentation[J]. IEEE Transactions on Medical Imaging, 2019, 39(6): 1856-1867., articleTitle=UNet++: redesigning skip connections to exploit multiscale features in image segmentation, refAbstract=null), Reference(id=1217860121884016651, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=4331, pageEnd=4339, url=null, language=null, rfNumber=[20], rfOrder=23, authorNames=Barron J T, journalName=Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, refType=null, unstructuredReference=Barron J T. A general and adaptive robust loss function[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 4331-4339., articleTitle=A general and adaptive robust loss function, refAbstract=null), Reference(id=1217860122060177430, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2004, volume=23, issue=4, pageStart=501, pageEnd=509, url=null, language=null, rfNumber=[21], rfOrder=24, authorNames=Staal J, Abràmoff M D, Niemeijer M, journalName=IEEE Transactions on Medical Imaging, refType=null, unstructuredReference=Staal J, Abràmoff M D, Niemeijer M, et al. Ridge-based vessel segmentation in color images of the retina[J]. IEEE Transactions on Medical Imaging, 2004, 23(4): 501-509., articleTitle=Ridge-based vessel segmentation in color images of the retina, refAbstract=null), Reference(id=1217860122173423647, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2009, volume=50, issue=5, pageStart=2004, pageEnd=2010, url=null, language=null, rfNumber=[22], rfOrder=25, authorNames=Owen C G, Rudnicka A R, Mullen R, journalName=Investigative Ophthalmology & Visual Science, refType=null, unstructuredReference=Owen C G, Rudnicka A R, Mullen R, et al. Measuring retinal vessel tortuosity in 10-year-old children: validation of the computer-assisted image analysis of the retina (CAIAR) program[J]. Investigative Ophthalmology & Visual Science, 2009, 50(5): 2004-2010., articleTitle=Measuring retinal vessel tortuosity in 10-year-old children: validation of the computer-assisted image analysis of the retina (CAIAR) program, refAbstract=null), Reference(id=1217860122269892649, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2023, volume=13, issue=null, pageStart=1077669, pageEnd=null, url=null, language=null, rfNumber=[23], rfOrder=26, authorNames=Tan T E, Wong T Y, journalName=Frontiers in Endocrinology, refType=null, unstructuredReference=Tan T E, Wong T Y. Diabetic retinopathy: looking forward to 2030[J]. Frontiers in Endocrinology, 2023, 13: 1077669., articleTitle=Diabetic retinopathy: looking forward to 2030, refAbstract=null), Reference(id=1217860122404110392, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=1607, pageEnd=null, url=null, language=null, rfNumber=[24], rfOrder=27, authorNames=Ba J L, Kiros J R, Hinton G E, journalName=ArXiv Preprint ArXiv, refType=null, unstructuredReference=Ba J L, Kiros J R, Hinton G E. Layer normalization[J]. ArXiv Preprint ArXiv, 2016: 1607.06450., articleTitle=Layer normalization, refAbstract=null), Reference(id=1217860122538328124, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=2102, pageEnd=null, url=null, language=null, rfNumber=[25], rfOrder=28, authorNames=Chen J, Lu Y, Yu Q, journalName=ArXiv Preprint ArXiv, refType=null, unstructuredReference=Chen J, Lu Y, Yu Q, et al. TransUNet: Transformers make strong encoders for medical image segmentation[J]. ArXiv Preprint ArXiv, 2021: 2102.04306., articleTitle=TransUNet: Transformers make strong encoders for medical image segmentation, refAbstract=null)], funds=[Fund(id=1217860117031207664, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, awardId=ZR2022MA027, language=CN, fundingSource=山东省自然科学基金面上项目(ZR2022MA027), fundOrder=null, country=null)], companyList=[AuthorCompany(id=1217860108902645877, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, xref=null, ext=[AuthorCompanyExt(id=1217860108915228791, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, companyId=1217860108902645877, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Mathematics and Statistics, Shandong University of Technology, Zibo 255000, China), AuthorCompanyExt(id=1217860108923617400, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, companyId=1217860108902645877, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=山东理工大学数学与统计学院, 淄博 255000)])], figs=[ArticleFig(id=1217860113478631911, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, label=Fig.1, caption=The network structure of AttenTransUnet, figureFileSmall=QIwryr0V8bzlZNIcWJG31g==, figureFileBig=2PNgwJXlUUr5PKHcwe0lvg==, tableContent=null), ArticleFig(id=1217860113617043961, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, label=图1, caption=AttenTransUnet的网络结构, figureFileSmall=QIwryr0V8bzlZNIcWJG31g==, figureFileBig=2PNgwJXlUUr5PKHcwe0lvg==, tableContent=null), ArticleFig(id=1217860113755456008, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, label=Fig.2, caption=The U-shaped structure of the Attention Block skip connection module, figureFileSmall=tCgpiwvkQSFq2DFZ3rH/Jw==, figureFileBig=Ukgv8PdeaeyWMq1zbcIkQw==, tableContent=null), ArticleFig(id=1217860113914839578, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, label=图2, caption=Attention Block跳跃连接模块的U形结构, figureFileSmall=tCgpiwvkQSFq2DFZ3rH/Jw==, figureFileBig=Ukgv8PdeaeyWMq1zbcIkQw==, tableContent=null), ArticleFig(id=1217860114095194672, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, label=Fig.3, caption=Gamma correction images for different channels, figureFileSmall=mkMeSXkBh8pcTX+3sRLHYQ==, figureFileBig=Nwo0yawK25B2Szfk64Lt3g==, tableContent=null), ArticleFig(id=1217860114191663678, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, label=图3, caption=不同通道的Gamma矫正图, figureFileSmall=mkMeSXkBh8pcTX+3sRLHYQ==, figureFileBig=Nwo0yawK25B2Szfk64Lt3g==, tableContent=null), ArticleFig(id=1217860114342658644, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, label=Fig.4, caption=Experimental results of different comparison methods, figureFileSmall=MJe9QEbe2b7ETCCmturcuQ==, figureFileBig=YnstDjPt5R9tdBvJbtTWRw==, tableContent=null), ArticleFig(id=1217860114531402346, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, label=图4, caption=各种比对算法的实验结果, figureFileSmall=MJe9QEbe2b7ETCCmturcuQ==, figureFileBig=YnstDjPt5R9tdBvJbtTWRw==, tableContent=null), ArticleFig(id=1217860114632065657, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, label=Fig.5, caption=Results of ablation experiments, figureFileSmall=Js3oUoJ99Vp8xypPPndTog==, figureFileBig=GIIWJuSVxcYyJrzkm8AIEA==, tableContent=null), ArticleFig(id=1217860115932299912, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, label=图5, caption=消融实验结果, figureFileSmall=Js3oUoJ99Vp8xypPPndTog==, figureFileBig=GIIWJuSVxcYyJrzkm8AIEA==, tableContent=null), ArticleFig(id=1217860116028768915, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, label=Table 1, caption=

