Article(id=1251457738845401878, tenantId=1146029695717560320, journalId=1251194938327613532, issueId=1251457733048877664, articleNumber=null, orderNo=null, doi=10.14132/j.cnki.1673-5439.2025.05.008, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1727625600000, receivedDateStr=2024-09-30, revisedDate=1734710400000, revisedDateStr=2024-12-21, acceptedDate=null, acceptedDateStr=null, onlineDate=1776300375900, onlineDateStr=2026-04-16, pubDate=null, pubDateStr=null, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1776300375900, onlineIssueDateStr=2026-04-16, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1776300375900, creator=13041195026, updateTime=1776300375900, updator=13041195026, issue=Issue{id=1251457733048877664, tenantId=1146029695717560320, journalId=1251194938327613532, year='2025', volume='45', issue='5', pageStart='1', pageEnd='118', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=1, specialIssue=null, createTime=1776300374519, creator=13041195026, updateTime=1776311882923, updator=13041195026, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1251506002898661500, tenantId=1146029695717560320, journalId=1251194938327613532, issueId=1251457733048877664, language=EN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1251506002898661501, tenantId=1146029695717560320, journalId=1251194938327613532, issueId=1251457733048877664, language=CN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=66, endPage=73, ext={EN=ArticleExt(id=1251457739185140515, articleId=1251457738845401878, tenantId=1146029695717560320, journalId=1251194938327613532, language=EN, title=Few-shot image segmentation based on multi-scale feature fusion and contrastive learning, columnId=1251457734063895255, journalTitle=Journal of Nanjing University of Posts and Telecommunications(Natural Science Edition), columnName=Computer and Automation, runingTitle=null, highlight=null, articleAbstract=

In response to the scarcity of annotated medical image data and the imitations of existing models in segmenting multi-scale target images, this paper proposes a few-shot medical image segmentation method based on multi-scale feature fusion and contrastive learning. First, a sequential concatenation-based multi-scale skip connection method is introduced to replace traditional skip connections, enabling effective fusion of multi-scale feature maps from the encoder and their transmission to the corresponding decoder. Second, considering the dual-branch structure of the model, a contrastive learning module based on multi-scale features is proposed, and a loss function is designed to enhance the model's discriminative ability at the pixel level. Experiments show that our method achieves cross-domain data segmentation for medical images, mitigates performance degradation due to dataset scarcity, and improves the segmentation accuracy and generalization for different-scale targets , outperforming current mainstream few-shot medical image segmentation methods.

, correspAuthors=null, 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=Xiaofei HU, Jiayun WU, Guichun ZOU, Lingzhi WU), CN=ArticleExt(id=1251457744138613683, articleId=1251457738845401878, tenantId=1146029695717560320, journalId=1251194938327613532, language=CN, title=基于多尺度特征融合和对比学习的小样本图像分割方法, columnId=1251457734210695897, journalTitle=南京邮电大学学报(自然科学版), columnName=计算机与自动控制, runingTitle=null, highlight=null, articleAbstract=

针对医学图像标注数据稀缺以及模型在分割多尺度目标时效果不佳的问题,提出一种基于多尺度特征融合和对比学习的小样本医学图像分割方法。首先,提出一种顺序拼接的多尺度跳跃连接方法替代传统的跳跃连接,有效地融合编码器中不同尺度的特征图并传递给相应的解码器。其次,鉴于模型的双分支结构,提出一种基于多尺度特征的对比学习模块,设计对比学习损失函数,提高模型对像素级特征的可分辨能力。实验结果表明,所提方法实现了医学图像的跨域数据分割,解决了因数据集稀缺造成的分割效果差的问题,提高模型对不同尺度目标的分割准确率和泛化能力,优于当今主流的小样本医学图像分割方法。

, correspAuthors=null, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=fE98rTjmHNOg40qAmMoKVA==, magXml=esUtV3Z/+yaenUQeKGPuoQ==, pdfUrl=null, pdf=LxG4NILzQ0WJZ1peeV0oZQ==, pdfFileSize=1410689, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=2HJ1QkfpeMxifboUn2mLew==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=9hGXONpBJXHermccFR2moA==, mapNumber=null, authorCompany=null, fund=null, authors=

胡晓飞,女,博士,副教授,

, authorsList=胡晓飞, 吴佳芸, 邹贵春, 武灵芝)}, authors=[Author(id=1251457744922948574, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=huxf@njupt.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1251457745052972006, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, authorId=1251457744922948574, language=EN, stringName=Xiaofei HU, firstName=Xiaofei, middleName=null, lastName=HU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1.School of Communications and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1251457745149441003, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, authorId=1251457744922948574, language=CN, stringName=胡晓飞, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1.南京邮电大学 通信与信息工程学院,江苏 南京 210003, bio={"content":"

