Article(id=1329420525178303191, tenantId=1146029695717560320, journalId=1324759914146529478, issueId=1329420435915125316, articleNumber=null, orderNo=null, doi=10.7544/issn1000-1239.202660468, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1780588800000, receivedDateStr=2026-06-05, revisedDate=1784217600000, revisedDateStr=2026-07-17, acceptedDate=null, acceptedDateStr=null, onlineDate=1794888152640, onlineDateStr=2026-11-17, pubDate=1788192000000, pubDateStr=2026-09-01, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1794888152640, onlineIssueDateStr=2026-11-17, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1794888152640, creator=13701087609, updateTime=1794888152640, updator=13701087609, issue=Issue{id=1329420435915125316, tenantId=1146029695717560320, journalId=1324759914146529478, year='2026', volume='63', issue='9', pageStart='2183', pageEnd='2477', issueExtLink='null', onlineDate='null', pubDate='1788192000000', pubDateStr='2026-09-01', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1794888131359, creator='13701087609', updateTime=1794888131359, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext=null, issueFiles=null, downloadFileDto=null}, startPage=2338, endPage=2354, ext={EN=ArticleExt(id=1329420525400601304, articleId=1329420525178303191, tenantId=1146029695717560320, journalId=1324759914146529478, language=EN, title=PathRoute: Active Evidence Routing for Cancer Survival Prediction on Whole Slide Pathology Images, columnId=1329420436590408262, journalTitle=Journal of Computer Research and Development, columnName=Special Issue on Frontier Technologies and Applications of Knowledge-Enhanced Artificial Intelligence, runingTitle=null, highlight=null, articleAbstract=

Whole slide images (WSIs) are central to cancer diagnosis and prognosis. However, mainstream survival prediction methods directly map massive tile features into a single global representation for risk readout, failing to explicitly distinguish complex histological composition, ignoring cancer-specific clinical phenotype differences, and leaving the model’s internal reasoning process opaque. To address these issues, we propose PathRoute, an active evidence routing framework for interpretable WSI survival prediction. The framework models WSI reasoning as a two-stage directional process that first forms evidence and then reads out risk: four role-specific learnable queries collaborate in a shared latent space, respectively performing tissue-composition sensing, active evidence routing, phenotype-semantic auxiliary supervision, and risk readout; a directional attention mask further constrains forward visibility between queries, so that risk prediction is structurally realized as a downstream readout of pre-formed evidence representations, rather than an end-to-end black-box mapping. On five TCGA solid tumor cohorts under unified inputs and identical patient-level 5-fold cross-validation splits, PathRoute using WSI alone achieves higher average C-index than representative multimodal and WSI-only methods, validating the effectiveness of structured evidence organization. Further analysis shows that routing queries consistently and spontaneously concentrate attention onto tumor regions across all five cancer types without any tissue-type supervision, providing direct interpretability support for the active routing mechanism. Code available at https://github.com/ZephyrGo/pathRoute.

, authors=Qianyi Lü1, 2, Hanwen Luo1, 2, Jun Wang2, Zhen Li3, Guoxian Yu1, 2, *, authorsList=Qianyi Lü, Hanwen Luo, Jun Wang, Zhen Li, Guoxian Yu, authorCompany=null, correspAuthors=Guoxian Yu, authorNote=

born in 2002. Master candidate. His main research interests include computational pathology and machine learning

born in 2002. PhD candidate. His main research interests include causal discovery and bioinformatics

born in 1983. PhD, professor, PhD supervisor. Senior member of CCF. Her main research interests include machine learning and bioinformatics

born in 1984. PhD, professor, chief physician. Her main research interest includes new technologies and methods for the study of digestive system diseases

born in 1985. PhD, professor, PhD supervisor. Distinguished member of CCF. His main research interests include data mining and bioinformatics

, correspAuthorsNote=null, copyrightStatement=Copyright © 2026 Journal of Computer Research and Development. All rights reserved., 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, fund=null), CN=ArticleExt(id=1329420527409672939, articleId=1329420525178303191, tenantId=1146029695717560320, journalId=1324759914146529478, language=CN, title=PathRoute:基于全切片图像的主动证据路由的癌症生存预测方法, columnId=1329420436737208904, journalTitle=计算机研究与发展, columnName=“知识增强的人工智能前沿技术及应用”专题, runingTitle=null, highlight=null, articleAbstract=

病理全切片图像(whole slide image, WSI)是癌症诊断与预后判断的核心载体。当前主流的病理全切片生存预测方法将海量的区块特征直接映射为全局表示并输出预后风险,不仅未能在建模中显式区分复杂的组织学构成、忽略了不同癌种特有的临床表型差异,也导致模型的内部推理过程缺乏可解释性。 针对上述问题,提出PathRoute,一个面向可解释生存预测的主动证据路由框架。该框架将WSI推理建模为先形成证据、再读出风险的两阶段有向过程:4类角色明确的可学习查询在统一隐空间中分工协作,分别承担组织组成感知、主动证据路由、表型语义辅助监督与风险读出;方向性注意力掩码进一步约束查询间的前向可见性,使风险预测在架构上成为前序证据表示的下游读出,而非端到端的黑盒映射。 在TCGA的5个常见实体瘤癌种上,采用统一输入特征与相同患者级5折交叉验证划分,PathRoute仅使用WSI作为输入,平均生存预测C-index优于代表性多模态融合方法与纯WSI方法,验证了结构化证据组织的有效性。进一步的分析表明,路由查询在未接受任何组织类型监督的条件下,仍在5个癌种上稳定地将注意力自发集中于肿瘤区域,为主动证据路由的可解释性提供了直接支撑。代码目前已在https://github.com/ZephyrGo/pathRoute开源。

, authors=吕千一1, 2, 罗汉文1, 2, 王峻2, 李真3, 余国先1, 2, *, authorsList=吕千一, 罗汉文, 王峻, 李真, 余国先, authorCompany=null, correspAuthors=余国先, authorNote=

作者贡献声明

吕千一负责研究构思、方法学、软件实现、验证、实验实施、形式分析、可视化并撰写论文;罗汉文负责方法学,撰写并审阅论文;王峻、李真、余国先负责监督指导、撰写并审阅论文。

2002年生。硕士研究生。主要研究方向为计算病理学、机器学习

2002年生。博士研究生。主要研究领域为因果发现、生物信息学

1983年生。博士,教授,博士生导师。CCF高级会员。主要研究方向为机器学习、生物信息学

1984年生。博士,教授,主任医师。主要研究方向为消化系统疾病研究新技术与新方法

1985年生。博士,教授,博士生导师。CCF杰出会员。主要研究方向为数据挖掘、生物信息学

, correspAuthorsNote=
余国先()
, copyrightStatement=版权所有 © 2026 计算机研究与发展编辑部, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=+XYNzhgln3tz9fu5H12YzA==, magXml=U3QimCy5ohyLjhLfb62hHw==, pdfUrl=null, pdf=itw1XCCbqN79aRVEbsjaFg==, pdfFileSize=3904089, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=8lqAAIScWeWXcaf3WN/DPg==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=fRiU0fJKySvAoAPDDZ4XOw==, mapNumber=null, fund=null)}, authors=[Author(id=1329420527871046390, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=qylv@mail.sdu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420527946543865, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420527871046390, language=EN, stringName=Qianyi Lü, firstName=Qianyi, middleName=null, lastName=Lü, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1School of Software, Shandong University, Jinan 250101
2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101, bio={"content":"

born in 2002. Master candidate. His main research interests include computational pathology and machine learning

"}, bioImg=null, bioContent=

born in 2002. Master candidate. His main research interests include computational pathology and machine learning

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528022041338, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420527871046390, language=CN, stringName=吕千一, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1山东大学软件学院 济南 250101
2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101, bio={"img":"ZPvOAfPOY3lFP1xPlR7ogA==","content":"

2002年生。硕士研究生。主要研究方向为计算病理学、机器学习

"}, bioImg=ZPvOAfPOY3lFP1xPlR7ogA==, bioContent=

2002年生。硕士研究生。主要研究方向为计算病理学、机器学习

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527640359660, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=1, ext=[AuthorCompanyExt(id=1329420527648748269, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Software, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527661331182, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1山东大学软件学院 济南 250101)]), AuthorCompany(id=1329420527728440047, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=2, ext=[AuthorCompanyExt(id=1329420527741022960, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527749411569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101)])]), Author(id=1329420528080761596, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, 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=1329420528156259071, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528080761596, language=EN, stringName=Hanwen Luo, firstName=Hanwen, middleName=null, lastName=Luo, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1School of Software, Shandong University, Jinan 250101
2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101, bio={"content":"

born in 2002. PhD candidate. His main research interests include causal discovery and bioinformatics

"}, bioImg=null, bioContent=

born in 2002. PhD candidate. His main research interests include causal discovery and bioinformatics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528219173632, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528080761596, language=CN, stringName=罗汉文, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1山东大学软件学院 济南 250101
2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101, bio={"img":"iFNHyiIrK6evcBD6qUjb8A==","content":"

2002年生。博士研究生。主要研究领域为因果发现、生物信息学

"}, bioImg=iFNHyiIrK6evcBD6qUjb8A==, bioContent=

2002年生。博士研究生。主要研究领域为因果发现、生物信息学

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527640359660, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=1, ext=[AuthorCompanyExt(id=1329420527648748269, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Software, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527661331182, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1山东大学软件学院 济南 250101)]), AuthorCompany(id=1329420527728440047, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=2, ext=[AuthorCompanyExt(id=1329420527741022960, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527749411569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101)])]), Author(id=1329420528277893890, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, 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=1329420528353391364, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528277893890, language=EN, stringName=Jun Wang, firstName=Jun, middleName=null, lastName=Wang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101, bio={"content":"

born in 1983. PhD, professor, PhD supervisor. Senior member of CCF. Her main research interests include machine learning and bioinformatics

"}, bioImg=null, bioContent=

born in 1983. PhD, professor, PhD supervisor. Senior member of CCF. Her main research interests include machine learning and bioinformatics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528416305925, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528277893890, language=CN, stringName=王峻, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101, bio={"img":"lvAveDzuhsTDz+9MclokOA==","content":"

1983年生。博士,教授,博士生导师。CCF高级会员。主要研究方向为机器学习、生物信息学

"}, bioImg=lvAveDzuhsTDz+9MclokOA==, bioContent=

1983年生。博士,教授,博士生导师。CCF高级会员。主要研究方向为机器学习、生物信息学

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527728440047, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=2, ext=[AuthorCompanyExt(id=1329420527741022960, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527749411569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101)])]), Author(id=1329420528475026183, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, 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=1329420528554717961, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528475026183, language=EN, stringName=Zhen Li, firstName=Zhen, middleName=null, lastName=Li, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3Qilu Hospital of Shandong University, Jinan 250012, bio={"content":"

born in 1984. PhD, professor, chief physician. Her main research interest includes new technologies and methods for the study of digestive system diseases

"}, bioImg=null, bioContent=

born in 1984. PhD, professor, chief physician. Her main research interest includes new technologies and methods for the study of digestive system diseases

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528609243914, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528475026183, language=CN, stringName=李真, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3山东大学齐鲁医院 济南 250012, bio={"img":"lmUf+SEbTMzepB36Qqb1+Q==","content":"

1984年生。博士,教授,主任医师。主要研究方向为消化系统疾病研究新技术与新方法

"}, bioImg=lmUf+SEbTMzepB36Qqb1+Q==, bioContent=

1984年生。博士,教授,主任医师。主要研究方向为消化系统疾病研究新技术与新方法

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527803937522, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=3, ext=[AuthorCompanyExt(id=1329420527812326131, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527803937522, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3Qilu Hospital of Shandong University, Jinan 250012), AuthorCompanyExt(id=1329420527816520436, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527803937522, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3山东大学齐鲁医院 济南 250012)])]), Author(id=1329420528672158476, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, orderNo=4, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=gxyu@sdu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1329420528760238863, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528672158476, language=EN, stringName=Guoxian Yu, firstName=Guoxian, middleName=null, lastName=Yu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, *, address=1School of Software, Shandong University, Jinan 250101
2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101, bio={"content":"

born in 1985. PhD, professor, PhD supervisor. Distinguished member of CCF. His main research interests include data mining and bioinformatics

"}, bioImg=null, bioContent=

born in 1985. PhD, professor, PhD supervisor. Distinguished member of CCF. His main research interests include data mining and bioinformatics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528818959121, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528672158476, language=CN, stringName=余国先, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, *, address=1山东大学软件学院 济南 250101
2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101, bio={"img":"r1VybPh6GVnmzwI5vQ94RA==","content":"