Acc, Sen, Spe, AUC of different algorithms

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 DRIVE CHASEBD1
Acc Sen Spe AUC Acc Sen Spe AUC
UNet 0.953 4 0.676 1 0.987 5 0.980 6 0.976 5 0.707 8 0.989 5 0.977 5
UNet++ 0.958 5 0.727 7 0.986 6 0.983 0 0.978 1 0.725 9 0.990 1 0.984 0
TransUNet 0.957 1 0.741 9 0.983 2 0.980 6 0.977 4 0.824 2 0.984 8 0.983 3
Swin-UNet 0.957 2 0.720 9 0.986 9 0.981 6 0.975 3 0.680 9 0.989 8 0.971 7
AttenTransUnet 0.960 3 0.757 7 0.984 9 0.985 0 0.979 4 0.822 1 0.987 0 0.987 3
), ArticleFig(id=1217860116167180960, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, label=表1, caption=

不同算法的Acc、Sen、Spe、AUC

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 DRIVE CHASEBD1
Acc Sen Spe AUC Acc Sen Spe AUC
UNet 0.953 4 0.676 1 0.987 5 0.980 6 0.976 5 0.707 8 0.989 5 0.977 5
UNet++ 0.958 5 0.727 7 0.986 6 0.983 0 0.978 1 0.725 9 0.990 1 0.984 0
TransUNet 0.957 1 0.741 9 0.983 2 0.980 6 0.977 4 0.824 2 0.984 8 0.983 3
Swin-UNet 0.957 2 0.720 9 0.986 9 0.981 6 0.975 3 0.680 9 0.989 8 0.971 7
AttenTransUnet 0.960 3 0.757 7 0.984 9 0.985 0 0.979 4 0.822 1 0.987 0 0.987 3
), ArticleFig(id=1217860116280427173, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, label=Table 2, caption=

Acc, Sen, Spe, AUC of ablation experiments

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 Acc Sen Spe AUC
模型1 0.953 4 0.676 1 0.987 5 0.980 6
模型2 0.958 5 0.738 7 0.985 2 0.982 8
模型3 0.957 3 0.738 3 0.984 1 0.982 9
模型4 0.958 0 0.744 5 0.984 0 0.982 3
模型5 0.958 6 0.728 5 0.986 3 0.983 2
模型6 0.960 3 0.764 4 0.984 1 0.985 3
), ArticleFig(id=1217860116477559479, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, label=表2, caption=

消融实验的Acc、Sen、Spe、AUC

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 Acc Sen Spe AUC
模型1 0.953 4 0.676 1 0.987 5 0.980 6
模型2 0.958 5 0.738 7 0.985 2 0.982 8
模型3 0.957 3 0.738 3 0.984 1 0.982 9
模型4 0.958 0 0.744 5 0.984 0 0.982 3
模型5 0.958 6 0.728 5 0.986 3 0.983 2
模型6 0.960 3 0.764 4 0.984 1 0.985 3
), ArticleFig(id=1217860116628554434, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=EN, label=Table 3, caption=

Acc, Sen, Spe, AUC of different loss functions

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 Acc Sen Spe AUC
L s 4 0.953 4 0.676 1 0.987 5 0.980 6
L s 4+ L s 5 0.956 8 0.727 6 0.984 8 0.980 9
L s 2+ L s 3+
L s 4+ L s 5
0.957 1 0.711 6 0.987 0 0.981 8
L s 1+ L s 2+ L s 3+
L s 4+ L s 5
0.956 3 0.708 4 0.986 6 0.980 0
L s 3+ L s 4+ L s 5 0.957 3 0.738 3 0.984 1 0.982 9
), ArticleFig(id=1217860116750189268, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1217789889345667072, language=CN, label=表3, caption=