胡晓飞,女,博士,副教授,

"}, bioImg=null, bioContent=

胡晓飞,女,博士,副教授,

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1251457744528683969, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, xref=1., ext=[AuthorCompanyExt(id=1251457744549655491, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744528683969, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1.School of Communications and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China), AuthorCompanyExt(id=1251457744566432708, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744528683969, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1.南京邮电大学 通信与信息工程学院,江苏 南京 210003)])]), Author(id=1251457745229132785, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, 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=1251457745346573304, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, authorId=1251457745229132785, language=EN, stringName=Jiayun WU, firstName=Jiayun, middleName=null, lastName=WU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2.School of Chemistry and Life Sciences, Nanjing University of Posts and Telecommunications, Nanjing 210023, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1251457745472401408, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, authorId=1251457745229132785, language=CN, stringName=吴佳芸, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2.南京邮电大学 化学与生命科学学院,江苏 南京 210023, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1251457744667096008, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, xref=2., ext=[AuthorCompanyExt(id=1251457744696456138, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744667096008, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2.School of Chemistry and Life Sciences, Nanjing University of Posts and Telecommunications, Nanjing 210023, China), AuthorCompanyExt(id=1251457744709039052, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744667096008, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2.南京邮电大学 化学与生命科学学院,江苏 南京 210023)])]), Author(id=1251457745568870405, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, orderNo=2, 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=1251457745648562188, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, authorId=1251457745568870405, language=EN, stringName=Guichun ZOU, firstName=Guichun, middleName=null, lastName=ZOU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2.School of Chemistry and Life Sciences, Nanjing University of Posts and Telecommunications, Nanjing 210023, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1251457745740836883, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, authorId=1251457745568870405, language=CN, stringName=邹贵春, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2.南京邮电大学 化学与生命科学学院,江苏 南京 210023, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1251457744667096008, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, xref=2., ext=[AuthorCompanyExt(id=1251457744696456138, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744667096008, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2.School of Chemistry and Life Sciences, Nanjing University of Posts and Telecommunications, Nanjing 210023, China), AuthorCompanyExt(id=1251457744709039052, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744667096008, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2.南京邮电大学 化学与生命科学学院,江苏 南京 210023)])]), Author(id=1251457745820528662, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, orderNo=3, 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=1251457745916997660, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, authorId=1251457745820528662, language=EN, stringName=Lingzhi WU, firstName=Lingzhi, middleName=null, lastName=WU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3.College of Science, Nanjing University of Posts and Telecommunications,Nanjing 210023, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1251457746000883743, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, authorId=1251457745820528662, language=CN, stringName=武灵芝, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3.南京邮电大学 理学院,江苏 南京 210023, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1251457744797119441, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, xref=3., ext=[AuthorCompanyExt(id=1251457744809702356, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744797119441, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3.College of Science, Nanjing University of Posts and Telecommunications,Nanjing 210023, China), AuthorCompanyExt(id=1251457744834868182, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744797119441, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3.南京邮电大学 理学院,江苏 南京 210023)])])], keywords=[Keyword(id=1251457746156073000, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, orderNo=1, keyword=deep learning), Keyword(id=1251457746277707824, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, orderNo=2, keyword=medical image segmentation), Keyword(id=1251457746449674293, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, orderNo=3, keyword=multi-scale feature fusion), Keyword(id=1251457746529366074, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, orderNo=4, keyword=contrastive learning), Keyword(id=1251457746705526852, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, orderNo=5, keyword=few-shot learning), Keyword(id=1251457746948796492, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, orderNo=1, keyword=深度学习), Keyword(id=1251457747066237008, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, orderNo=2, keyword=医学图像分割), Keyword(id=1251457747171094614, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, orderNo=3, keyword=多尺度特征融合), Keyword(id=1251457747254980699, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, orderNo=4, keyword=对比学习), Keyword(id=1251457747334672480, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, orderNo=5, keyword=小样本学习)], refs=[Reference(id=1251457749050142885, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2014, volume=31, issue=3, pageStart=199, pageEnd=213, url=null, language=null, rfNumber=[1], rfOrder=0, authorNames=NOROUZI A, RAHIM M S M, ALTAMEEM A, journalName=IETE Technical Review, refType=null, unstructuredReference=NOROUZI A, RAHIM M S M, ALTAMEEM A, et al. Medical image segmentation methods, algorithms, and applications[J]. IETE Technical Review, 2014, 31(3):199-213., articleTitle=Medical image segmentation methods, algorithms, and applications, refAbstract=null), Reference(id=1251457749159194794, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2022, volume=16, issue=5, pageStart=1243, pageEnd=1267, url=null, language=null, rfNumber=[2], rfOrder=1, authorNames=WANG R S, LEI T, CUI R X, journalName=IET Image Processing, refType=null, unstructuredReference=WANG R S, LEI T, CUI R X, et al. Medical image segmentation using deep learning: a survey[J]. IET Image Processing, 2022, 16(5): 1243-1267., articleTitle=Medical image segmentation using deep learning: a survey, refAbstract=null), Reference(id=1251457749234692270, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2024, volume=44, issue=4, pageStart=77, pageEnd=87, url=null, language=null, rfNumber=[3], rfOrder=2, authorNames=朱王令, 金正猛, 王皓, journalName=南京邮电大学学报(自然科学版), refType=null, unstructuredReference=朱王令,金正猛,王皓.融合注意力机制和边缘预测的医学图像分割网络算法[J].南京邮电大学学报(自然科学版), 2024, 44(4): 77-87., articleTitle=融合注意力机制和边缘预测的医学图像分割网络算法, refAbstract=null), Reference(id=1251457749347938483, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2024, volume=44, issue=4, pageStart=77, pageEnd=87, url=null, language=null, rfNumber=[3], rfOrder=3, authorNames=ZHU Wangling, JIN Zhengmeng, WANG Hao, journalName=Journal of Nanjing University of Posts and Telecommunications (Natural Science Edition), refType=null, unstructuredReference=ZHU Wangling, JIN Zhengmeng, WANG Hao. A segmentation network algorithm for medical images based on attention mechanism and edge predictor[J]. Journal of Nanjing University of Posts and Telecommunications (Natural Science Edition), 2024, 44(4): 77-87. (in Chinese), articleTitle=A segmentation network algorithm for medical images based on attention mechanism and edge predictor, refAbstract=null), Reference(id=1251457749452796086, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2020, volume=64, issue=2, pageStart=20508-1, pageEnd=20508-12, url=null, language=null, rfNumber=[4], rfOrder=4, authorNames=DU G T, CAO X, LIANG J M, journalName=Journal of Imaging Science and Technology, refType=null, unstructuredReference=DU G T, CAO X, LIANG J M, et al. Medical image segmentation based on U-Net: a review[J]. Journal of Imaging Science and Technology, 2020, 64(2): 20508-1-20508-12., articleTitle=Medical image segmentation based on U-Net: a review, refAbstract=null), Reference(id=1251457749553459386, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2020, volume=53, issue=8, pageStart=5637, pageEnd=5674, url=null, language=null, rfNumber=[5], rfOrder=5, authorNames=WANG Z B, WANG E, ZHU Y, journalName=Artificial Intelligence Review, refType=null, unstructuredReference=WANG Z B, WANG E, ZHU Y. Image segmentation evaluation: a survey of methods[J]. Artificial Intelligence Review, 2020, 53(8): 5637-5674., articleTitle=Image segmentation evaluation: a survey of methods, refAbstract=null), Reference(id=1251457749649928383, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2021, volume=41, issue=2, pageStart=29, pageEnd=35, url=null, language=null, rfNumber=[6], rfOrder=6, authorNames=乔艳, 苑金辉, 谢文鑫, journalName=南京邮电大学学报(自然科学版), refType=null, unstructuredReference=乔艳,苑金辉,谢文鑫,.一种基于U-Net网络心脏MR图像左室识别的图割后处理快速算法[J].南京邮电大学学报(自然科学版), 2021, 41(2): 29-35., articleTitle=一种基于U-Net网络心脏MR图像左室识别的图割后处理快速算法, refAbstract=null), Reference(id=1251457749729620162, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2021, volume=41, issue=2, pageStart=29, pageEnd=35, url=null, language=null, rfNumber=[6], rfOrder=7, authorNames=QIAO Yan, YUAN Jinhui, XIE Wenxin, journalName=Journal of Nanjing University of Posts and Telecommunications (Natural Science Edition), refType=null, unstructuredReference=QIAO Yan, YUAN Jinhui, XIE Wenxin, et al. Fast post-processing image processing algorithm based on U-Net network for left ventricular recognition of cardiac MR images[J]. Journal of Nanjing University of Posts and Telecommunications (Natural Science Edition), 2021, 41(2): 29-35. (in Chinese), articleTitle=Fast post-processing image processing algorithm based on U-Net network for left ventricular recognition of cardiac MR images, refAbstract=null), Reference(id=1251457749809311941, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2023, volume=59, issue=2, pageStart=1, pageEnd=11, url=null, language=null, rfNumber=[7], rfOrder=8, authorNames=韦婷, 李馨蕾, 刘慧, journalName=计算机工程与应用, refType=null, unstructuredReference=韦婷,李馨蕾,刘慧.小样本困境下的图像语义分割综述[J].计算机工程与应用, 2023, 59(2): 1-11., articleTitle=小样本困境下的图像语义分割综述, refAbstract=null), Reference(id=1251457749909975241, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2023, volume=59, issue=2, pageStart=1, pageEnd=11, url=null, language=null, rfNumber=[7], rfOrder=9, authorNames=WEI Ting, LI Xinlei, LIU Hui, journalName=Computer Engineering and Applications, refType=null, unstructuredReference=WEI Ting, LI Xinlei, LIU Hui. Survey on image semantic segmentation in dilemma of few-shot[J]. Computer Engineering and Applications, 2023, 59(2): 1-11. (in Chinese), articleTitle=Survey on image semantic segmentation in dilemma of few-shot, refAbstract=null), Reference(id=1251457750006444237, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2021, volume=3, issue=6, pageStart=17, pageEnd=34, url=null, language=null, rfNumber=[8], rfOrder=10, authorNames=陈琼, 杨咏, 黄天林, journalName=数据与计算发展前沿, refType=null, unstructuredReference=陈琼,杨咏,黄天林,.小样本图像语义分割综述[J].数据与计算发展前沿, 2021, 3(6): 17-34., articleTitle=小样本图像语义分割综述, refAbstract=null), Reference(id=1251457750086136017, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2021, volume=3, issue=6, pageStart=17, pageEnd=34, url=null, language=null, rfNumber=[8], rfOrder=11, authorNames=CHEN Qiong, YANG Yong, HUANG Tianlin, journalName=Frontiers of Data & Computing, refType=null, unstructuredReference=CHEN Qiong, YANG Yong, HUANG Tianlin, et al. A survey on few-shot image semantic segmentation[J]. Frontiers of Data & Computing, 2021, 3(6): 17-34. (in Chinese), articleTitle=A survey on few-shot image semantic segmentation, refAbstract=null), Reference(id=1251457750186799318, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2023, volume=55, issue=13, pageStart=1, pageEnd=40, url=null, language=null, rfNumber=[9], rfOrder=12, authorNames=SONG Y S, WANG T, CAI P Y, journalName=ACM Computing Surveys, refType=null, unstructuredReference=SONG Y S, WANG T, CAI P Y, et al. A comprehensive survey of few-shot learning: evolution, applications, challenges, and opportunities[J]. ACM Computing Surveys, 2023, 55(13): 1-40., articleTitle=A comprehensive survey of few-shot learning: evolution, applications, challenges, and opportunities, refAbstract=null), Reference(id=1251457750270685402, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[10], rfOrder=13, authorNames=SHABAN A, BANSAL S, LIU Z, journalName=null, refType=null, unstructuredReference=SHABAN A, BANSAL S, LIU Z, et al. One-shot learning for semantic segmentation[C]//British Machine Vision Conference. 