1985年生。博士,教授,博士生导师。CCF杰出会员。主要研究方向为数据挖掘、生物信息学

"}, bioImg=r1VybPh6GVnmzwI5vQ94RA==, bioContent=

1985年生。博士,教授,博士生导师。CCF杰出会员。主要研究方向为数据挖掘、生物信息学

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527640359660, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=1, ext=[AuthorCompanyExt(id=1329420527648748269, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Software, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527661331182, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1山东大学软件学院 济南 250101)]), AuthorCompany(id=1329420527728440047, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=2, ext=[AuthorCompanyExt(id=1329420527741022960, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527749411569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101)])])], keywords=[Keyword(id=1329420528915428114, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, orderNo=1, keyword=computational pathology), Keyword(id=1329420528982536979, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, orderNo=2, keyword=whole slide image (WSI)), Keyword(id=1329420529053840148, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, orderNo=3, keyword=survival analysis), Keyword(id=1329420529116754709, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, orderNo=4, keyword=evidence routing), Keyword(id=1329420529188057878, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, orderNo=5, keyword=interpretable reasoning), Keyword(id=1329420529255166743, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, orderNo=6, keyword=pan-cancer), Keyword(id=1329420529322275608, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, orderNo=1, keyword=计算病理学), Keyword(id=1329420529410355993, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, orderNo=2, keyword=全切片图像), Keyword(id=1329420529481659162, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, orderNo=3, keyword=生存分析), Keyword(id=1329420529552962331, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, orderNo=4, keyword=证据路由), Keyword(id=1329420529624265500, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, orderNo=5, keyword=可解释推理), Keyword(id=1329420529699762973, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, orderNo=6, keyword=多癌种)], refs=[Reference(id=1329420531096466220, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=1, rfOrder=0, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen R J, Ding T, Lu M Y, et al. Towards a general-purpose foundation model for computational pathology[J]. Nature Medicine, 2024, 30(3): 850−862, articleTitle=null, refAbstract=null), Reference(id=1329420531163575085, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=1, authorNames=null, journalName=null, refType=null, unstructuredReference=MahmoodLab. UNI2-h: An updated pathology foundation model [EB/OL]. 2025 [2026-07-10]. https://huggingface.co/MahmoodLab/UNI2-h, articleTitle=null, refAbstract=null), Reference(id=1329420531230683950, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=2, authorNames=null, journalName=null, refType=null, unstructuredReference=Zimmermann E, Vorontsov E, Viret J, et al. Virchow2: Scaling self-supervised mixed magnification models in pathology [EB/OL]. 2024 [2026-07-10]. https://arxiv.org/abs/2408.00738, articleTitle=null, refAbstract=null), Reference(id=1329420531301987119, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=4, rfOrder=3, authorNames=null, journalName=null, refType=null, unstructuredReference=Lu M Y, Chen B, Williamson D F K, et al. A visual-language foundation model for computational pathology[J]. Nature Medicine, 2024, 30(3): 863−874, articleTitle=null, refAbstract=null), Reference(id=1329420531364901680, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=5, rfOrder=4, authorNames=null, journalName=null, refType=null, unstructuredReference=Ma J, Guo Z, Zhou F, et al. A generalizable pathology foundation model using a unified knowledge distillation pretraining framework[J]. Nature Biomedical Engineering, 2026, 10: 545−564, articleTitle=null, refAbstract=null), Reference(id=1329420531427816241, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=5, authorNames=null, journalName=null, refType=null, unstructuredReference=Ilse M, Tomczak J M, Welling M. Attention-based deep multiple instance learning [C] //Proc of the 35th Int Conf on Machine Learning. Cambridge, MA: PMLR, 2018: 2127−2136, articleTitle=null, refAbstract=null), Reference(id=1329420531486536498, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=7, rfOrder=6, authorNames=null, journalName=null, refType=null, unstructuredReference=Shao Z, Bian H, Chen Y, et al. TransMIL: Transformer based correlated multiple instance learning for whole slide image classification [C] //Advances in Neural Information Processing Systems. Red Hook, NY: Curran Associates, 2021, 34: 2136−2147, articleTitle=null, refAbstract=null), Reference(id=1329420531549451059, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=8, rfOrder=7, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang H, Meng Y, Zhao Y, et al. DTFD-MIL: Double-tier feature distillation multiple instance learning for histopathology whole slide image classification [C] //Proc of the IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2022: 18802−18812, articleTitle=null, refAbstract=null), Reference(id=1329420531599782708, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=9, rfOrder=8, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen R J, Lu M Y, Weng W H, et al. Multimodal co-attention transformer for survival prediction in gigapixel whole slide images [C] //Proc of the IEEE/CVF Int Conf on Computer Vision. Piscataway, NJ: IEEE, 2021: 4015−4025, articleTitle=null, refAbstract=null), Reference(id=1329420531658502965, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=10, rfOrder=9, authorNames=null, journalName=null, refType=null, unstructuredReference=Jaume G, Vaidya A, Chen R J, et al. Modeling dense multimodal interactions between biological pathways and histology for survival prediction [C] //Proc of the IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2024: 11579−11590, articleTitle=null, refAbstract=null), Reference(id=1329420531725611830, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=11, rfOrder=10, authorNames=null, journalName=null, refType=null, unstructuredReference=Bu Y, Niu Q, Li Z, et al. Cancer survival prediction by cyclic generation and multi-grained alignment [C] //Proc of the AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2026, 40(24): 19781−19789, articleTitle=null, refAbstract=null), Reference(id=1329420531784332087, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=12, rfOrder=11, authorNames=null, journalName=null, refType=null, unstructuredReference=Hou J, Zhang R, Xie Y, et al. Multimodal deep learning for cancer prognosis prediction with clinical information prompts integration[J]. npj Digital Medicine, 2026, 9: 76, articleTitle=null, refAbstract=null), Reference(id=1329420531838858040, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=13, rfOrder=12, authorNames=null, journalName=null, refType=null, unstructuredReference=Oquab M, Darcet T, Moutakanni T, et al. DINOv2: Learning robust visual features without supervision[J/OL]. Transactions on Machine Learning Research, 2024 [2026-07-16]. https://openreview.net/forum?id=a68SUt6zFt, articleTitle=null, refAbstract=null), Reference(id=1329420531901772601, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=14, rfOrder=13, authorNames=null, journalName=null, refType=null, unstructuredReference=Mobadersany P, Yousefi S, Amgad M, et al. Predicting cancer outcomes from histology and genomics using convolutional networks[J]. Proceedings of the National Academy of Sciences, 2018, 115(13): E2970−E2979, articleTitle=null, refAbstract=null), Reference(id=1329420531981464378, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=14, authorNames=null, journalName=null, refType=null, unstructuredReference=Cox D R. Regression models and life-tables[J]. Journal of the Royal Statistical Society: Series B (Methodological), 1972, 34(2): 187−202, articleTitle=null, refAbstract=null), Reference(id=1329420532044378939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=15, authorNames=null, journalName=null, refType=null, unstructuredReference=Kather J N, Heij L R, Grabsch H I, et al. Pan-cancer image-based detection of clinically actionable genetic alterations[J]. Nature Cancer, 2020, 1(8): 789−799, articleTitle=null, refAbstract=null), Reference(id=1329420532107293500, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=17, rfOrder=16, authorNames=null, journalName=null, refType=null, unstructuredReference=王梅霞, 王慧颖, 黄苗苗, 等. 多示例多标记学习综述[J]. 计算机研究与发展, 2026, 63(1): 66−89, articleTitle=null, refAbstract=null), Reference(id=1329420532170208061, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=17, rfOrder=17, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang Meixia, Wang Huiying, Huang Miaomiao, et al. Survey on multi-instance multi-label learning[J]. Journal of Computer Research and Development, 2026, 63(1): 66−89 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329420532241511230, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=18, rfOrder=18, authorNames=null, journalName=null, refType=null, unstructuredReference=Campanella G, Hanna M G, Geneslaw L, et al. Clinical-grade computational pathology using weakly supervised deep learning on whole slide images[J]. Nature Medicine, 2019, 25(8): 1301−1309, articleTitle=null, refAbstract=null), Reference(id=1329420532296037183, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=19, rfOrder=19, authorNames=null, journalName=null, refType=null, unstructuredReference=丁世飞, 孙玉婷, 梁志贞, 等. 弱监督场景下的支持向量机算法综述[J]. 计算机学报, 2024, 47(5): 987−1009, articleTitle=null, refAbstract=null), Reference(id=1329420532358951744, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=19, rfOrder=20, authorNames=null, journalName=null, refType=null, unstructuredReference=Ding Shifei, Sun Yuting, Liang Zhizhen, et al. Survey on support vector machine algorithms in weakly supervised scenarios[J]. Chinese Journal of Computers, 2024, 47(5): 987−1009 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329420532421866305, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=20, rfOrder=21, authorNames=null, journalName=null, refType=null, unstructuredReference=Lu M Y, Williamson D F K, Chen T Y, et al. Data-efficient and weakly supervised computational pathology on whole-slide images[J]. Nature Biomedical Engineering, 2021, 5(6): 555−570, articleTitle=null, refAbstract=null), Reference(id=1329420532480586562, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=21, rfOrder=22, authorNames=null, journalName=null, refType=null, unstructuredReference=Li B, Li Y, Eliceiri K W. Dual-stream multiple instance learning network for whole slide image classification with self-supervised contrastive learning [C] //Proc of the IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 14318−14328, articleTitle=null, refAbstract=null), Reference(id=1329420532547695427, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=22, rfOrder=23, authorNames=null, journalName=null, refType=null, unstructuredReference=Carion N, Massa F, Synnaeve G, et al. End-to-end object detection with transformers [G] //LNCS 12346: Proc of the European Conf on Computer Vision. Cham: Springer, 2020: 213−229, articleTitle=null, refAbstract=null), Reference(id=1329420532618998596, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=23, rfOrder=24, authorNames=null, journalName=null, refType=null, unstructuredReference=李文生, 张菁, 卓力, 等. 基于Transformer的视觉分割技术进展[J]. 计算机学报, 2024, 47(12): 2760−2782, articleTitle=null, refAbstract=null), Reference(id=1329420532673524549, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=23, rfOrder=25, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Wensheng, Zhang Jing, Zhuo Li, et al. Overview of Transformer-based visual segmentation[J]. Chinese Journal of Computers, 2024, 47(12): 2760−2782 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329420532732244806, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=24, rfOrder=26, authorNames=null, journalName=null, refType=null, unstructuredReference=Jaegle A, Gimeno F, Brock A, et al. Perceiver: General perception with iterative attention [C] //Proc of the Int Conf on Machine Learning. Cambridge, MA: PMLR, 2021: 4651−4664, articleTitle=null, refAbstract=null), Reference(id=1329420532795159367, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=25, rfOrder=27, authorNames=null, journalName=null, refType=null, unstructuredReference=Jaegle A, Borgeaud S, Alayrac J B, et al. Perceiver IO: A general architecture for structured inputs and outputs[C/OL]// Proc of the 10th Int Conf on Learning Representations. Amherst, MA: OpenReview. net, 2022 [2026-07-16]. https://openreview.net/forum?id=fILj7WpI-g, articleTitle=null, refAbstract=null), Reference(id=1329420532849685320, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=26, rfOrder=28, authorNames=null, journalName=null, refType=null, unstructuredReference=Li J, Li D, Savarese S, et al. BLIP-2: Bootstrapping language-image pre-training with frozen image encoders and large language models [C] //Proc of the Int Conf on Machine Learning. Cambridge, MA: PMLR, 2023: 19730−19742, articleTitle=null, refAbstract=null), Reference(id=1329420532912599881, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=27, rfOrder=29, authorNames=null, journalName=null, refType=null, unstructuredReference=曾诚, 葛云洁, 赵令辰, 等. 多模态视觉语言表征学习模型及其对抗样本攻防技术综述[J]. 计算机研究与发展, 2025, 62(9): 2208−2232, articleTitle=null, refAbstract=null), Reference(id=1329420532971320138, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=27, rfOrder=30, authorNames=null, journalName=null, refType=null, unstructuredReference=Zeng Cheng, Ge Yunjie, Zhao Lingchen, et al. Survey of multimodal vision-language representation learning models and their adversarial examples attack and defense techniques[J]. Journal of Computer Research and Development, 2025, 62(9): 2208−2232 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329420533034234699, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=28, rfOrder=31, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang J, Kapse S, Ma K, et al. Prompt-MIL: Boosting multi-instance learning schemes via task-specific prompt tuning [C] //Proc of the Int Conf on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer, 2023: 624−634, articleTitle=null, refAbstract=null), Reference(id=1329420533097149260, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=29, rfOrder=32, authorNames=null, journalName=null, refType=null, unstructuredReference=Selvaraju R R, Cogswell M, Das A, et al. Grad-CAM: Visual explanations from deep networks via gradient-based localization [C] //Proc of the IEEE Int Conf on Computer Vision. Piscataway, NJ: IEEE, 2017: 618−626, articleTitle=null, refAbstract=null), Reference(id=1329420533155869517, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=30, rfOrder=33, authorNames=null, journalName=null, refType=null, unstructuredReference=Hashimoto N, Fukushima D, Koga R, et al. Multi-scale domain-adversarial multiple-instance CNN for cancer subtype classification with unannotated histopathological images [C] //Proc of the IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 3852−3861, articleTitle=null, refAbstract=null), Reference(id=1329420533214589774, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=31, rfOrder=34, authorNames=null, journalName=null, refType=null, unstructuredReference=Koh P W, Nguyen T, Tang Y S, et al. Concept bottleneck models [C] //Proc of the Int Conf on Machine Learning. Cambridge, MA: PMLR, 2020: 5338−5348, articleTitle=null, refAbstract=null), Reference(id=1329420533269115727, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=32, rfOrder=35, authorNames=null, journalName=null, refType=null, unstructuredReference=Kuhn H W. The Hungarian method for the assignment problem[J]. Naval Research Logistics Quarterly, 1955, 2(1-2): 83−97, articleTitle=null, refAbstract=null), Reference(id=1329420533323641680, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=33, rfOrder=36, authorNames=null, journalName=null, refType=null, unstructuredReference=Zadeh S G, Schmid M. Bias in cross-entropy-based training of deep survival networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(9): 3126−3137, articleTitle=null, refAbstract=null), Reference(id=1329420533386556241, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=34, rfOrder=37, authorNames=null, journalName=null, refType=null, unstructuredReference=Hutter C, Zenklusen J C. The cancer genome atlas: Creating lasting value beyond its data[J]. Cell, 2018, 173(2): 283−285, articleTitle=null, refAbstract=null), Reference(id=1329420533462053714, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=35, rfOrder=38, authorNames=null, journalName=null, refType=null, unstructuredReference=Harrell F E, Califf R M, Pryor D B, et al. Evaluating the yield of medical tests[J]. JAMA, 1982, 247(18): 2543−2546, articleTitle=null, refAbstract=null), Reference(id=1329420533520773971, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=36, rfOrder=39, authorNames=null, journalName=null, refType=null, unstructuredReference=Klambauer G, Unterthiner T, Mayr A, et al. Self-normalizing neural networks [C] //Advances in Neural Information Processing Systems. Red Hook, NY: Curran Associates, 2017, 30: 971−980, articleTitle=null, refAbstract=null), Reference(id=1329420533583688532, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=37, rfOrder=40, authorNames=null, journalName=null, refType=null, unstructuredReference=Yao J, Zhu X, Jonnagaddala J, et al. Whole slide images based cancer survival prediction using attention guided deep multiple instance learning networks[J]. Medical Image Analysis, 2020, 65: 101789, articleTitle=null, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1329420527640359660, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=1, ext=[AuthorCompanyExt(id=1329420527648748269, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Software, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527661331182, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1山东大学软件学院 济南 250101)]), AuthorCompany(id=1329420527728440047, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=2, ext=[AuthorCompanyExt(id=1329420527741022960, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527749411569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101)]), AuthorCompany(id=1329420527803937522, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=3, ext=[AuthorCompanyExt(id=1329420527812326131, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527803937522, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3Qilu Hospital of Shandong University, Jinan 250012), AuthorCompanyExt(id=1329420527816520436, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527803937522, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3山东大学齐鲁医院 济南 250012)])], figs=[ArticleFig(id=1329420529833980702, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, label=Figure 1, caption=Overview of the PathRoute framework, figureFileSmall=K7MhevZgzsX6t1UAIjPp/Q==, figureFileBig=8lqAAIScWeWXcaf3WN/DPg==, tableContent=null), ArticleFig(id=1329420529896895263, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, label=图1, caption=PathRoute框架总览, figureFileSmall=K7MhevZgzsX6t1UAIjPp/Q==, figureFileBig=8lqAAIScWeWXcaf3WN/DPg==, tableContent=null), ArticleFig(id=1329420530106610464, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, label=Figure 2, caption=Schematic diagram of active evidence routing, figureFileSmall=5A24j1FX8UOlCMBUBAd5WQ==, figureFileBig=XMTsuTWvVMOQEoAteteTdw==, tableContent=null), ArticleFig(id=1329420530177913633, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, label=图2, caption=主动证据路由示意图, figureFileSmall=5A24j1FX8UOlCMBUBAd5WQ==, figureFileBig=XMTsuTWvVMOQEoAteteTdw==, tableContent=null), ArticleFig(id=1329420530261799714, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, label=Figure 3, caption=Illustration of the directional attention mask, figureFileSmall=Ry9KX4LvR0m6o/OOabLJWA==, figureFileBig=SGdF+J3IxcoFlxnhaPTVVA==, tableContent=null), ArticleFig(id=1329420530337297187, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, label=图3, caption=方向性注意力掩码示意图, figureFileSmall=Ry9KX4LvR0m6o/OOabLJWA==, figureFileBig=SGdF+J3IxcoFlxnhaPTVVA==, tableContent=null), ArticleFig(id=1329420530400211748, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, label=Figure 4, caption=Interpretability snapshot of four query types in a BRCA slide (TCGA-3C-AALI) of PathRoute, figureFileSmall=pcCA4I5ePQGXtvZWsW0tjQ==, figureFileBig=NhTBltftucVZUEMEdYCJqQ==, tableContent=null), ArticleFig(id=1329420530471514917, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, label=图4, caption=PathRoute在一例BRCA切片(TCGA-3C-AALI)上的4类查询可解释性快照, figureFileSmall=pcCA4I5ePQGXtvZWsW0tjQ==, figureFileBig=NhTBltftucVZUEMEdYCJqQ==, tableContent=null), ArticleFig(id=1329420530530235174, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, label=Table 1, caption=