不同损失函数的Acc、Sen、Spe、AUC

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 Acc Sen Spe AUC
L s 4 0.953 4 0.676 1 0.987 5 0.980 6
L s 4+ L s 5 0.956 8 0.727 6 0.984 8 0.980 9
L s 2+ L s 3+
L s 4+ L s 5
0.957 1 0.711 6 0.987 0 0.981 8
L s 1+ L s 2+ L s 3+
L s 4+ L s 5
0.956 3 0.708 4 0.986 6 0.980 0
L s 3+ L s 4+ L s 5 0.957 3 0.738 3 0.984 1 0.982 9
)], attaches=null, journal=Journal(id=1146119176004939786, delFlag=0, nameCn=科学技术与工程, nameEn=Science Technology and Engineering, nameHistory1=null, nameHistory2=null, issn=1671-1815, eissn=, cn=11-4688/T, coden=null, periodic=4, language=CN, oaType=是, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=UKU/O7GSka5polgCTkbIIw==, journalPrice=null, startedYear=null, abbrevIsoEn=Sci Technol Eng, journalRemark=null, publicationField=null, createdTime=null, updatedTime=1754445529766, createdBy=null, updatedBy=13701087609, firstLetterCn=S, firstLetterEn=S, subjectCode=Natural Sciences, subjectName=自然科学, subjectCodeEn=Natural Sciences, subjectNameEn=null, picCn=UKU/O7GSka5polgCTkbIIw==, picEn=5hwlULoNwcbj3xUmVi9MAQ==, jcr=null, cjcr=null, exts=[JournalExt(id=1159791870395564357, language=CN, name=科学技术与工程, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=http://www.stae.com.cn/jsygc/home, createdTime=1754445529793, updatedTime=1754445529793, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=http://www.stae.com.cn/jsygc/site/menus/20090429150146001, submissionAuthorUrl=http://www.stae.com.cn/jsygc/author/login, submissionEditorUrl=http://www.stae.com.cn/jsygc/editor/login, submissionReviewUrl=http://www.stae.com.cn/jsygc/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1159791870441701702, language=EN, name=Science Technology and Engineering, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=http://www.stae.com.cn/jsygc/home, createdTime=1754445529804, updatedTime=1754445529804, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.stae.com.cn/jsygc/author/login, submissionEditorUrl=http://www.stae.com.cn/jsygc/editor/login, submissionReviewUrl=http://www.stae.com.cn/jsygc/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1146123166801305609, websiteList=[Website(id=1148243202391400884, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146123166801305609, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/kxjsygc/CN, language=CN, createTime=1751692112777, createBy=18614031015, updateTime=1753520965431, updateBy=18614031015, name=科学技术与工程-中文站点, tplId=1146099689490845704, title=科学技术与工程, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1148622798802673703, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=articleTextType, value=kx, createTime=1751782615614, updateTime=1751782615614, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798781702180, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=banner, value=null, createTime=1751782615609, updateTime=1751782615609, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798769119267, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=j86gbwi+p0Idkyl5SzIlmQ==, createTime=1751782615606, updateTime=1751782615606, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798794285094, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1751782615612, updateTime=1751782615612, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798790090789, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1751782615611, updateTime=1751782615611, creator=18614031015, updator=18614031015)]), Website(id=1155914124811976731, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146123166801305609, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/kxjsygc/EN, language=EN, createTime=1753521003206, createBy=18614031015, updateTime=1753521003206, updateBy=18614031015, name=科学技术与工程-英文站点, tplId=1146101810881728533, title=Science Technology and Engineering, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1155914371227308235, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=articleTextType, value=kx, createTime=1753521061952, updateTime=1753521061952, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371210531016, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=banner, value=null, createTime=1753521061947, updateTime=1753521061947, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371202142407, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=j86gbwi+p0Idkyl5SzIlmQ==, createTime=1753521061945, updateTime=1753521061945, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371223113930, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1753521061950, updateTime=1753521061950, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371218919625, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1753521061949, updateTime=1753521061949, creator=18614031015, updator=18614031015)])], journalTitle=科学技术与工程, weixinUrl=null, journalUrl=null, iacademicId=null, status=0, seqNo=null, journalTitleEn=Science Technology and Engineering, journalPhotoCn=UKU/O7GSka5polgCTkbIIw==, journalPhotoEn=5hwlULoNwcbj3xUmVi9MAQ==, journalFirstLetter=S, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=null, provinceCode=null, provinceName=null, collectFlag=false), detailUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2406155, detailUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/10.12404/j.issn.1671-1815.2406155, pdfUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/PDF/10.12404/j.issn.1671-1815.2406155, pdfUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/PDF/10.12404/j.issn.1671-1815.2406155, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
基于自注意力机制和侧输出损失函数的视网膜血管分割网络
收藏切换
PDF下载
于振华 , 闫本聪 , 王迎美 *
科学技术与工程 | 论文·自动化技术、计算机技术 2025,25(21): 8993-9001
收起
收藏切换
科学技术与工程 | 论文·自动化技术、计算机技术 2025, 25(21): 8993-9001
基于自注意力机制和侧输出损失函数的视网膜血管分割网络
全屏
于振华 , 闫本聪, 王迎美*
作者信息
  • 山东理工大学数学与统计学院, 淄博 255000
  • 于振华(2000—),男,汉族,山东滨州人,硕士研究生。研究方向:医学图像处理。E-mail:

通讯作者:

* 王迎美(1987—),女,汉族,山东新泰人,博士,副教授。研究方向:医学图像处理与重建。E-mail:
Retinal Vessel Segmentation Network Based on Self-attention Mechanism and Lateral Output Loss Function
Zhen-hua YU , Ben-cong YAN, Ying-mei WANG*
Affiliations
  • School of Mathematics and Statistics, Shandong University of Technology, Zibo 255000, China
出版时间: 2025-07-28 doi: 10.12404/j.issn.1671-1815.2406155
文章导航
收藏切换

图像分割是医学图像分析中的一个基本问题,基于深度学习的典型UNet架构(UNet architecture)分割网络模型及其变式被广泛应用于视网膜血管分割之中。但是UNet网络通过局部卷积模块提取图像的特征信息,难以关联图像的全局信息,无法有效捕捉像素之间的长距离依赖关系。考虑到UNet网络模型存在的问题和视网膜血管图像的特点,在UNet跳跃连接中加入注意力模块,可以捕捉血管之间的长距离依赖关系。此外,为增强网络的分割能力,使用群归一化(group normalization,GN)代替UNet网络模型原始的批归一化(batch normalization,BN),对不同的通道选择对应的分组。为更新参数和优化网络,利用侧输出层和最后的输出层设计了交叉熵损失函数。在DRIVE数据集和CHASEDB1数据集上进行了实验,实验结果表明所提出的模型有更好的图像分割效果。