2017., articleTitle=One-shot learning for semantic segmentation, refAbstract=null), Reference(id=1251457750358765791, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2020, volume=59, issue=null, pageStart=101587, pageEnd=null, url=null, language=null, rfNumber=[11], rfOrder=14, authorNames=GUHA R A, SIDDIQUI S, PÖLSTERL S, journalName=Medical Image Analysis, refType=null, unstructuredReference=GUHA R A, SIDDIQUI S, PÖLSTERL S, et al.‘Squeeze & excite' guided few-shot segmentation of volumetric images[J]. Medical Image Analysis, 2020, 59:101587., articleTitle=‘Squeeze & excite' guided few-shot segmentation of volumetric images, refAbstract=null), Reference(id=1251457750459429091, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=7132, pageEnd=7141, url=null, language=null, rfNumber=[12], rfOrder=15, authorNames=HU J, SHEN L, SUN G, journalName=null, refType=null, unstructuredReference=HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2018: 7132-7141., articleTitle=Squeeze-and-excitation networks, refAbstract=null), Reference(id=1251457750526537958, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2023, volume=87, issue=null, pageStart=102834, pageEnd=null, url=null, language=null, rfNumber=[13], rfOrder=16, authorNames=FENG Y, WANG Y H, LI H H, journalName=Medical Image Analysis, refType=null, unstructuredReference=FENG Y, WANG Y H, LI H H, et al. Learning what and where to segment: a new perspective on medical image few-shot segmentation[J]. Medical Image Analysis, 2023, 87: 102834., articleTitle=Learning what and where to segment: a new perspective on medical image few-shot segmentation, refAbstract=null), Reference(id=1251457750614618347, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=1, pageEnd=5, url=null, language=null, rfNumber=[14], rfOrder=17, authorNames=FAN J H, BOCUS M J, HOSKING B, journalName=null, refType=null, unstructuredReference=FAN J H, BOCUS M J, HOSKING B, et al. Multi-scale feature fusion: learning better semantic segmentation for road pothole detection[C]//IEEE International Conference on Autonomous Systems (ICAS). 2021: 1-5., articleTitle=Multi-scale feature fusion: learning better semantic segmentation for road pothole detection, refAbstract=null), Reference(id=1251457750711087344, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2020, volume=39, issue=6, pageStart=1856, pageEnd=1867, url=null, language=null, rfNumber=[15], rfOrder=18, authorNames=ZHOU Z W, SIDDIQUEE M M R, TAJBAKHSH N, journalName=IEEE Transactions on Medical Imaging, refType=null, unstructuredReference=ZHOU Z W, 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, 2020, 39(6): 1856-1867., articleTitle=UNet++: redesigning skip connections to exploit multiscale features in image segmentation, refAbstract=null), Reference(id=1251457750794973431, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2022, volume=71, issue=null, pageStart=103165, pageEnd=null, url=null, language=null, rfNumber=[16], rfOrder=19, authorNames=LIU X W, YANG L, CHEN J G, journalName=Biomedical Signal Processing and Control, refType=null, unstructuredReference=LIU X W, YANG L, CHEN J G, et al. Region-to-boundary deep learning model with multi-scale feature fusion for medical image segmentation[J]. Biomedical Signal Processing and Control, 2022, 71: 103165., articleTitle=Region-to-boundary deep learning model with multi-scale feature fusion for medical image segmentation, refAbstract=null), Reference(id=1251457750862082297, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2025, volume=158, issue=null, pageStart=111072, pageEnd=null, url=null, language=null, rfNumber=[17], rfOrder=20, authorNames=ZHANG G, LI Z Y, TANG C F, journalName=Pattern Recognition, refType=null, unstructuredReference=ZHANG G, LI Z Y, TANG C F, et al. CEDNet: a cascade encoder-decoder network for dense prediction[J]. Pattern Recognition, 2025, 158: 111072., articleTitle=CEDNet: a cascade encoder-decoder network for dense prediction, refAbstract=null), Reference(id=1251457750929191165, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2023, volume=49, issue=1, pageStart=15, pageEnd=39, url=null, language=null, rfNumber=[18], rfOrder=21, authorNames=张重生, 陈杰, 李岐龙, journalName=自动化学报, refType=null, unstructuredReference=张重生,陈杰,李岐龙,.深度对比学习综述[J].自动化学报, 2023, 49(1): 15-39., articleTitle=深度对比学习综述, refAbstract=null), Reference(id=1251457751000494338, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2023, volume=49, issue=1, pageStart=15, pageEnd=39, url=null, language=null, rfNumber=[18], rfOrder=22, authorNames=ZHANG Chongsheng, CHEN Jie, LI Qilong, journalName=Acta Automatica Sinica, refType=null, unstructuredReference=ZHANG Chongsheng, CHEN Jie, LI Qilong, et al. Deep contrastive learning: a survey[J]. Acta Automatica Sinica, 2023, 49(1): 15-39. (in Chinese), articleTitle=Deep contrastive learning: a survey, refAbstract=null), Reference(id=1251457751063408902, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2020, volume=33, issue=null, pageStart=12546, pageEnd=12558, url=null, language=null, rfNumber=[19], rfOrder=23, authorNames=CHAITANYA K, ERDIL E, KARANI N, journalName=Advances in Neural Information Processing Systems, refType=null, unstructuredReference=CHAITANYA K, ERDIL E, KARANI N, et al. Contrastive learning of global and local features for medical image segmentation with limited annotations[J]. Advances in Neural Information Processing Systems, 2020, 33:12546-12558., articleTitle=Contrastive learning of global and local features for medical image segmentation with limited annotations, refAbstract=null), Reference(id=1251457751143100682, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=3023, pageEnd=3032, url=null, language=null, rfNumber=[20], rfOrder=24, authorNames=WANG X L, ZHANG R F, SHEN C H, journalName=null, refType=null, unstructuredReference=WANG X L, ZHANG R F, SHEN C H, et al. Dense contrastive learning for self-supervised visual pre-training[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2021: 3023-3032., articleTitle=Dense contrastive learning for self-supervised visual pre-training, refAbstract=null), Reference(id=1251457751222792461, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2019, volume=41, issue=12, pageStart=2933, pageEnd=2946, url=null, language=null, rfNumber=[21], rfOrder=25, authorNames=ZHUANG X H, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=ZHUANG X H. Multivariate mixture model for myocardial segmentation combining multi-source images[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(12): 2933-2946., articleTitle=Multivariate mixture model for myocardial segmentation combining multi-source images, refAbstract=null), Reference(id=1251457751310872847, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[22], rfOrder=26, authorNames=ZHUANG X, journalName=Multivariate Mixture Model for Cardiac Segmentation from Multi-Sequence MRI, refType=null, unstructuredReference=ZHUANG X. Multivariate Mixture Model for Cardiac Segmentation from Multi-Sequence MRI[M]. Cham:Springer International Publishing, 2016., articleTitle=null, refAbstract=null), Reference(id=1251457751407341843, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=1945, volume=26, issue=3, pageStart=297, pageEnd=302, url=null, language=null, rfNumber=[23], rfOrder=27, authorNames=LEE R D C, journalName=Ecology, refType=null, unstructuredReference=LEE R D C. Measures of the amount of ecologic association between species[J]. Ecology, 1945, 26(3): 297-302., articleTitle=Measures of the amount of ecologic association between species, refAbstract=null), Reference(id=1251457751503810838, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=516, pageEnd=520, url=null, language=null, rfNumber=[24], rfOrder=28, authorNames=YU J, JIANG Y, WANG Z, journalName=null, refType=null, unstructuredReference=YU J, JIANG Y, WANG Z, et al. UnitBox: an advanced object detection network[C]//Proceedings of the 24th ACM International Conference on Multimedia. 2016:516-520., articleTitle=UnitBox: an advanced object detection network, refAbstract=null), Reference(id=1251457751591891225, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[25], rfOrder=29, authorNames=赵昊宸, journalName=基于双分支网络小样本学习的脑肿瘤图像分割算法研究, refType=null, unstructuredReference=赵昊宸.基于双分支网络小样本学习的脑肿瘤图像分割算法研究[D].南京:南京邮电大学,2023., articleTitle=null, refAbstract=null), Reference(id=1251457751721914654, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[25], rfOrder=30, authorNames=ZHAO Haochen, journalName=Research on brain tumor image segmentation algorithm based on few-shot learning of dualbranch network, refType=null, unstructuredReference=ZHAO Haochen. Research on brain tumor image segmentation algorithm based on few-shot learning of dualbranch network[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2023. (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1251457751809995041, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=21381, pageEnd=21394, url=null, language=null, rfNumber=[26], rfOrder=31, authorNames=BUTOI V I, GONZALEZ ORTIZ J J, MA T Y, journalName=null, refType=null, unstructuredReference=BUTOI V I, GONZALEZ ORTIZ J J, MA T Y, et al. UniverSeg: universal medical image segmentation[C]//IEEE/CVF International Conference on Computer Vision (ICCV). 2023: 21381-21394., articleTitle=UniverSeg: universal medical image segmentation, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1251457744528683969, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, xref=1., ext=[AuthorCompanyExt(id=1251457744549655491, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744528683969, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1.School of Communications and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China), AuthorCompanyExt(id=1251457744566432708, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744528683969, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1.南京邮电大学 通信与信息工程学院,江苏 南京 210003)]), AuthorCompany(id=1251457744667096008, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, xref=2., ext=[AuthorCompanyExt(id=1251457744696456138, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744667096008, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2.School of Chemistry and Life Sciences, Nanjing University of Posts and Telecommunications, Nanjing 210023, China), AuthorCompanyExt(id=1251457744709039052, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744667096008, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2.南京邮电大学 化学与生命科学学院,江苏 南京 210023)]), AuthorCompany(id=1251457744797119441, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, xref=3., ext=[AuthorCompanyExt(id=1251457744809702356, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744797119441, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3.College of Science, Nanjing University of Posts and Telecommunications,Nanjing 210023, China), AuthorCompanyExt(id=1251457744834868182, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, companyId=1251457744797119441, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3.南京邮电大学 理学院,江苏 南京 210023)])], figs=[ArticleFig(id=1251457747473084517, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, label=null, caption=null, figureFileSmall=cWniztBDgMEC56bfzIADgg==, figureFileBig=2HJ1QkfpeMxifboUn2mLew==, tableContent=null), ArticleFig(id=1251457747552776298, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, label=图1, caption=双分支小样本医学图像分割模型, figureFileSmall=cWniztBDgMEC56bfzIADgg==, figureFileBig=2HJ1QkfpeMxifboUn2mLew==, tableContent=null), ArticleFig(id=1251457747775074418, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, label=null, caption=null, figureFileSmall=vs/uxRI6bbY5fahTttmz/A==, figureFileBig=u+ScbaDUL9cZEbWrJ3nIqw==, tableContent=null), ArticleFig(id=1251457747867349112, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, label=图2, caption=基于顺序拼接的多尺度跳跃连接模块, figureFileSmall=vs/uxRI6bbY5fahTttmz/A==, figureFileBig=u+ScbaDUL9cZEbWrJ3nIqw==, tableContent=null), ArticleFig(id=1251457747955429499, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, label=null, caption=null, figureFileSmall=CLRdXh/ODJzolUAweWb6vw==, figureFileBig=RokY33o/iCnNwnTT3QigoA==, tableContent=null), ArticleFig(id=1251457748022538367, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, label=图3, caption=多尺度特征的对比学习模块, figureFileSmall=CLRdXh/ODJzolUAweWb6vw==, figureFileBig=RokY33o/iCnNwnTT3QigoA==, tableContent=null), ArticleFig(id=1251457748098035843, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, label=null, caption=null, figureFileSmall=ahO+7Fw1tQ4Mxn/d5TewNQ==, figureFileBig=smDmLpmiD/Ser02r7S4+YQ==, tableContent=null), ArticleFig(id=1251457748165144711, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, label=图4, caption=基于不同算法的MSCMR数据分割结果可视化, figureFileSmall=ahO+7Fw1tQ4Mxn/d5TewNQ==, figureFileBig=smDmLpmiD/Ser02r7S4+YQ==, tableContent=null), ArticleFig(id=1251457748257419402, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, label=null, caption=null, figureFileSmall=1E7eKtOrlGUaNhglCcxmfA==, figureFileBig=NlKyxM51rPEwH3resowRLg==, tableContent=null), ArticleFig(id=1251457748349694094, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, label=图5, caption=不同算法的IoU箱型图, figureFileSmall=1E7eKtOrlGUaNhglCcxmfA==, figureFileBig=NlKyxM51rPEwH3resowRLg==, tableContent=null), ArticleFig(id=1251457748441968786, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, label=null, caption=null, figureFileSmall=null, figureFileBig=null, tableContent=
算法1-shot5-shot
LV-MYOLVRVLV-MYOLVRV
SE-Net(Baseline)[11]18.1674.5960.5118.8174.4363.79
RAP-Net[13]31.0066.8756.0431.5869.4857.36
DSCSE-Net[25]24.1674.4962.0924.8674.6566.17
UniverSeg[26]18.5947.0442.3622.9450.4145.21
本文27.3674.9864.1027.5275.1266.89
), ArticleFig(id=1251457748525854869, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, label=表1, caption=