C-index for Survival Prediction on Five Cancer Types

, figureFileSmall=null, figureFileBig=null, tableContent=
方法Patho.Geno.BRCASTADCOADREADHNSCBLCAOverall
SNN$ \mathrm{\checkmark } $0.634$ \pm $0.1100.643 $ \pm $0.0560.649$ \pm $0.0620.572$ \pm $0.0200.612$ \pm $0.0610.622
MLP-Omics$ \mathrm{\checkmark } $0.609$ \pm $0.1400.539$ \pm $0.0940.687$ \pm $0.0590.602$ \pm $0.0640.647$ \pm $0.0690.617
ABMIL$ \mathrm{\checkmark } $0.531$ \pm $0.0710.506$ \pm $0.0920.692$ \pm $0.1520.583$ \pm $0.0250.639$ \pm $0.0250.590
DeepMISL$ \mathrm{\checkmark } $0.585$ \pm $0.0860.580$ \pm $0.0410.684$ \pm $0.1180.554$ \pm $0.0470.522$ \pm $0.0140.585
TransMIL$ \mathrm{\checkmark } $0.575$ \pm $0.0790.507$ \pm $0.1050.673$ \pm $0.1470.624$ \pm $0.0550.643$ \pm $0.0160.604
ABMIL+Path$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.589$ \pm $0.0840.531$ \pm $0.0970.685$ \pm $0.1290.561$ \pm $0.0370.620$ \pm $0.0470.597
DeepMISL+Path$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.593$ \pm $0.0500.548$ \pm $0.0600.592$ \pm $0.1170.555$ \pm $0.0590.597$ \pm $0.0520.577
TransMIL+Path$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.578$ \pm $0.1050.577$ \pm $0.1200.709 $ \pm $0.1320.623$ \pm $0.0460.630$ \pm $0.0290.623
MCAT$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.677 $ \pm $0.0870.576$ \pm $0.0770.692$ \pm $0.0840.617$ \pm $0.0390.649 $ \pm $0.0540.642
SurvPath$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.633$ \pm $0.1230.536$ \pm $0.0880.618$ \pm $0.1610.629 $ \pm $0.0580.629$ \pm $0.0270.609
CIMA$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.664$ \pm $0.0900.592$ \pm $0.0170.620$ \pm $0.0970.626$ \pm $0.0260.595$ \pm $0.0500.619
SurvPGC$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.638$ \pm $0.0380.628$ \pm $0.0610.691$ \pm $0.1380.598$ \pm $0.0230.658$ \pm $0.0360.643
PathRoute$ \mathrm{\checkmark } $0.680$ \pm $0.0500.665$ \pm $0.0750.717$ \pm $0.1430.646$ \pm $0.0750.621$ \pm $0.0730.666
), ArticleFig(id=1329420530614121255, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, label=表1, caption=

五癌种生存预测C-index

, figureFileSmall=null, figureFileBig=null, tableContent=
方法Patho.Geno.BRCASTADCOADREADHNSCBLCAOverall
SNN$ \mathrm{\checkmark } $0.634$ \pm $0.1100.643 $ \pm $0.0560.649$ \pm $0.0620.572$ \pm $0.0200.612$ \pm $0.0610.622
MLP-Omics$ \mathrm{\checkmark } $0.609$ \pm $0.1400.539$ \pm $0.0940.687$ \pm $0.0590.602$ \pm $0.0640.647$ \pm $0.0690.617
ABMIL$ \mathrm{\checkmark } $0.531$ \pm $0.0710.506$ \pm $0.0920.692$ \pm $0.1520.583$ \pm $0.0250.639$ \pm $0.0250.590
DeepMISL$ \mathrm{\checkmark } $0.585$ \pm $0.0860.580$ \pm $0.0410.684$ \pm $0.1180.554$ \pm $0.0470.522$ \pm $0.0140.585
TransMIL$ \mathrm{\checkmark } $0.575$ \pm $0.0790.507$ \pm $0.1050.673$ \pm $0.1470.624$ \pm $0.0550.643$ \pm $0.0160.604
ABMIL+Path$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.589$ \pm $0.0840.531$ \pm $0.0970.685$ \pm $0.1290.561$ \pm $0.0370.620$ \pm $0.0470.597
DeepMISL+Path$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.593$ \pm $0.0500.548$ \pm $0.0600.592$ \pm $0.1170.555$ \pm $0.0590.597$ \pm $0.0520.577
TransMIL+Path$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.578$ \pm $0.1050.577$ \pm $0.1200.709 $ \pm $0.1320.623$ \pm $0.0460.630$ \pm $0.0290.623
MCAT$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.677 $ \pm $0.0870.576$ \pm $0.0770.692$ \pm $0.0840.617$ \pm $0.0390.649 $ \pm $0.0540.642
SurvPath$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.633$ \pm $0.1230.536$ \pm $0.0880.618$ \pm $0.1610.629 $ \pm $0.0580.629$ \pm $0.0270.609
CIMA$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.664$ \pm $0.0900.592$ \pm $0.0170.620$ \pm $0.0970.626$ \pm $0.0260.595$ \pm $0.0500.619
SurvPGC$ \mathrm{\checkmark } $$ \mathrm{\checkmark } $0.638$ \pm $0.0380.628$ \pm $0.0610.691$ \pm $0.1380.598$ \pm $0.0230.658$ \pm $0.0360.643
PathRoute$ \mathrm{\checkmark } $0.680$ \pm $0.0500.665$ \pm $0.0750.717$ \pm $0.1430.646$ \pm $0.0750.621$ \pm $0.0730.666
), ArticleFig(id=1329420530693813032, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, label=Table 2, caption=

Component and Structural Ablations

, figureFileSmall=null, figureFileBig=null, tableContent=
分类配置BRCASTADCOADREADHNSCBLCAOverall
组件消融
(关闭单个
语义组件)
完整PathRoute0.680$ \pm $0.0500.665 $ \pm $0.0750.717$ \pm $0.1430.646$ \pm $0.0750.621 $ \pm $0.0730.666
w/o Tissue Census0.666$ \pm $0.0400.668$ \pm $0.0900.686$ \pm $0.1480.619 $ \pm $0.0310.612$ \pm $0.0830.650
Random Routing0.677 $ \pm $0.0470.668$ \pm $0.0600.712 $ \pm $0.1190.610$ \pm $0.0530.624$ \pm $0.0550.658
w/o Phenotype Supervision0.658$ \pm $0.0630.653$ \pm $0.1000.711$ \pm $0.1300.603$ \pm $0.0540.609$ \pm $0.0580.647
Survival-Only0.634$ \pm $0.0600.660$ \pm $0.0660.704$ \pm $0.1570.606$ \pm $0.0530.595$ \pm $0.0980.639
结构消融
(保留语义组件、
扰动结构约束)
完整PathRoute0.680$ \pm $0.0500.665 $ \pm $0.0750.717$ \pm $0.1430.646$ \pm $0.0750.621$ \pm $0.0730.666
Block-Directional Mask0.673 $ \pm $0.0340.665 $ \pm $0.0990.718$ \pm $0.1480.607 $ \pm $0.0440.624$ \pm $0.0720.657
Fully Bidirectional Mask0.666$ \pm $0.0400.671$ \pm $0.0960.720$ \pm $0.1500.604$ \pm $0.0410.620$ \pm $0.0740.656
Reordered Roles0.673 $ \pm $0.0400.661$ \pm $0.0950.727$ \pm $0.1440.596$ \pm $0.0410.626 $ \pm $0.0770.657
w/o Tissue Stop-Gradient0.680$ \pm $0.0390.658$ \pm $0.1020.723 $ \pm $0.1380.602$ \pm $0.0450.627$ \pm $0.0640.658
), ArticleFig(id=1329420530790282025, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, label=表2, caption=

组件与结构消融

, figureFileSmall=null, figureFileBig=null, tableContent=
分类配置BRCASTADCOADREADHNSCBLCAOverall
组件消融
(关闭单个
语义组件)
完整PathRoute0.680$ \pm $0.0500.665 $ \pm $0.0750.717$ \pm $0.1430.646$ \pm $0.0750.621 $ \pm $0.0730.666
w/o Tissue Census0.666$ \pm $0.0400.668$ \pm $0.0900.686$ \pm $0.1480.619 $ \pm $0.0310.612$ \pm $0.0830.650
Random Routing0.677 $ \pm $0.0470.668$ \pm $0.0600.712 $ \pm $0.1190.610$ \pm $0.0530.624$ \pm $0.0550.658
w/o Phenotype Supervision0.658$ \pm $0.0630.653$ \pm $0.1000.711$ \pm $0.1300.603$ \pm $0.0540.609$ \pm $0.0580.647
Survival-Only0.634$ \pm $0.0600.660$ \pm $0.0660.704$ \pm $0.1570.606$ \pm $0.0530.595$ \pm $0.0980.639
结构消融
(保留语义组件、
扰动结构约束)
完整PathRoute0.680$ \pm $0.0500.665 $ \pm $0.0750.717$ \pm $0.1430.646$ \pm $0.0750.621$ \pm $0.0730.666
Block-Directional Mask0.673 $ \pm $0.0340.665 $ \pm $0.0990.718$ \pm $0.1480.607 $ \pm $0.0440.624$ \pm $0.0720.657
Fully Bidirectional Mask0.666$ \pm $0.0400.671$ \pm $0.0960.720$ \pm $0.1500.604$ \pm $0.0410.620$ \pm $0.0740.656
Reordered Roles0.673 $ \pm $0.0400.661$ \pm $0.0950.727$ \pm $0.1440.596$ \pm $0.0410.626 $ \pm $0.0770.657
w/o Tissue Stop-Gradient0.680$ \pm $0.0390.658$ \pm $0.1020.723 $ \pm $0.1380.602$ \pm $0.0450.627$ \pm $0.0640.658
), ArticleFig(id=1329420530882556714, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=EN, label=Table 3, caption=