UNet  /  Swin-UNet  /  群归一化  /  侧输出层  /  视网膜血管分割

Image segmentation is a fundamental problem in medical image analysis, the typical deep learning based UNet architecture (UNet) and its variants are widely used in retinal vessel segmentation. However, the UNet network extracts feature information from images through local convolution modules, which makes the global information of the images difficult to be correlated and the long-distance dependencies between pixels difficult to be effectively captured. Considering the problems with the UNet network model and the characteristics of retinal vascular images, an attention module was added to the skip connections of UNet to capture long-distance dependencies between blood vessels. In addition, to enhance the segmentation ability of the network, the group normalization(GN) was used instead of the original batch normalization (BN) of the UNet network model, and the corresponding groups were selected for different channels. To update parameters and optimize the network, the final cross entropy loss function was designed using the side output layer and the final output layer. Experiments are implemented on the DRIVE dataset and CHASEDB1 dataset, and the experimental results show that the proposed model has better image segmentation performance.

UNet  /  Swin-UNet  /  group normalization  /  lateral output layers  /  retinal blood vessel segmentation
于振华, 闫本聪, 王迎美. 基于自注意力机制和侧输出损失函数的视网膜血管分割网络. 科学技术与工程, 2025 , 25 (21) : 8993 -9001 . DOI: 10.12404/j.issn.1671-1815.2406155
Zhen-hua YU, Ben-cong YAN, Ying-mei WANG. Retinal Vessel Segmentation Network Based on Self-attention Mechanism and Lateral Output Loss Function[J]. Science Technology and Engineering, 2025 , 25 (21) : 8993 -9001 . DOI: 10.12404/j.issn.1671-1815.2406155
眼科医学作为医学领域中的一个重要分支,在近几十年来取得了令人瞩目的进展。随着人口老龄化和生活方式的变化,眼科疾病的发病率呈上升趋势,因此对于眼科疾病的早期诊断和治疗提出了更高的要求[1]。在这一背景下,视网膜血管分割技术作为眼底图像处理的关键环节[2],成为学者们的研究热点。
眼底图像是医生获取患者眼部信息的重要工具之一,一张正常的眼底图像主要包括血管、视盘和黄斑3个部分,并且正常的眼底呈现橘红色,眼底区域呈现出明亮的微黄色。通过观察血管结构、视盘大小及形状、黄斑和出血情况等,医生可以判断眼部疾病的类型并给出对应治疗方案。然而眼底图像[3]中复杂的结构和组织,尤其是视网膜血管网络的复杂性,使得人工分析和诊断变得烦琐且容易出现主观误判。
基于视网膜血管结构的复杂性,传统方法在视网膜血管分割中表现并不是很好。近年来随着深度学习技术的发展,使用网络来解决视网膜血管分割的方法越来越多。UNet架构(UNet architecture)网络是图像分割领域的一个重要网络,该网络及其变式被广泛应用于各种图像分割任务之中。Ronneberger等[4]提出了由编码器和解码器构成的UNet网络,该网络使用局部卷积来提取图像的特征信息,并通过跳跃连接融合特征。近年来不断有学者对UNet网络进行改进[5],包括在UNet网络中加入残差结构[6]、注意力模块[7]、Inception模块[8]等。田会娟等[9]将残差结构加入UNet网络的编码器结构之中提出了SRN-UNet(squeeze and exceitation resnext UNet)网络,该网络可以激励图像的特征结构并抑制噪声,被用于低质量的虹膜图像分割之中。添加注意力模块的UNet网络结构包括在编码器结构中引入注意力机制,在解码器或跳跃连接中引入注意力机制等。郭宁等[10]将注意力门控添加到了网络的解码器,使得网络增加了对特征区域和结构的关注,减少了网络在分割图像时受图像背景的影响,在肺部图像的分割中有着较好的效果。除了添加各种模块以外,还有部分学者将UNet网络与其他网络结合构造了一些复合网络。Vaswani等[11]提出了用于文字处理的Transformer结构,该结构是一种自注意力模块,可以使得网络更好地挖掘输入信息的全局特征,一开始被用于语言处理之中[12]。Cao等[13]将Transformer结构与UNet模型相结合,提出了用于图像分割的Swin-UNet(Swin UNet architecture)网络模型,同时通过添加移动窗口解决了Transformer结构在大型图像分割上计算量过大的问题,使得Transformer结构在图像分割领域得到了很好的应用。
除了在UNet网络中添加各种模块,还有一些学者对其连接方式进行改进。Huang等[14]为弥补UNet模型各尺度信息利用不充分的情况,设计了具有全尺度跳跃链接的UNet3+模型。Xiang等[15]在编码器和解码器之间建立O形循环路径,这种路径可以多次递归提升网络的性能。
视网膜血管图像由主干血管和由主干血管延申出的众多分支血管构成,通过UNet网络模型进行视网膜血管分割时有以下问题亟待解决:一是视网膜血管分割的准确率较低,即在分割过程中网络不能很好的区分分割的主体和背景。二是对微小血管的分割效果较差,分割出的细小血管存在断裂、无法连续的现象。