不同算法Dice系数对比

, figureFileSmall=null, figureFileBig=null, tableContent=
算法1-shot5-shot
LV-MYOLVRVLV-MYOLVRV
SE-Net(Baseline)[11]18.1674.5960.5118.8174.4363.79
RAP-Net[13]31.0066.8756.0431.5869.4857.36
DSCSE-Net[25]24.1674.4962.0924.8674.6566.17
UniverSeg[26]18.5947.0442.3622.9450.4145.21
本文27.3674.9864.1027.5275.1266.89
), ArticleFig(id=1251457748626518167, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, label=null, caption=null, figureFileSmall=null, figureFileBig=null, tableContent=
Feature Maps for Skip Connection1-shot/%
d2d3d4LV-MYOLVRV
×××18.1674.5960.51
××25.6873.4261.38
××25.0368.4262.26
××22.1672.9363.95
×22.5073.1862.68
×24.6274.6961.06
×22.5970.6660.57
24.9673.6263.29
), ArticleFig(id=1251457748731375769, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, label=表2, caption=

不同尺度消融实验Dice系数对比

, figureFileSmall=null, figureFileBig=null, tableContent=
Feature Maps for Skip Connection1-shot/%
d2d3d4LV-MYOLVRV
×××18.1674.5960.51
××25.6873.4261.38
××25.0368.4262.26
××22.1672.9363.95
×22.5073.1862.68
×24.6274.6961.06
×22.5970.6660.57
24.9673.6263.29
), ArticleFig(id=1251457748840427678, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=EN, label=null, caption=null, figureFileSmall=null, figureFileBig=null, tableContent=
算法1-shot/%
SC-MSCMFCLLV-MYOLVRV
××18.1674.5960.51
×24.9673.6263.29
×23.2073.4362.16
27.3674.9864.10
), ArticleFig(id=1251457748941090977, tenantId=1146029695717560320, journalId=1251194938327613532, articleId=1251457738845401878, language=CN, label=表3, caption=