Tile-level Attention Concentration of Active Evidence Routing

, figureFileSmall=null, figureFileBig=null, tableContent=
癌种($ n $切片)肿瘤间质坏死脂肪组织正常组织淋巴细胞
BRCA ($ n=922 $)2.04/0.950.02/0.090.05/0.140.02/0.030.02/0.080.00/0.02
STAD ($ n=342 $)1.90/0.890.27/0.120.36/0.160.46/0.240.88/0.420.48/0.23
COADREAD ($ n=292 $)2.84/0.960.00/0.010.15/0.080.00/0.020.46/0.290.00/0.03
HNSC ($ n=412 $)2.56/0.980.14/0.140.13/0.050.00/0.030.13/0.150.03/0.07
BLCA ($ n=423 $)2.69/0.900.03/0.050.38/0.250.01/0.100.05/0.060.03/0.11
), ArticleFig(id=1329420530995802923, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, language=CN, label=表3, caption=

主动证据路由的区块级注意力集中度

, figureFileSmall=null, figureFileBig=null, tableContent=
癌种($ n $切片)肿瘤间质坏死脂肪组织正常组织淋巴细胞
BRCA ($ n=922 $)2.04/0.950.02/0.090.05/0.140.02/0.030.02/0.080.00/0.02
STAD ($ n=342 $)1.90/0.890.27/0.120.36/0.160.46/0.240.88/0.420.48/0.23
COADREAD ($ n=292 $)2.84/0.960.00/0.010.15/0.080.00/0.020.46/0.290.00/0.03
HNSC ($ n=412 $)2.56/0.980.14/0.140.13/0.050.00/0.030.13/0.150.03/0.07
BLCA ($ n=423 $)2.69/0.900.03/0.050.38/0.250.01/0.100.05/0.060.03/0.11
)], attaches=null, journal=Journal(id=1324759606947315908, delFlag=0, nameCn=计算机研究与发展, nameEn=Journal of Computer Research and Development, nameHistory1=null, nameHistory2=null, issn=1000-1239, eissn=null, cn=11-1777/TP, coden=null, periodic=0, 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=k2l8zgqwstuxxevFX2qxVA==, journalPrice=null, startedYear=null, abbrevIsoEn=Journal of Computer Research and Development, journalRemark=null, publicationField=null, createdTime=1793776903141, updatedTime=1793850698612, createdBy=18614031015, updatedBy=13701087609, firstLetterCn=J, firstLetterEn=J, subjectCode=Electronic Information, subjectName=null, subjectCodeEn=Electronic Information, subjectNameEn=null, picCn=k2l8zgqwstuxxevFX2qxVA==, picEn=r6k0Ry+RD+c/VXj5QarmDw==, jcr=null, cjcr=null, exts=[JournalExt(id=1325069127710286688, language=CN, name=计算机研究与发展, 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=1793850698640, updatedTime=1793850698640, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://crad.ict.ac.cn/journalx/authorLogOn.action, submissionEditorUrl=https://crad.ict.ac.cn/journalx/editorLogOn.action, submissionReviewUrl=https://crad.ict.ac.cn/journalx/expertLogOn.action, submissionCeEditorUrl=https://crad.ict.ac.cn/journalx/editorInChiefLogOn.action, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1325069127777395553, language=EN, name=Journal of Computer Research and Development, 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=1793850698656, updatedTime=1793850698656, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://crad.ict.ac.cn/journalx/authorLogOn.action, submissionEditorUrl=https://crad.ict.ac.cn/journalx/editorLogOn.action, submissionReviewUrl=https://crad.ict.ac.cn/journalx/expertLogOn.action, submissionCeEditorUrl=https://crad.ict.ac.cn/journalx/editorInChiefLogOn.action, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1324759914146529478, websiteList=[Website(id=1325070300139892771, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1324759914146529478, 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/jsjyjyfz/CN, language=CN, createTime=1793850978169, createBy=18614031015, updateTime=1793851040259, updateBy=18614031015, name=计算机研究与发展-中文, tplId=1146099689490845704, title=计算机研究与发展, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1325070893751349360, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=articleTextType, value=kx, createTime=1793851119697, updateTime=1793851119697, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893730377837, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=banner, value=null, createTime=1793851119692, updateTime=1793851119692, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893772320883, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=grayFlag, value=0, createTime=1793851119702, updateTime=1793851119702, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893721989228, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=logo, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/file/pic?fileId=U27tDsy717YF7XSPA+Vwdw==, createTime=1793851119690, updateTime=1793851119690, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893784903797, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=minRunFlag, value=0, createTime=1793851119705, updateTime=1793851119705, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893742960751, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/file/pic, createTime=1793851119695, updateTime=1793851119695, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893780709492, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=silenceFlag, value=0, createTime=1793851119704, updateTime=1793851119704, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893738766446, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1793851119694, updateTime=1793851119694, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893755543665, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=themeColor, value=null, createTime=1793851119698, updateTime=1793851119698, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893763932274, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=themeStyle, value=null, createTime=1793851119700, updateTime=1793851119700, creator=18614031015, updator=18614031015)]), Website(id=1325070300580294698, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1324759914146529478, 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/jsjyjyfz/EN, language=EN, createTime=1793850978274, createBy=18614031015, updateTime=1793851036638, updateBy=18614031015, name=计算机研究与发展-英文, tplId=1146101810881728533, title=Journal of Computer Research and Development, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1325070925288321148, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=articleTextType, value=kx, createTime=1793851127216, updateTime=1793851127216, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925263155321, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=banner, value=null, createTime=1793851127210, updateTime=1793851127210, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925309292671, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=grayFlag, value=0, createTime=1793851127221, updateTime=1793851127221, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925254766712, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=logo, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/file/pic?fileId=U27tDsy717YF7XSPA+Vwdw==, createTime=1793851127208, updateTime=1793851127208, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925326069889, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=minRunFlag, value=0, createTime=1793851127225, updateTime=1793851127225, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925279932539, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/file/pic, createTime=1793851127214, updateTime=1793851127214, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925317681280, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=silenceFlag, value=0, createTime=1793851127223, updateTime=1793851127223, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925271543930, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1793851127212, updateTime=1793851127212, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925292515453, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=themeColor, value=null, createTime=1793851127217, updateTime=1793851127217, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925300904062, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=themeStyle, value=null, createTime=1793851127219, updateTime=1793851127219, creator=18614031015, updator=18614031015)])], journalTitle=计算机研究与发展, weixinUrl=null, journalUrl=https://crad.ict.ac.cn/, iacademicId=null, status=1, seqNo=null, journalTitleEn=Journal of Computer Research and Development, journalPhotoCn=k2l8zgqwstuxxevFX2qxVA==, journalPhotoEn=r6k0Ry+RD+c/VXj5QarmDw==, 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, interPubPlatform=, interPubPlatformUrl=null), detailUrlCn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/10.7544/issn1000-1239.202660468, detailUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/10.7544/issn1000-1239.202660468, pdfUrlCn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/PDF/10.7544/issn1000-1239.202660468, pdfUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/PDF/10.7544/issn1000-1239.202660468, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1788192000000, fullTextJson=null, articleText=null, reference=null)
收藏切换
PathRoute:基于全切片图像的主动证据路由的癌症生存预测方法
吕千一 1, 2 , 罗汉文 1, 2 , 王峻 2 , 李真 3 , 余国先 1, 2, *
计算机研究与发展 | “知识增强的人工智能前沿技术及应用”专题 2026,63(9): 2338-2354
收起
收藏切换
计算机研究与发展 |“知识增强的人工智能前沿技术及应用”专题 2026 , 63 (9) : 2338 -2354
PathRoute:基于全切片图像的主动证据路由的癌症生存预测方法
全屏
[Author(id=1329420527871046390, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=qylv@mail.sdu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420527946543865, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420527871046390, language=EN, stringName=Qianyi Lü, firstName=Qianyi, middleName=null, lastName=Lü, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1School of Software, Shandong University, Jinan 250101
2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101, bio={"content":"

born in 2002. Master candidate. His main research interests include computational pathology and machine learning

"}, bioImg=null, bioContent=

born in 2002. Master candidate. His main research interests include computational pathology and machine learning

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528022041338, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420527871046390, language=CN, stringName=吕千一, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1山东大学软件学院 济南 250101
2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101, bio={"img":"ZPvOAfPOY3lFP1xPlR7ogA==","content":"

2002年生。硕士研究生。主要研究方向为计算病理学、机器学习

"}, bioImg=ZPvOAfPOY3lFP1xPlR7ogA==, bioContent=

2002年生。硕士研究生。主要研究方向为计算病理学、机器学习

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527640359660, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=1, ext=[AuthorCompanyExt(id=1329420527648748269, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Software, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527661331182, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1山东大学软件学院 济南 250101)]), AuthorCompany(id=1329420527728440047, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=2, ext=[AuthorCompanyExt(id=1329420527741022960, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527749411569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101)])]), Author(id=1329420528080761596, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, 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=1329420528156259071, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528080761596, language=EN, stringName=Hanwen Luo, firstName=Hanwen, middleName=null, lastName=Luo, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1School of Software, Shandong University, Jinan 250101
2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101, bio={"content":"

born in 2002. PhD candidate. His main research interests include causal discovery and bioinformatics

"}, bioImg=null, bioContent=

born in 2002. PhD candidate. His main research interests include causal discovery and bioinformatics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528219173632, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528080761596, language=CN, stringName=罗汉文, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1山东大学软件学院 济南 250101
2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101, bio={"img":"iFNHyiIrK6evcBD6qUjb8A==","content":"

2002年生。博士研究生。主要研究领域为因果发现、生物信息学

"}, bioImg=iFNHyiIrK6evcBD6qUjb8A==, bioContent=

2002年生。博士研究生。主要研究领域为因果发现、生物信息学

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527640359660, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=1, ext=[AuthorCompanyExt(id=1329420527648748269, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Software, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527661331182, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1山东大学软件学院 济南 250101)]), AuthorCompany(id=1329420527728440047, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=2, ext=[AuthorCompanyExt(id=1329420527741022960, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527749411569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101)])]), Author(id=1329420528277893890, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, 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=1329420528353391364, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528277893890, language=EN, stringName=Jun Wang, firstName=Jun, middleName=null, lastName=Wang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101, bio={"content":"

born in 1983. PhD, professor, PhD supervisor. Senior member of CCF. Her main research interests include machine learning and bioinformatics

"}, bioImg=null, bioContent=

born in 1983. PhD, professor, PhD supervisor. Senior member of CCF. Her main research interests include machine learning and bioinformatics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528416305925, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528277893890, language=CN, stringName=王峻, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101, bio={"img":"lvAveDzuhsTDz+9MclokOA==","content":"

1983年生。博士,教授,博士生导师。CCF高级会员。主要研究方向为机器学习、生物信息学

"}, bioImg=lvAveDzuhsTDz+9MclokOA==, bioContent=

1983年生。博士,教授,博士生导师。CCF高级会员。主要研究方向为机器学习、生物信息学

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527728440047, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=2, ext=[AuthorCompanyExt(id=1329420527741022960, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527749411569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101)])]), Author(id=1329420528475026183, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, 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=1329420528554717961, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528475026183, language=EN, stringName=Zhen Li, firstName=Zhen, middleName=null, lastName=Li, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3Qilu Hospital of Shandong University, Jinan 250012, bio={"content":"

born in 1984. PhD, professor, chief physician. Her main research interest includes new technologies and methods for the study of digestive system diseases

"}, bioImg=null, bioContent=

born in 1984. PhD, professor, chief physician. Her main research interest includes new technologies and methods for the study of digestive system diseases

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528609243914, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528475026183, language=CN, stringName=李真, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3山东大学齐鲁医院 济南 250012, bio={"img":"lmUf+SEbTMzepB36Qqb1+Q==","content":"

1984年生。博士,教授,主任医师。主要研究方向为消化系统疾病研究新技术与新方法

"}, bioImg=lmUf+SEbTMzepB36Qqb1+Q==, bioContent=

1984年生。博士,教授,主任医师。主要研究方向为消化系统疾病研究新技术与新方法

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527803937522, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=3, ext=[AuthorCompanyExt(id=1329420527812326131, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527803937522, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3Qilu Hospital of Shandong University, Jinan 250012), AuthorCompanyExt(id=1329420527816520436, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527803937522, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3山东大学齐鲁医院 济南 250012)])]), Author(id=1329420528672158476, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, orderNo=4, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=gxyu@sdu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1329420528760238863, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528672158476, language=EN, stringName=Guoxian Yu, firstName=Guoxian, middleName=null, lastName=Yu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, *, address=1School of Software, Shandong University, Jinan 250101
2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101, bio={"content":"

born in 1985. PhD, professor, PhD supervisor. Distinguished member of CCF. His main research interests include data mining and bioinformatics