分析上述亟待解决的问题,在视网膜血管的分割任务中分割主体和背景的像素数量存在巨大的差别,血管结构仅占眼底图像的7%左右,这容易引起网络对血管的关注不够,造成视网膜血管分割的准确率较低。一些细小血管的宽度仅由单像素构成,在网络分割的过程中很难识别该部分血管,并且血管的弯曲程度以及不同分支血管之间的重叠都会影响网络对血管的分割,造成细小血管分割较差。
针对上述问题,现添加可以关联全局信息的Transformer结构来捕捉血管之间的长距离依赖关系,通过局部卷积模块提取局部细节信息。为了增强视网膜血管图像的对比度,在数据预处理部分增加了原始图像的伽马矫正[16]。为进一步提高网络血管分割的准确率,使用群归一化(group normalization, GN)代替批归一化(batch normalization, BN),使网络在训练中有更好的效果,从而保证网络的分割能力。此外,充分利用解码器的侧输出层,通过侧输出层构造损失函数。为了验证本文网络的分割能力,在DRIVE数据集和CHASEDB1数据集中进行实验。
UNet网络模型由3个部分构成,分别是编码器、解码器以及跳跃连接。编码器结构共有4层,每一层都是由两步相同的操作组成,即使用大小为3的卷积核进行卷积,BN归一化以及Relu函数激活。然后进行下采样,同样进行这些操作,反复4次后网络就会得到一组只有原图1/16的特征图。
解码器首先将特征图进行上采样,每一次上采样将特征图大小扩大一倍,然后拼接编码器结构所提取的特征图,再进行下一步的卷积等操作。经过4次操作后,网络会输出与输入图像相同大小的分割图像。UNet网络通过这种编码器解码器结构能够较好地挖掘图像的细节和结构信息,但是仅仅通过卷积模块进行信息的挖掘无法很好地挖掘图像的全局特征。在视网膜血管图像中,视网膜血管网络呈现复杂的树状结构,拥有丰富的分支,每一根细小的血管都依附于主干血管,同时细小血管与背景之间的对比度较弱。在视网膜血管分割的任务中,考虑图像的全局特征尤为重要,用于视网膜血管分割的网络应该具有图像的全局信息挖掘能力。
Transformer结构首次在自然语言处理领域被提出,该结构主要由输入、多层编码器、多层解码器以及输出4部分构成。其中输入包括单词嵌入和位置编码[17]处理,输出包括输出线性层以及Softmax激活函数处理。该结构的核心在于它的编码器结构和解码器结构中的自注意力模块(self-attention)和多头注意力模块(multi-head attention)。
自注意力模块通过查询向量 Q、键向量 K和值向量 V来计算注意力,计算公式为
Q = X W Q K = X W K V = X W v
X R m × D W Q , W K R D × d k W V R D × d v
式中: X为输入的特征向量矩阵; D为每个特征向量的维度; d k Q K中每个单词嵌入的维度; d v V中每个单词嵌入的维度;m为特征向量的数量。该结构通过将输入的 X传递进一个线性层进行处理计算出 Q K V,即式(1)的过程。然后通过得到的 Q K V进行注意力的计算,公式为
A t t e n t i o n ( Q , K , V ) = S o f t m a x Q K T d k V
式(3)中:
Q R m × d k K R m × d k V R m × d v
自注意力模块能够关联语句中单词之间的长距离关系。多头注意力模块是对自注意力模块的进一步拓展,该模块使用了多个式(1)中的权重矩阵 W Q W K W V来计算多个 W i,然后使用这些分段后的 W i计算 Q K V,然后通过式(3)计算注意力,最后将这些计算后的注意力拼接起来,计算公式为
M u l t i H e a d A t t e n t i o n ( Q , K , V ) =     C o n c a t ( h e a d 1 , h e a d 2 , , h e a d n ) W p
式(5)中:head1,head2,…,headn为使用不同的 W i计算出的Attention, n为使用权重矩阵的数量;Concat为将多个headi进行拼接; W p为一个矩阵,该矩阵的作用是将结果恢复到原本的维度。
Transformer模型被广泛地应用在自然语言处理任务之中,Dosovitskiy等[11]将该模型应用于图像识别任务中。在该模型最初并不能广泛应用于图像处理任务中,由于在Transformer的位置编码中,将图像进行编码形成的向量过长,计算量过大。Swin-UNet模型[13]改进了Transformer,通过移动窗口[18]进行自注意力模块计算,大大减少了计算量,使得该模型可以在大型图片上进行运算,提高了Transformer结构的实用性。
Swin-UNet网络采用了UNet的U形结构,由编码器、解码器及其之间的跳跃连接构成。该网络的基础模块是Swin Transformer Block,利用该模块进行特征的提取,通过移动窗口进行Transformer处理,即在不同的窗口中计算注意力进一步提取图像的特征,该方法使得Transformer结构可以在较大的图片中使用,捕捉图片中的全局信息。
此外,该网络使用Patch Merging模块和Patch Expending模块进行下采样、上采样和通道数增减操作。每进行一次Patch Merging操作,特征块的大小将变为原来的一半,通道数增加一倍,Patch Expending则将特征块的大小翻倍,通道数减半。Skip Connection则是将编码器和解码器对应层次的特征进行拼接。Swin-Unet网络完全使用Transformer结构提取图像的特征,然而该结构无法像卷积神经网络(convolutional neural network, CNN)结构一样捕捉图像局部细节特征,并且在较小的数据集上并没有很好的效果。
提出一种基于Transformer和UNet的网络结构,命名为AttenTransUnet,该网络结构既可以保证局部卷积对图像细节特征的提取能力,又可以兼顾Transformer模块联系全局的能力,网络结构如图1所示。首先对视网膜图像的绿色通道图像进行伽马矫正实现图像增强,然后将增强后的图像作为网络输入。该网络结构使用UNet网络的编码器、解码器及跳跃连接结构。编码器部分包括4层,每一层包括两步卷积核大小为3的卷积操作,GN归一化和Relu激活。然后经过下采样后进入下一层,重复操作4次后进入解码器,解码器包括卷积核大小为3的卷积、GN归一化、Relu激活和上采样操作。
考虑到UNet网络在跳跃连接中将不同层次的特征图直接拼接时,可能会由于不同特征的语义差导致网络性能下降,同时单纯的CNN结构无法关联图像的全局信息,为减缓上述问题以及提高网络的全局信息关联能力,设计了简化Swin-Unet的U形结构作为编码器和解码器特征图之间的跳跃连接,命名为Attention Block。即在特征图被卷积模块处理之后,会进入AttentionBlock模块,该操作会捕捉特征图中像素之间的全局联系,并且进行处理后的特征图再进行拼接会消除由直接拼接带来的影响,从而提高网络的分割能力。
所提网络结构充分利用侧输出层特征信息来构造损失函数,如图1右侧所示。通过上采样将解码器中每一层的特征图与输入图像大小保持相同,然后计算交叉熵损失函数。
伽马矫正通过改变像素值的分布来实现对图像对比度增强,公式表示为
O u t p u t = I n p u t M a x γ M a x