不同模块消融实验Dice系数对比

, figureFileSmall=null, figureFileBig=null, tableContent=
算法1-shot/%
SC-MSCMFCLLV-MYOLVRV
××18.1674.5960.51
×24.9673.6263.29
×23.2073.4362.16
27.3674.9864.10
)], attaches=null, journal=Journal(id=1251194371677143129, delFlag=0, nameCn=南京邮电大学学报(自然科学版), nameEn=Journal of Nanjing University of Posts and Telecommunications(Natural Science Edition), nameHistory1=null, nameHistory2=null, issn=1673-5439, eissn=null, cn=32-1772/TN, coden=null, periodic=1, language=CN, oaType=null, 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=Qa2F6AV6BkPtme75zFagKw==, journalPrice=null, startedYear=null, abbrevIsoEn=Journal of Nanjing University of Posts and Telecommunications(Natural Science Edition), journalRemark=null, publicationField=null, createdTime=1776237584277, updatedTime=1776238222056, createdBy=18614031015, updatedBy=13701087609, firstLetterCn=J, firstLetterEn=J, subjectCode=Natural Sciences, subjectName=null, subjectCodeEn=Natural Sciences, subjectNameEn=null, picCn=Qa2F6AV6BkPtme75zFagKw==, picEn=QkGGvm8WyxRIWi2yYZFUAA==, jcr=null, cjcr=null, exts=[JournalExt(id=1251197046796005699, language=CN, name=南京邮电大学学报(自然科学版), nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1776238222073, updatedTime=1776238222073, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=http://nyzr.njupt.edu.cn/ch/author/login.aspx, submissionEditorUrl=http://nyzr.njupt.edu.cn/ch/login.aspx, submissionReviewUrl=http://nyzr.njupt.edu.cn/ch/auditor/login.aspx, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1251197046842143044, language=EN, name=Journal of Nanjing University of Posts and Telecommunications(Natural Science Edition), nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1776238222084, updatedTime=1776238222084, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=http://nyzr.njupt.edu.cn/ch/author/login.aspx, submissionEditorUrl=http://nyzr.njupt.edu.cn/ch/login.aspx, submissionReviewUrl=http://nyzr.njupt.edu.cn/ch/auditor/login.aspx, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1251194938327613532, websiteList=[Website(id=1251197148423991666, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1251194938327613532, 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/njydxxxb/CN, language=CN, createTime=1776238246303, createBy=18614031015, updateTime=1776238749027, updateBy=18614031015, name=南京邮电大学学报(自然科学版)-中文, tplId=1146099689490845704, title=南京邮电大学学报(自然科学版), delFlag=0, indexPage=/home, props=[WebsiteProps(id=1251199318661743185, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=articleTextType, value=kx, createTime=1776238763728, updateTime=1776238763728, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199318636577358, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=banner, value=null, createTime=1776238763722, updateTime=1776238763722, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199318695297620, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=grayFlag, value=0, createTime=1776238763736, updateTime=1776238763736, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199318623994445, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=logo, value=https://castjournals.cast.org.cn/joweb/njydxxxb/CN/file/pic?fileId=C6FsguZTTyyR4cFBBnCSMw==, createTime=1776238763719, updateTime=1776238763719, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199318712074838, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=minRunFlag, value=0, createTime=1776238763740, updateTime=1776238763740, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199318653354576, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/njydxxxb/CN/file/pic, createTime=1776238763726, updateTime=1776238763726, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199318703686229, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=silenceFlag, value=0, createTime=1776238763738, updateTime=1776238763738, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199318644965967, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1776238763724, updateTime=1776238763724, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199318670131794, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=themeColor, value=null, createTime=1776238763730, updateTime=1776238763730, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199318682714707, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148423991666, code=themeStyle, value=null, createTime=1776238763733, updateTime=1776238763733, creator=18614031015, updator=18614031015)]), Website(id=1251197148524654966, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1251194938327613532, 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/njydxxxb/EN, language=EN, createTime=1776238246327, createBy=18614031015, updateTime=1776238738734, updateBy=18614031015, name=南京邮电大学学报(自然科学版)-英文, tplId=1146101810881728533, title=Journal of Nanjing University of Posts and Telecommunications(Natural Science Edition), delFlag=0, indexPage=/home, props=[WebsiteProps(id=1251199351561863771, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=articleTextType, value=kx, createTime=1776238771572, updateTime=1776238771572, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199351536697944, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=banner, value=null, createTime=1776238771566, updateTime=1776238771566, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199351582835294, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=grayFlag, value=0, createTime=1776238771577, updateTime=1776238771577, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199351532503639, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=logo, value=https://castjournals.cast.org.cn/joweb/njydxxxb/EN/file/pic?fileId=C6FsguZTTyyR4cFBBnCSMw==, createTime=1776238771565, updateTime=1776238771565, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199351595418208, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=minRunFlag, value=0, createTime=1776238771580, updateTime=1776238771580, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199351553475162, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/njydxxxb/EN/file/pic, createTime=1776238771570, updateTime=1776238771570, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199351591223903, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=silenceFlag, value=0, createTime=1776238771579, updateTime=1776238771579, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199351545086553, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1776238771568, updateTime=1776238771568, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199351570252380, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=themeColor, value=null, createTime=1776238771574, updateTime=1776238771574, creator=18614031015, updator=18614031015), WebsiteProps(id=1251199351578640989, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1251197148524654966, code=themeStyle, value=null, createTime=1776238771576, updateTime=1776238771576, creator=18614031015, updator=18614031015)])], journalTitle=南京邮电大学学报(自然科学版), weixinUrl=null, journalUrl=https://nyzr.njupt.edu.cn/, iacademicId=null, status=1, seqNo=null, journalTitleEn=Journal of Nanjing University of Posts and Telecommunications(Natural Science Edition), journalPhotoCn=Qa2F6AV6BkPtme75zFagKw==, journalPhotoEn=QkGGvm8WyxRIWi2yYZFUAA==, journalFirstLetter=J, 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=, provinceCode=null, provinceName=null, collectFlag=false), detailUrlCn=https://castjournals.cast.org.cn/joweb/njydxxxb/CN/10.14132/j.cnki.1673-5439.2025.05.008, detailUrlEn=https://castjournals.cast.org.cn/joweb/njydxxxb/EN/10.14132/j.cnki.1673-5439.2025.05.008, pdfUrlCn=https://castjournals.cast.org.cn/joweb/njydxxxb/CN/PDF/10.14132/j.cnki.1673-5439.2025.05.008, pdfUrlEn=https://castjournals.cast.org.cn/joweb/njydxxxb/EN/PDF/10.14132/j.cnki.1673-5439.2025.05.008, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
基于多尺度特征融合和对比学习的小样本图像分割方法
收藏切换
PDF下载
胡晓飞 1 , 吴佳芸 2 , 邹贵春 2 , 武灵芝 3
南京邮电大学学报(自然科学版) | 计算机与自动控制 2025,45(5): 66-73
收起
收藏切换
南京邮电大学学报(自然科学版) | 计算机与自动控制 2025, 45(5): 66-73
基于多尺度特征融合和对比学习的小样本图像分割方法
全屏
胡晓飞1 , 吴佳芸2, 邹贵春2, 武灵芝3
作者信息
  • 1.南京邮电大学 通信与信息工程学院,江苏 南京 210003
  • 2.南京邮电大学 化学与生命科学学院,江苏 南京 210023
  • 3.南京邮电大学 理学院,江苏 南京 210023
  • 胡晓飞,女,博士,副教授,

Few-shot image segmentation based on multi-scale feature fusion and contrastive learning
Xiaofei HU1 , Jiayun WU2, Guichun ZOU2, Lingzhi WU3
Affiliations
  • 1.School of Communications and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
  • 2.School of Chemistry and Life Sciences, Nanjing University of Posts and Telecommunications, Nanjing 210023, China
  • 3.College of Science, Nanjing University of Posts and Telecommunications,Nanjing 210023, China
doi: 10.14132/j.cnki.1673-5439.2025.05.008
文章导航
收藏切换