"}, bioImg=null, bioContent=

born in 1985. PhD, professor, PhD supervisor. Distinguished member of CCF. His main research interests include data mining and bioinformatics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420528818959121, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, authorId=1329420528672158476, language=CN, stringName=余国先, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, *, address=1山东大学软件学院 济南 250101
2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101, bio={"img":"r1VybPh6GVnmzwI5vQ94RA==","content":"

1985年生。博士,教授,博士生导师。CCF杰出会员。主要研究方向为数据挖掘、生物信息学

"}, bioImg=r1VybPh6GVnmzwI5vQ94RA==, bioContent=

1985年生。博士,教授,博士生导师。CCF杰出会员。主要研究方向为数据挖掘、生物信息学

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420527640359660, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=1, ext=[AuthorCompanyExt(id=1329420527648748269, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Software, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527661331182, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527640359660, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1山东大学软件学院 济南 250101)]), AuthorCompany(id=1329420527728440047, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, xref=2, ext=[AuthorCompanyExt(id=1329420527741022960, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101), AuthorCompanyExt(id=1329420527749411569, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420525178303191, companyId=1329420527728440047, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101)])])]
吕千一1, 2 , 罗汉文1, 2, 王峻2, 李真3, 余国先1, 2, *
作者信息
  • 1山东大学软件学院 济南 250101
  • 2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101
  • 3山东大学齐鲁医院 济南 250012
通讯作者:
余国先()
作者简介:

作者贡献声明

吕千一负责研究构思、方法学、软件实现、验证、实验实施、形式分析、可视化并撰写论文;罗汉文负责方法学,撰写并审阅论文;王峻、李真、余国先负责监督指导、撰写并审阅论文。

2002年生。硕士研究生。主要研究方向为计算病理学、机器学习

2002年生。博士研究生。主要研究领域为因果发现、生物信息学

1983年生。博士,教授,博士生导师。CCF高级会员。主要研究方向为机器学习、生物信息学

1984年生。博士,教授,主任医师。主要研究方向为消化系统疾病研究新技术与新方法

1985年生。博士,教授,博士生导师。CCF杰出会员。主要研究方向为数据挖掘、生物信息学

PathRoute: Active Evidence Routing for Cancer Survival Prediction on Whole Slide Pathology Images
Qianyi Lü1, 2 , Hanwen Luo1, 2, Jun Wang2, Zhen Li3, Guoxian Yu1, 2, *
Affiliations
  • 1School of Software, Shandong University, Jinan 250101
  • 2Joint SDU-NTU Centre for Artificial Intelligence Research, Shandong University, Jinan 250101
  • 3Qilu Hospital of Shandong University, Jinan 250012
  • born in 2002. Master candidate. His main research interests include computational pathology and machine learning

    born in 2002. PhD candidate. His main research interests include causal discovery and bioinformatics

    born in 1983. PhD, professor, PhD supervisor. Senior member of CCF. Her main research interests include machine learning and bioinformatics

    born in 1984. PhD, professor, chief physician. Her main research interest includes new technologies and methods for the study of digestive system diseases

    born in 1985. PhD, professor, PhD supervisor. Distinguished member of CCF. His main research interests include data mining and bioinformatics

出版时间: 2026-09-01 doi: 10.7544/issn1000-1239.202660468
文章导航
收藏切换

病理全切片图像(whole slide image, WSI)是癌症诊断与预后判断的核心载体。当前主流的病理全切片生存预测方法将海量的区块特征直接映射为全局表示并输出预后风险,不仅未能在建模中显式区分复杂的组织学构成、忽略了不同癌种特有的临床表型差异,也导致模型的内部推理过程缺乏可解释性。 针对上述问题,提出PathRoute,一个面向可解释生存预测的主动证据路由框架。该框架将WSI推理建模为先形成证据、再读出风险的两阶段有向过程:4类角色明确的可学习查询在统一隐空间中分工协作,分别承担组织组成感知、主动证据路由、表型语义辅助监督与风险读出;方向性注意力掩码进一步约束查询间的前向可见性,使风险预测在架构上成为前序证据表示的下游读出,而非端到端的黑盒映射。 在TCGA的5个常见实体瘤癌种上,采用统一输入特征与相同患者级5折交叉验证划分,PathRoute仅使用WSI作为输入,平均生存预测C-index优于代表性多模态融合方法与纯WSI方法,验证了结构化证据组织的有效性。进一步的分析表明,路由查询在未接受任何组织类型监督的条件下,仍在5个癌种上稳定地将注意力自发集中于肿瘤区域,为主动证据路由的可解释性提供了直接支撑。代码目前已在https://github.com/ZephyrGo/pathRoute开源。

计算病理学  /  全切片图像  /  生存分析  /  证据路由  /  可解释推理  /  多癌种

Whole slide images (WSIs) are central to cancer diagnosis and prognosis. However, mainstream survival prediction methods directly map massive tile features into a single global representation for risk readout, failing to explicitly distinguish complex histological composition, ignoring cancer-specific clinical phenotype differences, and leaving the model’s internal reasoning process opaque. To address these issues, we propose PathRoute, an active evidence routing framework for interpretable WSI survival prediction. The framework models WSI reasoning as a two-stage directional process that first forms evidence and then reads out risk: four role-specific learnable queries collaborate in a shared latent space, respectively performing tissue-composition sensing, active evidence routing, phenotype-semantic auxiliary supervision, and risk readout; a directional attention mask further constrains forward visibility between queries, so that risk prediction is structurally realized as a downstream readout of pre-formed evidence representations, rather than an end-to-end black-box mapping. On five TCGA solid tumor cohorts under unified inputs and identical patient-level 5-fold cross-validation splits, PathRoute using WSI alone achieves higher average C-index than representative multimodal and WSI-only methods, validating the effectiveness of structured evidence organization. Further analysis shows that routing queries consistently and spontaneously concentrate attention onto tumor regions across all five cancer types without any tissue-type supervision, providing direct interpretability support for the active routing mechanism. Code available at https://github.com/ZephyrGo/pathRoute.