式(6)中:Input为输入图像像素的灰度值;Max为输入图像像素灰度值的最大值;Output为对应于输入图像位置的输出像素灰度值; γ为伽马值,通过调整 γ大小可以调整图像亮暗区域的对比度。当 γ 1时拉伸亮区的灰度范围起到增强亮区图像的对比度的作用;当 γ 1时拉伸暗区图像的灰度范围,使图像暗区对比度增加。
针对视网膜血管图像,实验数据预处理使用 γ = 1.5的伽马矫正进行了图像增强。通过实验证明,对绿色通道使用伽马矫正相比于对其他通道或是三通道同时进行伽马矫正有着更好的效果,因此使用绿色通道伽马矫正后的图像作为网络输入。
原始UNet网络直接将编码器特征图与解码器特征图进行特征融合,由于编码器特征图与解码器的特征图包含着不同层级的信息,这种特征图之间的信息差会导致网络分割能力下降。针对这种问题,Zhou等[19]提出了UNet++模型,该模型通过在跳跃连接中添加卷积块的方式对UNet网络进行了改进,减弱了直接跳跃连接带来的影响。
考虑到直接拼接的问题和局部卷积无法很好地提取图像全局信息的缺点,设计了注意力跳跃连接模块,结构如图2所示。该模块是一个简化的Swin-UNet的U形结构,包括编码器和解码器两部分组成,编码器由3层移动窗口自注意力模块(Attention Block)构成,该模块的原理是将整幅图片通过移动窗口分成不同的小的特征块,然后再对特征块进行Transformer处理,计算他们的注意力权重。每次进行完该处理后会通过块合并(patch merging)模块进行下采样以及提升特征层的数量。与编码器类似,解码器也由3层Attention Block构成,并通过块拓展(patch expending)进行上采样以及降低特征层的数量。提出的AttenTransUNet编码器通过CNN结构提取图像的细节特征,将特征图传递给Attention Block跳跃连接模块,该模块使用Transformer结构对特征图进行进一步处理,可以加强特征图中像素之间的全局联系。
通过实验对比了Attention Block模块深度的影响,最后将跳跃连接模块的U形结构确定为3层,并且对比了在不同位置该模块所带来的影响,发现仅在第一层、第二层和第三层加入该模块有较好的效果,同时可以减少计算量。
为了参数优化和更好地网络训练,对原始UNet网络中的损失函数[20]进行了改进。原始UNet模型利用最后输出层与标签图像进行交叉熵函数运算,以此作为最后的损失函数,计算公式如下。
Lossfunction=CrossEntropy(side4,label)
式(7)中:side4为网络最后输出图像;label为对应图像的标签。为了充分利用侧输出信息,将解码器中每层的输出层通过上采样处理,将图像大小恢复到与输入图像相同的尺寸,得到 s i d e i , i = 1,2 , 3,4 然后计算交叉熵,公式如下。
L s 1 = C r o s s E n t r o p y ( s i d e 1 , l a b e l )
L s 2 = C r o s s E n t r o p y ( s i d e 2 , l a b e l )
L s 3 = C r o s s E n t r o p y ( s i d e 3 , l a b e l )
L s 4 = C r o s s E n t r o p y ( s i d e 4 , l a b e l )
L s 5 = C r o s s E n t r o p y [ ( s i d e 1 + s i d e 2 + s i d e 3 + s i d e 4 ) / 4 , l a b e l ]
通过这些侧输出层构造了以上损失函数 L s i , i = 1,2 , 3 , , 5 通过消融实验,确定损失函数为
L o s s f u n c t i o n = L s 3 + L s 4 + L s 5
为了验证网络分割的有效性,使用DRIVE数据集[21]和CHASEDB1数据集[22]进行实验。DRIVE数据集(digital retinal images for vessel extraction)是研究视网膜病变的数据集,该数据集的图像来自荷兰的一个糖尿病性视网膜病变[21]的筛查项目,筛查人群是400名25~90岁的糖尿病受试者,并在图片中随机选取了40张图片。该数据集中的图片大小为584×565,在实验中使用其中的20张作为训练图像,另外的20张作为测试图片。CHASEDB1数据集(combined healthy and diabetic retinopathy database 1)是一个用于眼底图像分析的公开数据集,该数据集中的图像是由14名健康人和14名糖尿病性视网膜患者中采集而来,该数据集包含28幅视网膜图像,每张图像的大小为999×960,在实验中使用20张图片进行训练,8张图片进行测试。
使用伽马矫正进行了图像增强预处理,从而增强视网膜图像中血管的对比度,实验显示同时对RGB三色通道进行增强效果并不明显,而只对绿色通道进行图像增强会使得血管结构更加清晰,因此图3列出了在DRIVE数据集中不同通道使用 γ = 1.5进行矫正的结果图。除了伽马矫正,也进行了随机翻转预处理扩大数据量。
模型在Windows环境下实现,使用深度学习框架PyTorch实现所提出的网络架构。实验所用电脑硬件环境为Windows 11系统,搭载装有集成显卡的Intel 酷睿i5 1135G7处理器。采用Adam[23]优化器对模型进行优化,Batch_size =1,Transformer模块依然使用层归一化[24]进行归一化,其他的归一化使用GN。总epoch数设为100,初始学习率设置为0.001,学习率随着迭代更新,每30次迭代缩小为原本的1/10。
采用以下评价参数进行各个网络视网膜血管分割能力的评估。
图像分割的准确率(accuracy,Acc)是评估图像分割算法性能的重要指标之一,它反映了算法对图像中目标物体分割的正确程度。准确率的计算通常基于像素级别的分类结果,即算法对图像中每个像素的类别判断与真实标签进行比较,计算正确分类的像素比例,公式为
A c c = T N + T P T P + F N + T N + F P
式(14)中:TP为真正例(算法正确检测的血管像素数);FN为假负例(算法未能检测到的血管像素数);TN为真负例(算法正确排除的非血管像素数);FP为假正例(算法错误地将非血管像素标记为血管的像素数)。图像分割的灵敏度(sensitivity, Sen)计算的是分割图像中分割主体即血管像素被分割正确的比例。该参数可以体现网络模型对分割主体的识别能力,即分割网络对视网膜血管的识别能力,公式为
S e n = T P T P + F N
图像分割中的特异度(specificity, Spe)是分割图像中分割背景被正确分割的比例。该参数体现的是分割网络对分割背景即非血管像素的识别能力,计算公式为
S p e = T N T N + F P
曲线下的面积(area under curve,AUC)是指受试者操作特征(receiver operating characteristics,ROC)曲线下面积的数值,ROC曲线的横轴是假阳性率,纵轴是真阳性率。AUC的范围在0~1,AUC越大,模型的分割性能就越好。当AUC=1时,表示模型可以完美地区分正例和负例,越接近于1,模型的分类能力越强。当AUC=0.5时,表示模型的分类能力和随机猜测的结果相同。