针对医学图像标注数据稀缺以及模型在分割多尺度目标时效果不佳的问题,提出一种基于多尺度特征融合和对比学习的小样本医学图像分割方法。首先,提出一种顺序拼接的多尺度跳跃连接方法替代传统的跳跃连接,有效地融合编码器中不同尺度的特征图并传递给相应的解码器。其次,鉴于模型的双分支结构,提出一种基于多尺度特征的对比学习模块,设计对比学习损失函数,提高模型对像素级特征的可分辨能力。实验结果表明,所提方法实现了医学图像的跨域数据分割,解决了因数据集稀缺造成的分割效果差的问题,提高模型对不同尺度目标的分割准确率和泛化能力,优于当今主流的小样本医学图像分割方法。

深度学习  /  医学图像分割  /  多尺度特征融合  /  对比学习  /  小样本学习

In response to the scarcity of annotated medical image data and the imitations of existing models in segmenting multi-scale target images, this paper proposes a few-shot medical image segmentation method based on multi-scale feature fusion and contrastive learning. First, a sequential concatenation-based multi-scale skip connection method is introduced to replace traditional skip connections, enabling effective fusion of multi-scale feature maps from the encoder and their transmission to the corresponding decoder. Second, considering the dual-branch structure of the model, a contrastive learning module based on multi-scale features is proposed, and a loss function is designed to enhance the model's discriminative ability at the pixel level. Experiments show that our method achieves cross-domain data segmentation for medical images, mitigates performance degradation due to dataset scarcity, and improves the segmentation accuracy and generalization for different-scale targets , outperforming current mainstream few-shot medical image segmentation methods.