computational pathology  /  whole slide image (WSI)  /  survival analysis  /  evidence routing  /  interpretable reasoning  /  pan-cancer
吕千一, 罗汉文, 王峻, 李真, 余国先. PathRoute:基于全切片图像的主动证据路由的癌症生存预测方法. 计算机研究与发展, 2026 , 63 (9) : 2338 -2354 . DOI: 10.7544/issn1000-1239.202660468
Qianyi Lü, Hanwen Luo, Jun Wang, Zhen Li, Guoxian Yu. PathRoute: Active Evidence Routing for Cancer Survival Prediction on Whole Slide Pathology Images[J]. Journal of Computer Research and Development, 2026 , 63 (9) : 2338 -2354 . DOI: 10.7544/issn1000-1239.202660468
病理全切片图像(whole slide image, WSI)是癌症诊断、分期与预后判断的核心载体,也是计算病理学最具挑战性的数据形态之一。一张切片在$ 20\times $放大倍率下可被切分为$ N\sim {10}^{4} $个区块(tile),但与患者预后真正相关的诊断证据在切片中稀疏分布且事先难以定位;与此同时,临床可获得的生存标签通常仅有数百例,远小于切片本身的视觉复杂度。这一特殊的数据形态决定了WSI生存预测无法依靠端到端微调一个超大视觉骨干来解决:模型既需要在大量背景区块中识别出真正具有预后价值的局部证据,又必须在弱监督信号下避免过拟合。
近年来,UNI[1]、UNI2[2]、Virchow2[3]、CONCH[4]与 GPFM[5]等病理基础模型在大规模自监督预训练下提供了高质量的通用区块表征,使下游任务无需再从头训练视觉骨干。在这一新范式下,WSI 生存预测的瓶颈从底层视觉特征学习转向在强表征之上的证据组织,即如何从数千个区块特征中识别、聚合并利用与预后相关的诊断信息。然而,主流的多示例学习方法[6-8]与多模态融合方法[9-12]大多沿用单步注意力聚合范式:将整张切片压缩为单一向量后直接预测风险。这一范式带来2个相互关联的问题。1)证据选择、临床先验注入与风险判断被叠加到同一个表示之中,模型究竟依据哪些区块、哪些临床语义形成最终预后判断难以被审计。2)对全量区块直接执行稠密自注意力的计算代价随区块数$ N $平方增长,对动辄$ N\sim {10}^{4} $的 WSI 不可行;硬选择不可微,随机采样亦无法保证命中稀疏的诊断性证据。
针对上述问题,本文提出 PathRoute,一个构建于冻结病理基础模型之上、面向可解释生存预测的主动证据路由框架。其核心思想是将 WSI 推理过程从单步聚合改造为先形成证据、再读出风险的两阶段有方向过程,并由一组角色明确的可学习查询在统一隐空间中分别承担4类语义功能:组织组成感知查询可以感知切片的整体组织构成;路由查询以软注意力方式从全量区块中聚合固定数量的诊断证据;表型语义查询通过癌种相关辅助监督接入临床先验;风险查询在前序信息就位的条件下完成最终预后判断。为使该信息流在架构层面具备结构性,PathRoute 进一步引入方向性注意力掩码,从前向计算上约束不同查询之间的可见性方向,使风险预测成为前序证据表示的下游读出,具体框架如图1。本文的主要贡献如下:
1)面向全切片生存预测的主动证据路由框架。 提出了一种面向WSI生存预测的主动证据路由框架 PathRoute。该框架打破了主流方法端到端“单步全局聚合”的黑盒范式,首创了“先提取证据、再预测风险”的两阶段定向推理过程。通过在统一隐空间中解耦4类功能明确的可学习查询(组织感知、路由聚合、表型语义、风险读出),并引入方向性注意力掩码,实现了对复杂病理特征的结构化建模与信息流的严格约束。
2)无监督可解释性机制。 赋予了模型高度贴合临床直觉的可解释推理能力。实验证实,在未接受任何病理组织类型直接监督的情况下,PathRoute中的路由查询能够跨越5个解剖部位完全不同的实体瘤,自发且稳定地将注意力精准聚焦于肿瘤区域。这不仅解决了WSI中诊断证据高度稀疏的痛点,更为生存预测中“肿瘤感知证据组织过程”提供了直接的视觉解释依据。
3)显著提升的跨癌种泛化与预测性能。 在5个代表性的 TCGA 实体瘤数据集(BRCA、STAD、COADREAD、HNSC、BLCA)的严谨交叉验证中,PathRoute仅依赖WSI单模态输入,其平均生存预测性能便全面超越了当前主流的纯WSI基础模型方法与多模态融合方法,充分验证了主动证据路由机制在临床预后评估中的有效性与鲁棒性。
大规模自监督预训练为计算病理学提供了高质量的通用区块表征,使下游任务对标注数据与任务特异性骨干训练的依赖显著降低。CONCH[4]基于对比语言-图像预训练范式,在病理图像-文本配对数据上学习跨模态语义表示。UNI[1]与其后续版本 UNI2[2]基于 DINOv2[13]自蒸馏框架,在超过10万张 WSI 上进行大规模预训练,所学表征在下游任务中展现出稳健的零样本与少样本迁移能力。Virchow2[3]采用掩码自编码器预训练策略,在混合放大倍率的病理图像上训练。GPFM[5]同时融合多个专家模型与自蒸馏目标,在 72 项病理下游任务的基准评测中取得平均排名第1。
本文选择 UNI2 作为冻结特征提取器,将视觉表征学习与下游推理建模解耦。这一设计带来2点直接好处:其一,无需针对特定癌种或任务微调骨干,规避了在仅有数百例生存标签条件下端到端训练大模型的不稳定性;其二,区块特征可预提取并缓存,训练与推理时无需加载完整骨干,显著降低显存占用。
基于 WSI 的生存预测方法经历了从手工特征到深度学习再到多模态融合的演进。早期工作从病理图像中提取手工设计的形态学特征并结合 Cox 比例风险模型进行生存回归[14-15]。随后 Kather 等人[16]证明卷积神经网络可从病理图像中学习与多癌种分子改变相关的形态表征,说明深度视觉特征能够捕获具有临床意义的组织学信号。
多示例学习(MIL)框架将 WSI 视为由数千区块组成的“包”,区块为“实例”,从而把千兆像素级图像分析转化为集合级预测问题[17]。Campanella等人[18]在大规模临床数据上验证了弱监督 MIL 的可行性[19]。ABMIL[6]引入参数化注意力机制对每个区块独立评分并加权聚合,成为后续方法的标准基线[7-8,20-21]。为克服 ABMIL忽略区块间关系的局限,TransMIL[7]引入Transformer自注意力建模区块间的全局交互;DTFD-MIL[8]通过双层特征蒸馏在效率与精度之间取得平衡。多模态融合方法进一步扩展了输入信息源:MCAT[9]设计协同注意力机制融合 WSI形态特征与基因组通路特征;SurvPath[10]以密集交互方式融合 WSI 与生物通路token;CIMA[11]显式建模WSI与基因组通路之间的跨模态对齐;而SurvPGC[12]在SurvPath基础上以双路径跨注意力进一步纳入临床文本信息。
上述方法在 WSI 侧的处理高度同构:大多以单步注意力聚合将整张切片压缩为全局向量后送入风险预测头,使证据选择、临床先验注入与风险判断混叠在同一表示之中。多模态融合虽能提升精度,但依赖额外基因组或临床数据。
用一组可学习查询从大量输入 token 中聚合信息,已成为视觉与多模态领域处理大规模或不定长输入的成熟方法学。DETR[22-23]在目标检测中引入一组目标查询向量,通过交叉注意力从图像特征中读取目标候选;Perceiver 与 PerceiverIO[24-25]用一组潜在查询向量把任意大规模、任意模态的输入压缩到固定长度,使后续 Transformer 计算与输入规模解耦;BLIP-2 的 Q-Former[26-27]通过 32 个可学习查询把视觉表征桥接到语言模型,实现了跨模态信息的紧凑路由。在病理领域,已有部分工作尝试将类似思路引入 WSI 聚合:例如基于可学习提示词元的MIL变体[28]在ABMIL框架上引入额外查询以增强表示能力。这些工作共同表明,查询式注意力路由能够同时实现可微的软选择、与输入规模解耦的固定长度推理,以及对稀疏证据的有效聚焦。
然而,已有查询式方法仍难以满足 WSI 生存预测中的结构化推理需求。首先,查询之间缺乏明确的角色分化。无论是 DETR 中的目标查询,还是 Q-Former 中的潜在查询,其查询通常作为一组同构、可互换的表示槽位存在,模型并未显式规定不同查询分别承担组织感知、证据聚合、临床先验注入或风险判断等功能。这会导致不同预后相关信息在同一组查询中混合编码,使查询语义难以稳定解释,也削弱了中间表示与具体临床功能之间的对应关系。其次,已有方法通常允许查询之间进行充分的双向交互,缺乏面向推理流程的方向性约束。这样虽然有利于信息融合,但也会使证据形成与风险读出在前向计算中相互混杂:用于最终风险预测的判别信息可能提前影响中间证据表示,使模型难以区分哪些表示对应上游病理证据、哪些表示服务于下游风险决策。因此,现有查询聚合机制虽能获得紧凑的全局表征,却难以显式刻画“证据先形成、风险后判断”的推理结构。
本文的PathRoute在这条方法学脉络上的定位是:将“查询式路由”扩展为角色化、有方向的查询式推理。具体来说,PathRoute 为不同认知功能定义独立的查询子集(组织组成感知、证据路由、表型语义、风险读出),并通过方向性注意力掩码在前向计算中显式分离证据形成阶段与风险读出阶段,使最终预测在架构上成为前序证据表示的下游读出。
可解释性是计算病理方法走向临床落地的关键瓶颈之一。现有解释方法大多属于事后归因范畴:注意力可视化将 ABMIL 或 TransMIL 中的区块级注意力权重映射回 WSI 空间,指示模型“关注”了哪些区域[6-7];Grad-CAM 类方法[29]通过梯度归因近似输入对输出的贡献;多尺度证据图[30]进一步将归因从单层注意力扩展至跨尺度证据定位。这些方法的共同优势是不修改原模型即可应用,但其局限也很明确:归因是在模型训练完成后外加的近似估计,而模型的内部信息流本身仍是不可审计的黑箱。
与事后归因互补的另一条路径,是在架构设计阶段就为模型提供具有显式语义角色的中间槽位。概念瓶颈模型(concept bottleneck model, CBM)[31]强制风险预测必须经由一组人类可读的中间概念,是这一路径的代表性尝试。但 CBM 类方法把全部预测压力施加在唯一的概念瓶颈上,当瓶颈内的概念无法覆盖所有相关信息时,模型性能会显著下降。
本文的 PathRoute在事后归因与架构层可解释这2条路径之间走出了一条折中路线:不依赖事后注意力可视化,而是在架构层面为不同认知功能划分独立的可读语义槽位(组织、证据、表型、风险);不同于概念瓶颈模型把表型语义设为风险预测的唯一中介路径,本文将其作为多个并列的可监督槽位之一,通过辅助监督向共享表示注入临床先验。这一设计在保留事后归因方法可应用性的同时,使模型的中间表示天然可读、可监督,从而提供“设计时可解释”的额外保障。
本文提出 PathRoute,一个构建于冻结病理基础模型之上的主动证据路由框架,其核心设计是将全切片推理过程组织为先形成证据、再读出风险的两阶段结构。基础模型输出的区块特征首先被统一映射至后续推理所在的隐空间,并注入空间位置与癌种条件信息(2.2节);同时进入证据形成阶段(2.3节),该阶段并行完成2类互补操作:组织组成感知在全局视野下建立切片的组织学先验,主动证据路由通过可学习查询从全量区块中聚合固定数量的诊断证据。在此基础上,风险读出阶段(2.4节)通过方向性注意力掩码约束不同语义查询之间的前向可见性,使风险预测在架构上成为前序证据表示的下游读出,并借助癌种相关的表型辅助监督向共享表示注入临床先验。训练目标与计算复杂度分析分别在2.5节和2.6节给出。
本节中,$ N $表示全切片的区块总数,$ d $与$ D $分别表示基础模型输出维度与 PathRoute 隐空间维度,$ c $表示癌种 ID。PathRoute 引入4组角色明确的可学习查询,分别由$ {Q}_{1} $、$ {Q}_{2} $、$ {Q}_{3} $、$ {Q}_{4} $个查询构成,记为$ {\boldsymbol{Q}}_{{\mathrm{tissue}}} $(组织组成感知)、$ {\boldsymbol{Q}}_{{\mathrm{route}}} $(证据路由)、$ {\boldsymbol{Q}}_{{\mathrm{phen}}} $(表型语义)与$ {\boldsymbol{Q}}_{{\mathrm{risk}}} $(风险读出)。证据形成阶段使用一个独立的$ {L}_{{\mathrm{routing}}} $层路由 Transformer 来承担证据路由,并在推理 Transformer 的前$ {L}_{1} $层完成组织组成感知;风险读出阶段则在推理 Transformer 的后$ L-{L}_{1} $层(其中$ L $为推理 Transformer 总层数)展开。生存预测部分将连续生存时间离散为$ K $个风险区间。其余局部符号在各节首次出现时定义。
大规模自监督预训练使病理基础模型在海量切片数据上学习到了高质量的通用区块表征。因此,区块特征编码的目标是在不破坏基础模型所学表征的前提下,将其映射至推理模块所在的隐空间,并叠加空间位置与癌种条件,使每个区块的表示在进入证据形成阶段之前即携带完整的上下文信息。
具体地,设全切片经预处理裁剪为$ N $个区块,由冻结的 UNI2 基础模型提取得到区块特征矩阵$ \boldsymbol{F}\in {{\mathbb{R}}}^{N\times d} $,其中$ d=1\;536 $为 UNI2 的输出特征维度;同时记录每个区块在切片中的归一化坐标$ \boldsymbol{C}\in{{\mathbb{R}}}^{N\times 4} $(中心坐标$ x,y $与尺寸$ w,h $)。为使区块特征与后续模块的查询向量在同一隐空间中进行注意力交互,本节将$ \boldsymbol{F} $投影至$ D $维隐空间。线性投影矩阵$ {\boldsymbol{W}}_{{\mathrm{proj}}}\in{{\mathbb{R}}}^{d\times D} $与偏置$ {\boldsymbol{b}}_{{\mathrm{proj}}}\in{{\mathbb{R}}}^{D} $给出:
$ {\boldsymbol{V}}_{{0}}={\boldsymbol{FW}}_{{{\mathrm{proj}}}}+{\boldsymbol{b}}_{{{\mathrm{proj}}}}。 $
空间位置嵌入$ {\boldsymbol{E}}_{{\mathrm{spatial}}}\in {{\mathbb{R}}}^{N\times D} $由多层感知机以归一化坐标$ \boldsymbol{C} $为输入,生成编码区块在切片中的绝对位置与尺寸信息。癌种嵌入$ {\boldsymbol{e}}_{{\mathrm{cancer}}}(c)\in {{\mathbb{R}}}^{D} $为以癌种 ID$ c $为索引的可学习向量,加和注入所有查询的初始化,使后续推理以癌种类型为条件。投影与位置嵌入之和经层归一化后得到视觉特征库:
$ {{\boldsymbol{V}}}={{LayerNorm}}({\boldsymbol{V}}_{{0}}+{\boldsymbol{E}}_{{{\mathrm{spatial}}}})\in {\mathbb{R}}^{{N}\times{D}}, $
作为证据形成阶段2个子模块的共同输入。
获得全切片的统一区块表示$ \boldsymbol{V} $之后,证据形成阶段的任务是从中提炼出2类互补的前驱表示,供风险读出阶段使用:其一是反映切片整体组织构成的组织学先验,其二是凝聚诊断相关区块信息的证据向量集合。值得注意的是,一张切片通常包含$ N\sim {10}^{4} $个区块,对全量区块执行稠密自注意力的计算代价随$ N $平方增长,在实践中不可行。
为此,2个子模块均以从$ \boldsymbol{V} $中均匀采样得到的上下文子集$ {\boldsymbol{V}}_{{\mathrm{ctx}}}\in {{\mathbb{R}}}^{C\times D} $作为共享输入,其中$ m\ll N $为预设的上下文 token 数。该子集既保留了全切片的空间代表性,又把后续 Transformer 模块内部的自注意力序列长度从$ O(N) $降至常数。2个子模块以相同的$ {\boldsymbol{V}}_{{\mathrm{ctx}}} $为输入并行展开,互不依赖。