对比实验在DRIVE数据集和CHASEBD1数据集两个数据集中进行,给出了UNet模型[4]、TransUNet模型[25]、Swin-UNet模型[13]以及所提出的AttenTransUnet模型的实验效果。通过精确率、灵敏度、特异度和AUC对分割效果进行了评估,不同模型在两个数据集上的具体结果如表1所示。图4是不同网络结构的分割结果图,选择了DRIVE数据集的第1幅图和第7幅图进行展示。
图3可以看出,对绿色通道进行伽马矫正的结果有着更好的效果,即使是细小血管也有很好的对比度,在分割结果中应该更能表现出更好的效果。在图4的细节图中不难看出,在血管的连接性上,本文网络有着更好的效果,血管断联较少,并且在血管分支上,分割出了更多的细节信息。特别是从第三行和第四行图像中可以看出,本文网络模型分割出的血管与标签相近,误判的像素更少。这些结论可以通过表1中展示的各种算法的评价指标得到进一步验证,相比于同样使用注意力模块和卷积模块的TransUNet模型,本文网络模型在两个数据集上精确率分别提升了0.003 2和0.002。
为了验证本文网络模型设计的GN、伽马矫正、Attention Block跳跃连接的有效性,进行了消融实验,消融实验包括使用BN的UNet分割网络(模型1);使用GN的UNet分割网络(模型2);含有本文网络模型设置侧输出层的UNet网络(模型3);含有本文网络模型跳跃连接结构的UNet网络(模型4);使用伽马矫正的UNet网络(模型5)以及含有跳跃连接,GN和侧输出层的最终分割网络(模型6),在DRIVE数据集上进行训练和测试。图5是消融实验中不同模块的分割图,可以看出,进行伽马矫正后的分割效果比原始的UNet网络有着更好的分割效果,这说明该处理对网络效果的提升发挥了作用。同样,加入注意力模块的网络结构相比于之前的网络细节信息更多。从表2 可以看出本文网络模型的不同模块对整个网络结构都有着不同程度的提升。综上所述,本文网络结构在血管分割方面有着更好的效果。
为了显示本文网络模型设计的损失函数过程,对比了本文网络模型AttenTransUnet中加入不同层损失函数的参数表现,如表3所示,确定使得各项评价指标最高的损失函数为式(8)。损失函数的消融实验包括无侧输出损失函数( L s 4模型),加入图2最后一层侧输出的损失函数( L s 4 + L s 5模型),加入图2的全部侧输出的损失函数( L s 1 + L s 2 + L s 3 + L s 4 + L s 5模型),加入图2的后三层侧输出的损失函数( L s 2 + L s 3 + L s 4 + L s 5模型),加入图2后两层的侧输出的损失函数( L s 3 + L s 4 + L s 5模型)。
提出了基于自注意力模块和侧输出损失函数的视网膜血管分割网络结构。自注意力模块可以加强血管之间的全局联系,增强血管的连接性。通过侧输出层构建的损失函数可以提高网络的分割能力。在进行分割之前,使用了绿色通道的伽马矫正,该过程增强了血管信息的对比度,使得网络可以更好地区别血管结构,从而增强网络的分割效果。同时在训练过程中,在网络中使用GN使得网络在卷积之后有更好的归一化结果,从而提升了网络的分割效果。
  • 山东省自然科学基金面上项目(ZR2022MA027)
参考文献 引证文献
排序方式:
[1]
Soleimani M, Cheraqpour K, Koganti R, et al. Cellular senescence and ophthalmic diseases: narrative review[J]. Graefe's Archive for Clinical and Experimental Ophthalmology, 2023, 261(11): 3067-3082.
[2]
Yu L, Qin Z, Zhuang T, et al. A framework for hierarchical division of retinal vascular networks[J]. Neurocomputing, 2020, 392: 221-232.
[3]
Qin Q, Chen Y. A review of retinal vessel segmentation for fundus image analysis[J]. Engineering Applications of Artificial Intelligence, 2024, 128: 107454.
[4]
Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation[C]// International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin:Springer International Publishing, 2015: 234-241.
[5]
汪南洋, 沈疆海. 基于MSHAM-UNet的岩心孔洞图像分割方法[J]. 科学技术与工程, 2024, 24(24): 10362-10369.
Wang Nanyang, Shen Jianghai. Image segmentation method of rock core hole based on MSHAM-UNet[J]. Science Technology and Engineering, 2024, 24(24): 10362-10369.
[6]
Zhang K, Sun M, Han T X, et al. Residual networks of residual networks: multilevel residual networks[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2017, 28(6): 1303-1314.
[7]
Baozhou Z, Hofstee P, Lee J, et al. An attention module for convolutional neural networks[J]. ArXiv, 2021: 2108.08205.
[8]
Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 2818-2826.
[9]
田会娟, 翟佳豪, 柳建新, 等. 基于SRN-UNet 的低质量虹膜分割算法[J]. 光子学报, 2022, 51(2): 241-249.
Tian Huijuan, Zhai Jiahao, Liu Jianxin, et al. A low-quality iris image segmentation algorithm based on SRN-UNet[J]. Acta Photonica Sinica, 2022, 51(2): 241-249.
[10]
郭宁, 柏正尧. 注意力机制下密集空洞卷积的肺部图像分割[J]. 中国图象图形学报, 2021: 26(9): 2146-2155.
Guo Ning, Bai Zhengyao. The integration of attention mechanism and dense atrous convolution for lung image segmentation[J]. Journal of Image and Graphics, 2021, 26(9): 2146-2155.
[11]
Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. ArXiv Preprint ArXiv, 2017: 1706.03762.
[12]
汤翔中, 高丙朋. 融合注意力空洞卷积和Transformer的矿石图像分割[J]. 科学技术与工程, 2023, 23(16): 6974-6982.
Tang Xiangzhong, Gao Bingpeng. Ore image segmentation based on attention hole convolution and Transformer[J]. Science Technology and Engineering, 2023, 23(16): 6974-6982.
[13]
Cao H, Wang Y, Chen J, et al. Swin-UNet: UNet-like pure Transformer for medical image segmentation[C]// European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 205-218.
[14]
Huang H, Lin L, Tong R, et al. UNet 3+: a full-scale connected UNet for medical image segmentation[C]// ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Barcelona: IEEE, 2020: 1055-1059.
[15]
Xiang T, Zhang C, Liu D, et al. BiO-Net: learning recurrent bi-directional connections for encoder-decoder architecture[C]// Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin:Springer, 2020: 74-84.
[16]
Ibrahim H A. Mohamed A E N, Ammar R, et al. Efficient color image enhancement using piecewise linear transformation and gamma correction[J]. Journal of Optics, 2024, 53(3): 2027-2037.
[17]
Wu K, Peng H, Chen M, et al. Rethinking and improving relative position encoding for vision transformer[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Montreal: IEEE, 2021: 10033-10041.
[18]
Song Z, Yu J, Chen Y P P, et al. Transformer tracking with cyclic shifting window attention[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans: IEEE, 2022: 8791-8800.
[19]
Zhou Z, Siddiquee M M R, Tajbakhsh N, et al. UNet++: redesigning skip connections to exploit multiscale features in image segmentation[J]. IEEE Transactions on Medical Imaging, 2019, 39(6): 1856-1867.
[20]
Barron J T. A general and adaptive robust loss function[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 4331-4339.
[21]
Staal J, Abràmoff M D, Niemeijer M, et al. Ridge-based vessel segmentation in color images of the retina[J]. IEEE Transactions on Medical Imaging, 2004, 23(4): 501-509.
[22]
Owen C G, Rudnicka A R, Mullen R, et al. Measuring retinal vessel tortuosity in 10-year-old children: validation of the computer-assisted image analysis of the retina (CAIAR) program[J]. Investigative Ophthalmology & Visual Science, 2009, 50(5): 2004-2010.
[23]
Tan T E, Wong T Y. Diabetic retinopathy: looking forward to 2030[J]. Frontiers in Endocrinology, 2023, 13: 1077669.
[24]
Ba J L, Kiros J R, Hinton G E. Layer normalization[J]. ArXiv Preprint ArXiv, 2016: 1607.06450.
[25]
Chen J, Lu Y, Yu Q, et al. TransUNet: Transformers make strong encoders for medical image segmentation[J]. ArXiv Preprint ArXiv, 2021: 2102.04306.
2025年第25卷第21期
PDF下载
223
96
引用本文
BibTeX
文章信息
doi: 10.12404/j.issn.1671-1815.2406155
  • 接收时间:2024-08-17
  • 首发时间:2026-01-13
  • 出版时间:2025-07-28
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2024-08-17
  • 修回日期:2025-04-10
基金
山东省自然科学基金面上项目(ZR2022MA027)
作者信息
    山东理工大学数学与统计学院, 淄博 255000

通讯作者:

* 王迎美(1987—),女,汉族,山东新泰人,博士,副教授。研究方向:医学图像处理与重建。E-mail:
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2406155
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数

Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)

Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