deep learning  /  medical image segmentation  /  multi-scale feature fusion  /  contrastive learning  /  few-shot learning
胡晓飞, 吴佳芸, 邹贵春, 武灵芝. 基于多尺度特征融合和对比学习的小样本图像分割方法. 南京邮电大学学报(自然科学版), 2025 , 45 (5) : 66 -73 . DOI: 10.14132/j.cnki.1673-5439.2025.05.008
Xiaofei HU, Jiayun WU, Guichun ZOU, Lingzhi WU. Few-shot image segmentation based on multi-scale feature fusion and contrastive learning[J]. Journal of Nanjing University of Posts and Telecommunications(Natural Science Edition), 2025 , 45 (5) : 66 -73 . DOI: 10.14132/j.cnki.1673-5439.2025.05.008
医学图像处理在研究和临床实践中起着重要作用,语义分割是医学图像分析的关键技术之一[1-3]。U-Net[4]基于像素级的标签实现端到端的模型训练将卷积神经网络应用至图像分割领域,但像素级的标注十分费时,特别是在医学领域,导致数据成本变高[5-6]。同时,临床上需要针对特别目标进行训练,这样训练后的泛化能力较差[7]。小样本语义分割(FSS)[8]旨在应对这两个难题:标注数据集少和模型的泛化能力不佳。
双分支神经网络[9]能够在仅有少量样本的情况下,有效地对新目标进行分类,并且在语义分割任务中同样展现出较好的性能。Shaban等[10]从分类中受到启发,设计了经典的双分支小样本语义分割模型。SE-Net[11]不依赖预先训练的模型,使用注意力机制SE(squeeze & excite)[12]挖掘查询集和支持集的相关性。RAP-Net[13]利用医学先验知识缩小假设空间,将转化后的空间信息传递给分割分支。双分支并行的操作可以获得高级特征空间中的信息,适用于类别多、样本少的情况,但较多的参数导致在预测新类别图像时容易过拟合且效率低。
多尺度特征融合[14]应用在医学图像分割领域有助于提高分割的准确性、鲁棒性和泛化能力,对于临床诊断和治疗计划的制定具有实际意义。UNet++[15]重新设计跳跃连接,在解码器中聚合不同尺度的语义特征,实现高度灵活的特征融合方案;Liu等[16]提出的多尺度特征模块融合相邻层特征图,减少不同特征通道层之间的语义差距;CED-Net[17]早期阶段整合高级特征指导低级特征学习,增强了多尺度特征间的融合。多尺度特征融合可以结合高分辨率特征图中的细节信息和低分辨率特征图中的上下文信息,而保留更多的细节,跳跃连接通过特征的重用提高了网络的学习效率,减少过拟合的风险。
早期的对比学习(Contrastive Learning,CL)起源于自监督学习[18],像素级的对比学习能提高模型的图像分割精度。Chaitanya等[19]利用医学图像本身的结构来实现局部和全局对比学习损失;稠密对比学习[20]针对像素级的下游任务将空间域上稠密分类的方式和自监督学习融合到一起。对比学习在医学图像分割领域尤其是处理有限标注数据时,显示出巨大的潜力。
针对小样本医学图像分割模型泛化能力和分割精度不佳的问题,本文提出基于多尺度特征融合和对比学习的小样本医学图像分割方法。首先,利用顺序拼接的多尺度跳跃连接代替传统跳跃连接,将不同尺度的编码器特征图进行融合,并通过跳跃连接传递给解码器,使网络同时考虑到局部细节和全局上下文信息;其次,充分利用小样本双分支结构,提出多尺度特征的对比学习模块,通过优化对比损失函数提高模型对像素级特征的分辨能力,从而提高模型的泛化能力和分割精度。
网络架构如图1所示,双分支小样本学习网络分为:(1)条件分支(Conditioner Branch)。获取支持集图像以及提取图像特征;(2)交互模块。通过通道挤压和空间激励交互模块(Channel Squeeze &Spatial Excitation,sSE)将条件分支的特征提供给分割分支;(3)分割分支(Segmenter Branch)。输入来自交互模块的特征图,分割查询集图像;(4)多尺度跳跃连接模块。用基于顺序拼接的多尺度跳跃连接(Sequential Concatenation based Multiscale Skip Connections,SC-MSC)方法替代传统跳跃连接,融合来自编码器的不同尺度特征,并将它们传递给相应解码器;(5)对比学习模块。基于多尺度特征的对比学习模块(Multiscale Feature based Contrastive Learning Module,MFCL),目标是从多尺度跳跃连接模块中获取像素级的对比学习样本,并通过改进损失函数来增强模型的图像分割性能。
条件分支和分割分支构成双分支U-Net结构,包含8个编码器(i=1,2,3,4),两个瓶颈块BCBS和8个解码器(i=1,2,3,4),以及一个分类层CB(一层核为1的卷积层);每一个解码块、瓶颈块与编码块都由卷积层及归一化层的组合构成,编码块和解码块之间使用了有3次顺序拼接模块Cii=1,2,3)的多尺度跳跃连接;4次下采样使用最大池化,4次上采样使用反卷积恢复图像分辨率,最后通过卷积运算,再使用Sigmoid激活函数输出结果。条件分支与分割分支之间由sSE连接,该模块主要使用了卷积操作对条件分支的输出特征图进行了空间上的激励以及通道上的挤压,计算公式如下
其中,ucon为条件分支卷积块输出的特征向量,useg为分割分支卷积块输出的特征向量,P( · )函数为1维的核为1的卷积操作,σ( · )函数为激活函数,Us代表模块的输出。
SC-MSC模块主要针对医学图像分割目标的尺度大小不一和小样本模型数据少的问题。该模块允许网络在解码器部分复用编码器中的特征,减少对大量训练样本的依赖,增强对小样本数据的细节保留和特征表示能力,即使样本数量有限,网络也能通过跳跃连接捕获到不同尺度的特征。
所提出的SC-MSC模块如图2所示,对来自4个编码器输出的不同尺度的特征图e1e2e3e4,进行顺序拼接的操作。
特征图e4通过卷积层来减少通道数至与e3相当,再经过激活层和上采样,与e3在通道上拼接得到作为解码器D4的跳跃连接输入特征图d4d4包含了e3e4两个尺度特征图的信息;d4又通过相同操作与e2在通道上拼接作为解码器D3的跳跃连接输入特征图d3d3包含了e2e3e4三个尺度特征图信息;以此类推,d2包含了e1e2e3e4四个尺度特征图的信息;最后解码器D1使用了传统跳跃连接。
该模块通过逐级上采样操作逐步恢复特征图分辨率,跳跃拼接复用编码器的特征信息,使浅层解码器接收到的特征信息包含不同分辨率的语义信息,有助于网络更好地理解图像结构和上下文信息,在分割时能够更准确地识别目标的边界和细节。
MFCL的核心目标是借助正样本对比损失函数来增强模型在像素级别上的特征分辨能力,使双分支解码部分获得的模块特征表示尽可能接近,分割分支可以学习到条件分支的特征表示。
像素级别特征和图像级别特征的区别在于:图像级别特征更多关注于图像特征的整体属性,而不是单个像素,它们提取的是图像的全局属性,常用的方法是对特征图进行最大池化或平均池化等方法来获得一个值,该值用于表示整个图像的内容或类别。而本文所说的像素级别特征[20]提取是通过直接对特征中的像素点的值进行计算,这种特征提取方法涉及到对每个像素的详细分析,以预测其类别或数值。像素级图像理解通常用于语义分割、单目深度估计和图像增强等任务,目的是让计算机能够更精细地对每个像素进行区分和预测。
本文的像素级别特征来自1.2小节的SC-MSC模块的跳跃连接输出的特征图的像素点,作为深层特征表示是比较抽象的像素级别特征,这些特征虽然比原始像素级别的颜色和亮度信息更加抽象,但仍然保持了一定的空间分辨率,可以用于精细的语义分割任务。
图3所示,对特征图的像素值在通道上进行归一化计算后获得密集的向量样本集kq,将其直接作为正样本对比学习损失函数的输入,因此将其称为像素级对比学习样本。
为了使分割分支学习到条件分支的空间信息,计算kq相同位置的正样本对之间的欧氏距离获取正样本对比损失LMFCL
正样本对比损失函数公式表示为
其中,LMFCL为基于多尺度特征的对比损失函数,N为特征向量的个数,di为第i对特征向量之间的欧氏距离,具体表示为
其中,di(ab)表示第i对特征向量对(ab)之间欧氏距离,向量a来自分割分支多尺度融合特征,向量b来自条件分支的ajbj表示向量ab在第j维的值,n为向量维度。
将对比学习函数和交叉熵损失函数加权求和,得到混合损失函数
其中,LM为混合损失函数,L为交叉熵损失函数值,LMFCL为对比损失函数值,λ为权重系数,其为常值,本实验设置为0.8。
交叉熵损失函数L表示为
其中,yi为心脏MRI标签图,p(yi )为心脏MRI预测图,N为标签数量。
为了验证所提网络的有效性,实验使用两个解剖部位不同的核磁共振(MRI)图像。一是CHAOS-T2数据集的T2SPIR序列,目标为腹部器官:肝脏、左肾、右肾和脾脏。CHAOS-T2数据集分训练集与验证集,训练集和验证集又分为支持集(图像和标签成对输入)和查询集(只输入图像),支持集和查询集的数量相当。二是心脏MRI数据的b-SSFP序列,该数据集为MSCMR Seg2019(Multi-sequence Cardiac MRI Segmentation)挑战数据集[21-22],分割目标包括:左心室(LV)、右心室(RV)和左心室心肌(LV-MYO),特别地,由于左心室心肌的特殊形状,分割结果是左心室心肌外膜减去内膜(即左心室外圈)。这两个数据集都经过相同的标准化,分辨率为256*256 dpi,CHAOS-T2经过了几何变换和伽马变换的数据增强。
MSCMR数据作为测试集,以5-shot为例,测试时选择一个分割目标,为了增加查询集和支持集的相似度,所有卷同时被分为5个部分,取同一个卷(实验选择的序号为11)的5个部分的中心切片及其标签作为支持集,5例支持样本与其他卷的5个部分对应,每次输入一个支持样本切片,查询样本为对应部分的一个查询样本。
实验在Ubuntu20.04.2 64位系统上进行,采用的CPU型号为Intel(R)Xeon(R)Gold6240 CPU7700,GPU型号为NVIDIA GeForce RTX2080ti。实验一共训练50轮,每轮采样800个数据,40 000次迭代;训练时设置批量大小为5。优化器采用SGD,学习率的表达式为
其中,αlr为模型学习率,αbase=0.001为基础学习率,0.98为衰减因子,Nepoch为当前训练轮数,取值范围为[1,50]。
为了评价训练后的模型在测试集中的表现,采用医学图像分割常用的Dice[23]系数与交并比(IoU)[24]对模型进行评价。
为了直观展示算法的分割效果,5-shot设置下小样本医学网络的可视化分割结果对比如图4所示,从上到下分别是:标签(Ground Truth)、提出的模型、SE-Net[11]、DSCSE-Net[25]和RAP-Net[13],红色曲线为分割结果的轮廓。从图4可以看出,所提出的算法与其他模型相比,过分割和欠分割的情况比较少,特别是在分割LV和RV时,分割结果的轮廓都较为平滑;LV-MYO的目标形状奇特分割难度较大,即使用外膜减去内膜的方式,这几种模型的分割结果都有错分和难以成环的问题。但总的来说,所提出的模型在主观上看比其他模型轮廓更加平滑,错分的部分更少,与真实值标签更接近。
为了系统展示算法的分割效果,基于1-shot和5-shot两种实验设置,不同算法在MSCMR测试集上的平均Dice系数如表1所示,如图5所示为基于5-shot的不同算法的IoU分布情况箱型图。
表1的数据可以看出RAP-Net[13]在LV-MYO的分割上取得了最高的Dice值,但是LV和RV的分割结果不佳;而提出的模型对LV和RV的分割结果比较好。在5-shot的实验设置下,LV-MYO、LV和RV的Dice系数分别达到了27.52%、75.12%和66.89%,与SE-Net(Baseline)相比分别提高了8.7个百分点、0.7个百分点和3.1个百分点。
箱体的长度代表了数据的分布范围,箱体越长,数据分布越分散。由图5可知,本算法的IoU值显示出较高的集中性,表现在箱型图中即为较短的箱体;同时,本算法的数据最小值,即箱体的底部位置,较其他算法更高。这表明本算法在改善分割效果较差的IoU方面取得了显著成效,从而提升了整体分割质量的下限。从图5也可以看出,LV-MYO和LV的IoU有一些异常值,主要原因是一些数据的左心室目标特别小,同时有些左心室外膜不是一个完整的圆环等特殊情况。
综上所述,所提出的网络虽然在LV-MYO的分割结果上不如RAP-Net[13],但与Baseline相比,模型对3个分割目标特别是对RV的分割精度提升较高,主要是SC-MSC和MFCL模块提高了网络对目标的识别和定位能力,从而提高了对不同尺度目标的分割能力。
MSCMR三个分割目标的尺度略有不同,LV-MYO外膜大尺度类型偏多,LV(LV-MYO内膜)则大中小3种尺度的数量比较均衡,RV则是大尺度和小尺度类型数据偏多。SC-MSC代替原始跳跃连接共使用了3个不同尺度的多尺度特征融合图,为了对比这3个跳跃连接对模型的性能提升效果,在MSCMR数据集上进行消融实验,Dice系数的实验结果如表2所示,其中“×”表示使用传统跳跃连接。可以看出,3个层次的多尺度跳跃连接对不同目标的分割精度影响略有不同。
表2可以看出,浅层多尺度特征图d2对LV-MYO的分割精度提升较高,中间层d3对LV-MYO和RV,特别是对LV-MYO提升约6.87个百分点,深层多尺度特征图d4则对RV提高了约3.44个百分点。为了均衡考虑,最终选择了三层特征都使用。