1)组织组成感知
推理 Transformer 由$ L $层注意力块构成,其前$ {L}_{1} $层($ {L}_{1}<L $)专门承担组织组成感知的计算,负责建立切片整体的组织学先验。由$ {Q}_{1} $个角色对称的可学习向量构成的组织查询$ {\boldsymbol{Q}}_{{\mathrm{tissue}}}\in {{\mathbb{R}}}^{{{Q}_{1}}\times D} $,与共享上下文拼接形成序列$ [{\boldsymbol{V}}_{{\mathrm{ctx}}}||{\boldsymbol{Q}}_{{\mathrm{tissue}}}] $后送入$ {L}_{1} $层注意力块。在自注意力中,视觉 token 之间保持双向交互以充分建模组织间的空间依赖,组织查询单向读取视觉 token,并按位置序排列以诱导查询间的语义分化,避免不同查询坍缩至同一组织类型。该阶段输出组织先验表示$ {\boldsymbol{H}}_{{\mathrm{tissue}}}\in {{\mathbb{R}}}^{{{Q}_{1}}\times D} $,经层归一化后送入组织类型分类头,预测切片的组织类型构成;同时以梯度截断的形式注入下一阶段。
组织类型共设肿瘤、基质、坏死、脂肪、正常上皮、淋巴细胞浸润6 类,监督标签由 CONCH[4]零样本分类生成的区块级伪标签提供。由于组织查询预测的是一个无序集合,训练时采用 Hungarian 匹配[32]将查询预测与真值组织类型一一对齐,在匹配对上计算交叉熵损失。
2)主动证据路由
与组织组成感知并行,主动证据路由这一分支负责从全量区块中聚合固定数量的诊断证据,从而使下游推理的序列长度与全量区块数$ N $解耦,如图2所示。PathRoute 为此设计了一个独立的$ {L}_{{\mathrm{routing}}} $层路由 Transformer,与上述组织组成感知所在的推理 Transformer 在参数上完全独立。
由$ {Q}_{2} $个可学习向量构成的路由查询$ {\boldsymbol{Q}}_{{\mathrm{route}}}\in {{\mathbb{R}}}^{{{Q}_{2}}\times D} $与上下文$ {\boldsymbol{V}}_{{\mathrm{ctx}}} $拼接后送入该子模块,其在路由 Transformer 顶层的输出表示记为$ {\boldsymbol{Q}}_{{\mathrm{routing}}}'\in {{\mathbb{R}}}^{{{Q}_{2}}\times D} $。$ {\boldsymbol{Q}}_{{\mathrm{routing}}}' $经线性投影矩阵$ {\boldsymbol{W}}_{{\mathrm{routing}}}\in {{\mathbb{R}}}^{D\times D} $与全量区块特征$ \boldsymbol{V} $做缩放点积,得到路由得分:
$ {\boldsymbol{S}}_{{{\mathrm{route}}}}=\frac{{\boldsymbol{Q}}_{{{\mathrm{routing}}}}'{\boldsymbol{W}}_{{{\mathrm{routing}}}}{\boldsymbol{V}}^{{\mathrm{T}}}}{\sqrt{{D}}}\in {{\mathbb{R}}}^{{{\boldsymbol{Q}}_{\mathbf{2}}}\times \boldsymbol{N}}。 $
在路由得分上施加$ {softmax} $归一化(令$ {\boldsymbol{A}}= {{softmax}}({\boldsymbol{S}}_{{\mathrm{route}}}) $)后对全量区块进行连续加权聚合:
$ {\boldsymbol{V}}_{{{\mathrm{sel}}}}={\boldsymbol{A}}\cdot {\boldsymbol{V}}\in {{\mathbb{R}}}^{{{{Q}}_{\mathbf{2}}}\times {D}}。 $
每个路由查询生成一个聚合证据向量,共同构成诊断证据$ {\boldsymbol{V}}_{{\mathrm{sel}}} $。该设计使路由器可在生存损失监督下联合优化,同时对低得分区块中潜在的弱预后信号保留梯度通道。路由查询自身的表示在生成$ {\boldsymbol{V}}_{{\mathrm{sel}}} $后即被舍弃。
3)阶段间衔接
至此,2条并行的分支分别产出了诊断证据集合$ {\boldsymbol{V}}_{{\mathrm{sel}}} $与组织先验$ {\boldsymbol{H}}_{{\mathrm{tissue}}} $,它们将汇入风险读出阶段。聚合证据$ {\boldsymbol{V}}_{{\mathrm{sel}}} $与上下文$ {\boldsymbol{V}}_{{\mathrm{ctx}}} $拼接,构成增强的视觉序列;而组织先验$ {\boldsymbol{H}}_{{\mathrm{tissue}}} $则以梯度截断方式注入:
$ {{\boldsymbol{H}}}_{{{\mathrm{tissue}}}}^{\dagger}={{{stopgrad}}}\left({\boldsymbol{H}}_{{{\mathrm{tissue}}}}\right)。 $
截断操作确保了组织组成感知与风险读出阶段在优化目标上保持严格隔离,防止生存损失回传导致组织查询的语义纯度受损。
风险读出阶段直接建立在证据形成阶段所提供的两路前驱表示之上。推理 Transformer 的后$ L-{L}_{1} $层在组装好的综合序列上展开计算:
$ {\boldsymbol{X}}_{{{\mathrm{read}}}}=\left[{\boldsymbol{V}}_{{{\mathrm{ctx}}}}||{\boldsymbol{V}}_{{{\mathrm{sel}}}}||{\boldsymbol{H}}_{{t}{{\mathrm{issue}}}}^{\dagger}||{\boldsymbol{Q}}_{{{\mathrm{phen}}}}||{\boldsymbol{Q}}_{{{\mathrm{risk}}}}\right]\;, $
并在自注意力运算中施加方向性注意力掩码$ {M} $。
1)方向性注意力掩码
如图3所示,方向性注意力掩码$ {M} $旨在约束序列$ {\boldsymbol{X}}_{{\mathrm{read}}} $的前向信息流向,规则如下。
底层证据双向互见,即视觉证据 token ($ {\boldsymbol{V}}_{{\mathrm{ctx}}} $与$ {\boldsymbol{V}}_{{\mathrm{sel}}} $) 之间保持双向交互;
高层查询单向读取,即所有查询 token 可单向读取视觉证据,但查询对视觉证据token不可见;
查询角色严格排序,即查询 token 之间按角色逻辑先后施加下三角约束:$ \boldsymbol{H}_{{\mathrm{tissue}}}^{\dagger}\to{\boldsymbol{Q}}_{{\mathrm{phen}}}\to{\boldsymbol{Q}}_{{\mathrm{risk}}} $。
这一掩码机制保障了风险查询必须在视觉证据、组织先验与表型语义均就位的条件下才能完成预后判断;同时,表型查询在计算时无法窥视风险查询的中间表征,保证了其作为独立临床监督接口的可读性。
2)表型语义辅助监督
在完成序列交互后,由$ {Q}_{3} $个向量构成的表型查询$ {\boldsymbol{Q}}_{{\mathrm{phen}}} $输出表示$ {\boldsymbol{H}}_{{\mathrm{phen}}}\in {\mathbb{R}}^{{{Q}_{3}}\times D} $,经平均池化送入癌种相关的辅助监督头,用于预测受体状态、组织学分级等形态学相关的临床表型。该设计旨在为共享推理表示注入临床先验知识,其角色定位为辅助监督而非风险预测的强制中介。
3)风险读出
作为整个模型的最终输出端,由$ {Q}_{4} $个向量构成的风险查询$ {\boldsymbol{Q}}_{{\mathrm{risk}}} $输出表示$ {\boldsymbol{H}}_{{\mathrm{risk}}}\in {{\mathbb{R}}}^{{{Q}_{4}}\times D} $,经平均池化后送入生存预测头。模型采用离散时间生存预测框架[10],预测头输出$ K $个风险区间的 hazard 率$ {h}_{1},…,{h}_{K} $,对应的离散生存函数为:
$ {S}({{t}}_{{j}})=\prod\limits_{{k}={1}}^{{j}}({1}-{{h}}_{{k}})。 $
最终计算得出的风险分数为$ r=-\displaystyle\sum\limits_{j=1}^{K}S({t}_{j}) $。风险区间边界由训练集中未删失样本生存时间的$ K $分位数事先确定。
PathRoute 在训练阶段同时优化3类监督损失与1项查询多样性正则。各项损失分别对应不同的设计目标,而非简单的多任务加权:组织监督约束组织组成感知查询学习稳定的全局组织学先验;表型监督为共享推理表示注入癌种相关的临床先验;生存监督直接优化最终风险输出;多样性正则约束同组查询不发生表示坍缩。总损失$ \mathcal{L} $定义为:
$ {\cal L}={ {\lambda }}_{ {{\mathrm{tissue}}}}{\cal L}_{ {{\mathrm{tissue}}}}+{ {\lambda }}_{ {{\mathrm{phen}}}}{\cal L}_{ {{\mathrm{phen}}}}+{ {\lambda }}_{ {{\mathrm{surv}}}}{\cal L}_{ {{\mathrm{surv}}}}+{ {\lambda }}_{ {{\mathrm{div}}}}{\cal L}_{ {{\mathrm{div}}}}。 $
1)组织监督$ {\cal{L}}_{{\mathrm{tissue}}} $。组织组成感知查询的预测与无序的真值组织集合通过 Hungarian 匹配[32]一一对齐后计算交叉熵。组织类型标签由 CONCH[4]零样本分类生成的区块级伪标签提供,5个癌种共用同一套 6 类组织定义。
2)表型监督$ {\cal{L}}_{{\mathrm{phen}}} $。由通用调度器根据癌种注册表自动派发。调度器扫描当前癌种注册的所有监督任务,根据预测 logit 维度自动推断损失类型:输出维度为 1 时使用二元交叉熵(BCE);目标为整数序数时使用累积 BCE;其余情况使用多类交叉熵。
3)生存监督$ {\cal L}_{{\mathrm{surv}}} $。采用离散时间负对数似然损失[33]约束风险输出。设$ B $为当前训练批次的样本数,$ {c}_{i}\in \{0{,}1\} $为样本$ i $的删失指示变量($ {c}_{i}=1 $表示删失),$ {y}_{i}\in \{1,… ,K\} $为该样本所属的风险区间索引(全文以$ k\in \{1,… ,K\} $表示风险区间索引,以$ i\in \{1,… ,B\} $表示批内样本索引,二者不混用;式中$ S(\cdot ) $依式(7)定义,$ {h}_{k} $为第$ k $个区间的 hazard 率),则生存损失定义为:
$ \begin{split}{\cal L}_{{{\mathrm{surv}}}}=\;& -\frac{{1}}{{B}}\sum\limits_{{i}={1}}^{{B}}[({1}-{{c}}_{{i}})(\log {S}({{t}}_{{{{y}}_{{i}}}-{1}})+\log {{h}}_{{{{y}}_{{i}}}})+\\& {{c}}_{{i}}\log {S}({{t}}_{{{{y}}_{{i}}}})]。\end{split} $
4)多样性正则$ {\cal{L}}_{{\mathrm{div}}} $。为了促进不同查询编码不同的语义,模型对组织、表型、风险这3组查询的输出向量分别计算归一化余弦相似度矩阵,并惩罚同组查询之间的非对角相似度均值。设$ s \in \{{\rm{tissue}}, {\rm{phen}}, {\rm{risk}}\} $为查询组索引,$ {Q}_{s} $为该组内的查询数量,$ {{\boldsymbol{q}}}_{i}^{(s)} $为该组的第$ i $个查询向量表示,正则项计算为:
$ {\cal L}_{{{\mathrm{div}}}}=\frac{{1}}{{3}}\sum\limits_{{s}}\frac{{1}}{{{Q}}_{{s}}\left({{Q}}_{{s}}-{1}\right)}\sum\limits_{{i}\neq {j}}|\cos ({{\boldsymbol{q}}}_{{i}}^{\left({s}\right)},{{\boldsymbol{q}}}_{{j}}^{\left({s}\right)})|。 $
需要特别指出的是,路由查询$ {\boldsymbol{Q}}_{{\mathrm{route}}} $不参与此项正则,因为它们的“多样性”已经由其对全切片不同区域的注意力分布间接体现和保障。
式(8)的4项损失并非在所有训练步骤中同等激活。$ {\cal{L}}_{{\mathrm{div}}} $在每次前向传播中无条件计算;$ {\cal{L}}_{{\mathrm{tissue}}} $仅在当前切片具有 CONCH 伪标签时激活;$ {\cal{L}}_{{\mathrm{phen}}} $仅在当前癌种已注册表型监督任务且批次中存在对应标签时生效;$ {\cal{L}}_{{\mathrm{surv}}} $则在所有具有生存时间与事件标注的样本上均参与计算,在本文5个癌种数据集中无缺失。
算法1 给出了 PathRoute 单次前向传播的完整流程,与上述各节的方法描述保持一致,可作为实现参考。
算法1. PathRoute 端到端推理与损失计算流程。
输入: 基础模型提取的区块特征$ {\boldsymbol{F}}\in {{\mathbb{R}}}^{N\times d} $,归一化坐标$ {\boldsymbol{C}}\in {{\mathbb{R}}}^{N\times 4} $,癌种 ID$ c $;
输出: 风险分数$ r $,联合损失$ \mathcal{L} $。
/* 1. 区块特征编码*/
① $ {\boldsymbol{V}}_{0}\leftarrow {\boldsymbol{FW}}_{{\mathrm{proj}}}+{\boldsymbol{b}}_{{\mathrm{proj}}} $;
② $ {\boldsymbol{V}}\leftarrow{{{LayerNorm}}}({\boldsymbol{V}}_{0}+{{MLP}}({\boldsymbol{C}})) $;/*癌种嵌入加 入各查询初始化*/
③ $ {\boldsymbol{V}}_{{\mathrm{ctx}}}\leftarrow {UniformSample}({\boldsymbol{V}},m) $;/*保留$ m $个 token*/
/* 2. 证据形成阶段:双分支并行*/
/* %分支2a:组织组成感知*/
④ $ {\boldsymbol{H}}_{{\mathrm{tissue}}}\leftarrow{{Transformer}}_{1\to{{L}_{1}}}([{\boldsymbol{V}}_{{\mathrm{ctx}}}||{\boldsymbol{Q}}_{{\mathrm{tissue}}}]) $;
⑤ $ {{\boldsymbol{H}}}_{{\mathrm{tissue}}}^{\dagger}\leftarrow {stop}\_{gradient}({\boldsymbol{H}}_{{\mathrm{tissue}}}) $;/*维持语义纯度 的梯度截断*/
/* % 分支 2b:主动证据路由*/