总之,多尺度跳跃连接使网络在不同尺度的特征图之间实现信息融合,深层特征包含丰富的语义信息,对大尺度类型数据分割有优势;而浅层特征包含更多的细节信息,对小尺度类型数据分割有优势。MSCMR的小目标数据经过池化可能丢失细节信息,经过SC-MSC模块提升网络捕捉不同大小目标的细节的能力,从而提升模型性能。
为了对比主干网络SE-Net、SC-MSC代替原始跳跃连接和加入MFCL模块的模型分割能力,进行分割精度的消融实验,Dice系数的实验结果如表3所示。
表3可以看出,加入SC-MSC模块后,LV-MYO和RV的Dice都得到了提升,LV的效果不明显;仅使用MFCL对模型的提升没有SC-MSC明显;但是同时加入这两个模块后3个目标的Dice系数与Baseline比,平均提升了4.4个百分点左右。可见,通过结合多尺度特征和对比学习策略,使用正样本对比损失函数在训练阶段指导SC-MSC模块的参数优化,可以使网络更加准确地识别和分割图像中的目标区域。
提出了一种基于多尺度特征融合和对比学习的小样本图像分割方法,将CHAOS-T2作为训练集和验证集,MSCMR作为测试集。首先,提出了一种创新的顺序拼接多尺度跳跃连接策略,代替传统的跳跃连接方式,允许网络复用4个编码器中的特征,提升对小样本数据的细节保留和特征表示能力,提高小样本模型对MRI医学图像不同尺度目标的分割能力。接着,基于双分支架构,引入基于多尺度特征的新型对比学习模块,使用像素级正样本对比损失函数指导多尺度融合模块的参数优化,使得分割分支在多尺度融合模块的作用下能够进一步利用条件分支的相关信息优化分割。通过这些改进,模型在分割准确率上取得了显著提升,相较于现有的主流方法具有明显优势。
但是,模型依然存在待改进的地方,对难分割小目标会出现过分割和欠分割的现象。另外不可避免的是增加模块导致计算量增加,训练时间更长,因此后续会针对如何在尽量不增加计算量的前提下提高网络的分割能力进行相关工作。
参考文献 引证文献
排序方式:
[1]
NOROUZI A, RAHIM M S M, ALTAMEEM A, et al. Medical image segmentation methods, algorithms, and applications[J]. IETE Technical Review, 2014, 31(3):199-213.
[2]
WANG R S, LEI T, CUI R X, et al. Medical image segmentation using deep learning: a survey[J]. IET Image Processing, 2022, 16(5): 1243-1267.
[3]
朱王令,金正猛,王皓.融合注意力机制和边缘预测的医学图像分割网络算法[J].南京邮电大学学报(自然科学版), 2024, 44(4): 77-87.
ZHU Wangling, JIN Zhengmeng, WANG Hao. A segmentation network algorithm for medical images based on attention mechanism and edge predictor[J]. Journal of Nanjing University of Posts and Telecommunications (Natural Science Edition), 2024, 44(4): 77-87. (in Chinese)
[4]
DU G T, CAO X, LIANG J M, et al. Medical image segmentation based on U-Net: a review[J]. Journal of Imaging Science and Technology, 2020, 64(2): 20508-1-20508-12.
[5]
WANG Z B, WANG E, ZHU Y. Image segmentation evaluation: a survey of methods[J]. Artificial Intelligence Review, 2020, 53(8): 5637-5674.
[6]
乔艳,苑金辉,谢文鑫,.一种基于U-Net网络心脏MR图像左室识别的图割后处理快速算法[J].南京邮电大学学报(自然科学版), 2021, 41(2): 29-35.
QIAO Yan, YUAN Jinhui, XIE Wenxin, et al. Fast post-processing image processing algorithm based on U-Net network for left ventricular recognition of cardiac MR images[J]. Journal of Nanjing University of Posts and Telecommunications (Natural Science Edition), 2021, 41(2): 29-35. (in Chinese)
[7]
韦婷,李馨蕾,刘慧.小样本困境下的图像语义分割综述[J].计算机工程与应用, 2023, 59(2): 1-11.
WEI Ting, LI Xinlei, LIU Hui. Survey on image semantic segmentation in dilemma of few-shot[J]. Computer Engineering and Applications, 2023, 59(2): 1-11. (in Chinese)
[8]
陈琼,杨咏,黄天林,.小样本图像语义分割综述[J].数据与计算发展前沿, 2021, 3(6): 17-34.
CHEN Qiong, YANG Yong, HUANG Tianlin, et al. A survey on few-shot image semantic segmentation[J]. Frontiers of Data & Computing, 2021, 3(6): 17-34. (in Chinese)
[9]
SONG Y S, WANG T, CAI P Y, et al. A comprehensive survey of few-shot learning: evolution, applications, challenges, and opportunities[J]. ACM Computing Surveys, 2023, 55(13): 1-40.
[10]
SHABAN A, BANSAL S, LIU Z, et al. One-shot learning for semantic segmentation[C]//British Machine Vision Conference. 2017.
[11]
GUHA R A, SIDDIQUI S, PÖLSTERL S, et al.‘Squeeze & excite' guided few-shot segmentation of volumetric images[J]. Medical Image Analysis, 2020, 59:101587.
[12]
HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2018: 7132-7141.
[13]
FENG Y, WANG Y H, LI H H, et al. Learning what and where to segment: a new perspective on medical image few-shot segmentation[J]. Medical Image Analysis, 2023, 87: 102834.
[14]
FAN J H, BOCUS M J, HOSKING B, et al. Multi-scale feature fusion: learning better semantic segmentation for road pothole detection[C]//IEEE International Conference on Autonomous Systems (ICAS). 2021: 1-5.
[15]
ZHOU Z W, 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, 2020, 39(6): 1856-1867.
[16]
LIU X W, YANG L, CHEN J G, et al. Region-to-boundary deep learning model with multi-scale feature fusion for medical image segmentation[J]. Biomedical Signal Processing and Control, 2022, 71: 103165.
[17]
ZHANG G, LI Z Y, TANG C F, et al. CEDNet: a cascade encoder-decoder network for dense prediction[J]. Pattern Recognition, 2025, 158: 111072.
[18]
张重生,陈杰,李岐龙,.深度对比学习综述[J].自动化学报, 2023, 49(1): 15-39.
ZHANG Chongsheng, CHEN Jie, LI Qilong, et al. Deep contrastive learning: a survey[J]. Acta Automatica Sinica, 2023, 49(1): 15-39. (in Chinese)
[19]
CHAITANYA K, ERDIL E, KARANI N, et al. Contrastive learning of global and local features for medical image segmentation with limited annotations[J]. Advances in Neural Information Processing Systems, 2020, 33:12546-12558.
[20]
WANG X L, ZHANG R F, SHEN C H, et al. Dense contrastive learning for self-supervised visual pre-training[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2021: 3023-3032.
[21]
ZHUANG X H. Multivariate mixture model for myocardial segmentation combining multi-source images[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(12): 2933-2946.
[22]
ZHUANG X. Multivariate Mixture Model for Cardiac Segmentation from Multi-Sequence MRI[M]. Cham:Springer International Publishing, 2016.
[23]
LEE R D C. Measures of the amount of ecologic association between species[J]. Ecology, 1945, 26(3): 297-302.
[24]
YU J, JIANG Y, WANG Z, et al. UnitBox: an advanced object detection network[C]//Proceedings of the 24th ACM International Conference on Multimedia. 2016:516-520.
[25]
赵昊宸.基于双分支网络小样本学习的脑肿瘤图像分割算法研究[D].南京:南京邮电大学,2023.
ZHAO Haochen. Research on brain tumor image segmentation algorithm based on few-shot learning of dualbranch network[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2023. (in Chinese)
[26]
BUTOI V I, GONZALEZ ORTIZ J J, MA T Y, et al. UniverSeg: universal medical image segmentation[C]//IEEE/CVF International Conference on Computer Vision (ICCV). 2023: 21381-21394.
2025年第45卷第5期
PDF下载
83
40
引用本文
BibTeX
文章信息
doi: 10.14132/j.cnki.1673-5439.2025.05.008
  • 接收时间:2024-09-30
  • 首发时间:2026-04-16
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2024-09-30
  • 修回日期:2024-12-21
基金
作者信息
    1.南京邮电大学 通信与信息工程学院,江苏 南京 210003
    2.南京邮电大学 化学与生命科学学院,江苏 南京 210023
    3.南京邮电大学 理学院,江苏 南京 210023
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/njyddxxb/CN/10.14132/j.cnki.1673-5439.2025.05.008
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