⑥ $ {\boldsymbol{Q}}_{{\mathrm{routing}}}'\leftarrow{{RoutingTransformer}}_{1\to{{L}_{{\mathrm{routing}}}}} $ ($[{\boldsymbol{V}}_{{\mathrm{ctx}}}||{\boldsymbol{Q}}_{{\mathrm{route}}}] $);
⑦ $ {\boldsymbol{S}}_{{\mathrm{route}}}\leftarrow \dfrac{{\boldsymbol{Q}}_{{\mathrm{routing}}}'{\boldsymbol{W}}_{{\mathrm{routing}}}{\boldsymbol{V}}^{{\mathrm{T}}}}{\sqrt{D}} $,$ {\boldsymbol{A}}\leftarrow {softmax}({\boldsymbol{S}}_{{\mathrm{route}}}) $;
⑧ $ {\boldsymbol{V}}_{{\mathrm{sel}}}\leftarrow {{\boldsymbol{AV}}} $(加权聚合得到诊断证据集合);
/* 3. 风险读出阶段*/
⑨ $ {\boldsymbol{X}}_{{\mathrm{read}}}\leftarrow[{\boldsymbol{V}}_{{\mathrm{ctx}}}||{\boldsymbol{V}}_{{\mathrm{sel}}}||{\boldsymbol{H}}_{{\mathrm{tissue}}}^{\dagger}||{\boldsymbol{Q}}_{{\mathrm{phen}}}||{\boldsymbol{Q}}_{{\mathrm{risk}}}] $; /*构造方向 性注意力掩码$M $,约束前向可见性*/
⑩ $ {\boldsymbol{H}}_{{\mathrm{out}}}\leftarrow{{Transformer}}_{{{L}_{1}}+1\to L}({\boldsymbol{X}}_{{\mathrm{read}}},{{{mask}}={{M}})} $;
⑪ 从$ {\boldsymbol{H}}_{{\mathrm{out}}} $中按序拆分出表型输出$ {\boldsymbol{H}}_{{\mathrm{phen}}} $与风险输 出$ {\boldsymbol{H}}_{{\mathrm{risk}}} $;
/* 4. 读出与监督计算*/
⑫ $ {\hat{y}}_{{\mathrm{tissue}}}\leftarrow {TissueHead(}{\boldsymbol{H}}_{{\mathrm{tissue}}}) $;
⑬ $ {\hat{y}}_{{\mathrm{phen}}}\leftarrow{PhenotypeHead({{MeanPool}}(}{\boldsymbol{H}}_{{\mathrm{phen}}})) $;
⑭ $ {h}_{1},…,{h}_{K}\leftarrow {SurvivalHead({{MeanPool}}(}{\boldsymbol{H}}_{{\mathrm{risk}}})) $;
⑮ $ S({t}_{j})\leftarrow \prod\limits_{k=1}^{j}(1-{h}_{k}) $,其中$ j\in \{1,…,K\} $;
⑯ $ r\leftarrow-\displaystyle\sum\limits_{j=1}^{K}S({t}_{j}) $;
⑰ $ {\cal L}\leftarrow{\lambda }_{{\mathrm{tissue}}}{\cal{L}}_{{\mathrm{tissue}}}+{\lambda }_{{\mathrm{phen}}}{\cal{L}}_{{\mathrm{phen}}}+{\lambda }_{{\mathrm{surv}}}{\cal{L}}_{{\mathrm{surv}}}+{\lambda }_{{\mathrm{div}}}{\cal{L}}_{{\mathrm{div}}} $;
⑱ 返回$ r,{\cal L} $。
WSI生存预测的现实瓶颈在于:对$ N\sim {10}^{4} $的全量区块直接执行稠密自注意力计算在显存和算力上均不可行。PathRoute 通过两阶段设计将主要的注意力计算转化为固定长度的交互,使风险读出阶段的计算量与切片物理大小彻底解耦。
具体而言,各阶段的计算复杂度如下:
1)主动证据路由。该阶段的输入序列长度为$ C+{Q}_{2} $;此外,路由查询还需与全量$ N $个区块进行缩放点积以计算路由得分。因此,其计算复杂度为$ {{\cal O}((}C+{Q}_{2}{)}^{2}+{Q}_{2}N) $。
2)组织组成感知。参与计算的序列长度为$ C+ {Q}_{1} $,其注意力计算复杂度为$ {{\cal O}((}C+{Q}_{1}{)}^{2}) $,与区块总数$ N $无关。
3)风险读出。参与前向交互的综合序列长度为$ C+{Q}_{2}+{Q}_{1}+{Q}_{3}+{Q}_{4} $,其复杂度为$ {{\cal O}((}C+{Q}_{2}+{Q}_{1}+ {Q}_{3}+ {Q}_{4}{)}^{2}) $,同样与区块总数$ N $无关。
综合来看,得益于主动证据路由的降维作用,PathRoute 的端到端注意力计算复杂度仅线性依赖于区块总数$ N $(主要来源为路由得分计算项$ {Q}_{2}N $),其余所有的深层语义交互计算均与切片的区块总数$ N $严格解耦。
实验在TCGA[34]PanCanAtlas的5个癌种(BRCA、STAD、COADREAD、HNSC、BLCA)上进行,覆盖乳腺癌、胃癌、结直肠癌、头颈癌和膀胱癌等常见实体瘤;样本量分别为1 111、398、565、467、455。所有实验采用相同的患者级5折交叉验证划分,确保与现有方法的公平比较。各癌种样本量见本节说明。
生存预测的评估指标为Harrell’s C-index[35],对每个癌种以5折交叉验证给出验证集的C-index均值$ \pm $标准差,表1中Overall列为5个癌种均值的算术平均。C-index衡量模型预测的风险排序与实际生存时间排序的一致性,取值范围为$ [0,1] $,0.5对应随机预测。
1)特征提取。所有WSI方法均使用UNI2[2]作为冻结骨干网络,输出特征维度$ d=1\;536 $;WSI经20倍放大裁剪为$ 256\times 256 $像素的区块,每张切片预提取得到的切片级特征以HDF5格式缓存为统一输入。
2)模型架构。PathRoute的架构超参数在所有癌种和所有实验中统一固定:隐藏维度$ D=1\;024 $;4类查询数$ {Q}_{1}/{Q}_{2}/{Q}_{3}/{Q}_{4}=8/16/16/8 $;上下文采样数$ C=200 $;路由Transformer层$ {L}_{{\mathrm{routing}}}=4 $,推理Transformer层$ L= 12 $(前4层用于组织组成感知);注意力头数为16,FFN维度为4096;组织类型数$ T=6 $,生存风险区间数$ K=4 $。所有实验采用统一训练设置。查询数($ {Q}_{3} $)、组织组成感知层数($ {L}_{1} $)与表型损失权重($ {\lambda }_{{\mathrm{phen}}} $)等核心超参数的敏感性分析见附录C,模型对其在合理范围内不敏感(C-index最大波动小于0.02)。
本文选取12种代表性相关方法进行比较:
1)纯基因组方法。SNN[36]为自归一化神经网络,仅以基因组通路特征为输入,代表基因组单模态基线的性能上界。MLP-Omics为基因组通路特征上的多层感知机基线,作为相对于SNN的简单参照模型,用以衡量基因组单模态在最简结构下可达的水平。
2)纯WSI方法。ABMIL[6]通过参数化注意力权重对区块特征做加权池化,将WSI压缩为单一向量后预测生存;DeepMISL[37]在注意力聚合的基础上引入深度多示例学习框架,对区块进行分层聚合以捕获更丰富的切片级表示;TransMIL[7]引入Transformer自注意力机制建模区块间的全局交互关系。上述方法均为单步聚合范式,缺乏显式的中间推理阶段。
3)多模态方法(WSI + 基因组)。上述3种纯WSI方法均有对应的“+Path”多模态变体,通过拼接WSI聚合特征与基因组通路特征实现早期融合。MCAT[9]设计了协同注意力机制,实现WSI形态特征与基因组通路特征的深度交互融合。SurvPath[10]以密集交互方式融合WSI token与生物通路token进行生存预测。CIMA[11]面向多模态不完整场景,通过循环模态重建模块相互填补缺失的病理与基因组数据,并以多粒度对齐显式建模WSI区块与生物通路之间的粗粒度与细粒度交互。SurvPGC[12]在SurvPath基础上引入双路径跨注意力融合结构,原文同时纳入临床信息文本提示词;本文为对齐统一的WSI与基因组输入条件,仅采用其病理与基因组双模态变体进行公平比较。
需要指出的是,PathRoute仅使用WSI图像输入,不依赖任何基因组数据,因此与纯WSI方法(ABMIL、DeepMISL、TransMIL)一样为同类可比方法,而与多模态方法的比较进一步刻画了结构化证据组织的相对作用。
为保证公平比较,所有WSI方法均使用相同的UNI2编码特征作为输入,所有多模态方法使用相同的基因组通路输入,并采用相同的患者级5折交叉验证划分。 在此条件下,表1比较的是不同模型结构对同一输入表征的利用能力。
表2展示了PathRoute与基线方法在5个癌种上的生存预测C-index对比。PathRoute在BRCA(0.680)、STAD(0.665)、COADREAD(0.717)、HNSC(0.646)四个癌种上取得最高C-index,5个癌种Overall为0.666。与最强的多模态基线SurvPGC(0.643)相比,Overall提升0.023,在HNSC、STAD、COADREAD上优势尤为明显,分别提升0.048,0.037,0.026;与最强的纯WSI基线TransMIL(0.604)相比,Overall提升0.062。值得注意的是,BRCA上PathRoute(0.680)仅小幅优于该癌种上最强的多模态基线MCAT(0.677),差距0.003,两者差距已在标准差范围内,这一结果与BRCA已有较成熟的形态学表型监督信号、信息增益空间有限相一致。上述结果表明,在统一输入条件下,结构化证据路由与方向性风险读出能够更有效地利用WSI表征完成生存预测;同时也说明,在强视觉基础表征(UNI2)支撑下,仅凭WSI本身即可提供较强的预后信号,不必需要额外的基因组通路输入作为补充。
在BLCA上,PathRoute(0.621)低于多模态方法SurvPGC(0.658),降低0.037。这与膀胱癌预后对基因组通路信息(如FGFR通路突变状态)的依赖较强结论一致,纯WSI输入难以完全恢复其分子层面的预后线索。
就结果方差而言,COADREAD的标准差在各方法中普遍偏高(PathRoute为±0.143),这与结直肠癌队列较小、5折划分时各折的终点事件分布不均衡有关:个别折的验证集中终点事件高度集中,风险排序难度显著高于其他折,从而拉大了折间C-index的波动幅度。其他癌种标准差相对较小(BRCA为±0.050、STAD为±0.075、HNSC为±0.075、BLCA为±0.073),整体结果可信。
为系统评估PathRoute不同设计对最终生存预测的贡献,本节将消融实验分为2组,如表2所示。组件消融关闭一个语义模块,但不改动其余模块的结构与监督,回答“哪些结构化语义模块是有效的”;结构消融保持4类查询及其监督全部就位,仅替换或扰动方向性注意力掩码、查询顺序、梯度截断等结构机制,回答“为何采用当前的信息流组织方式”。所有消融均沿用主实验的输入特征、训练超参数与 5 折交叉验证划分。下文按组件消融、结构消融、综合分析依次展开。
组件消融逐一关闭一个语义模块。w/o Tissue Census 移除组织组成感知分支,下游查询直接从视觉 token 中获取信息,用以检验全局组织学先验对共享表示的塑形作用。Random Routing 用均匀随机采样替代主动证据路由,但保持下游推理阶段的视觉序列长度不变,用以检验软注意力路由相对于无偏采样的增益。w/o Phenotype Supervision 保留表型查询及其位置,仅去除表型监督损失(受体状态、组织学分级、MSI 等标签均不使用),用以检验表型语义辅助监督对共享表示学习的帮助。Survival-Only 是合并消融,同时关闭组织组成感知、将证据路由替换为随机采样、并取消所有表型监督,使风险查询成为唯一受监督信号驱动的组件,估计上述3类辅助模块作为整体的贡献区间。
表2 显示,移除任一组件均会造成五癌种 Overall 平均 C-index 下降,说明组织组成感知、主动证据路由与表型辅助监督均对最终预测产生贡献,且贡献来源各不相同。w/o Tissue Census 对C-index带来平均1.6个百分点的下降,验证了全局组织学先验作为共享上下文的塑形作用。w/o Phenotype Supervision 带来最大的单组件下降(1.9 个百分点),且在乳腺癌(BRCA)这一组织异质性最高的癌种上下降最为显著(从 0.680 降至 0.658)。这与表型监督的设计目标相符:通过受体状态等癌种相关标签为共享推理表示注入临床先验,使学习信号不仅来自稀疏的生存事件。需要强调的是,表型监督的作用是训练阶段的多任务正则,而非风险预测的强制中介路径,这与方法节中关于方向性掩码“不强制风险经由表型语义中转”的论述一致。
相比之下,Random Routing 在五癌种C-index 的Overall 上仅造成平均 0.8 个百分点的下降,是3个单组件消融中降幅最小的,这一数字需结合3点解读。其一,Overall 平均掩盖了单癌种上的显著差异:在头颈癌(HNSC)上主动路由(0.646)较随机路由( 0.610)高3.6个百分点,难癌种上的稳定性收益不可忽略。其二,在 UNI2 强表征下随机采样已能给出有竞争力的排序,故主动路由的必要性主要体现在可解释的证据组织:3.6节展示主动路由能在 2 391 张切片上稳定将注意力质量自发集中于肿瘤(集中度中位数为2.30,Wilcoxon p<10−48),而随机路由集中度恒为 1且不产生任何可对齐肿瘤的证据聚合;一旦移除主动路由,本文3.6节与3.7节所依托的整个可解释性分析将不复成立。其三,如实而言,纯 C-index 维度上主动路由相对随机采样增益有限;PathRoute 将其定位为“在几乎不损失预测精度的前提下换取可审计的证据组织”,而非仅仅提升C-index 的机制。
Survival-Only 同时关闭3类辅助模块,带来平均2.7个百分点的下降,幅度大于任一单组件消融。这一结果表明3类辅助模块的贡献并非相互冗余,而是共同在共享表示上施加了来自不同语义层面的正则信号。
结构消融保留全部4类查询不变,仅扰动其实现机制中的某一结构约束。Block-Directional Mask 与 Fully Bidirectional Mask 分别将方向性注意力掩码松弛为“组内双向、跨组单向”和“完全双向”,对应文献中2类常见的多步推理掩码方案。Reordered Roles 保持掩码形式不变,但将表型查询物理上前置于组织组成感知查询之前,破坏组织先验在前向计算中的优先性。w/o Tissue Stop-Gradient 移除组织先验的梯度截断,允许生存损失沿组织先验回传至组织组成感知查询,用以检验梯度截断对维持组织组成感知作为可读组织学先验槽位是否必要。
表2 显示,结构消融的整体降幅小于组件消融,且不同癌种之间存在波动。Overall从 0.666 下降至 0.656~0.658,绝对差距处于跨折标准差范围内。这一观察与本文对方向性掩码的方法定位一致:方向性掩码的核心作用并非作为单纯的性能放大技巧,而是在前向计算层面规定不同语义角色的可见性方向,使风险查询成为前序证据表示的下游读出。
Reordered Roles 与 w/o Tissue Stop-Gradient 进一步表明,结构化查询的语义角色需要与其在推理流中的位置和监督边界保持一致。当组织先验被允许在生存损失下被反向更新,或被排在表型查询之后时,模型仍能学习到具有竞争力的风险预测,但其内部表示与“组织先验—诊断证据—表型语义—风险读出”的设计对应关系被削弱,组织组成感知作为可读组织学先验的槽位意义随之减弱。
综合2类消融可以看到,组件消融表明组织组成感知、主动证据路由与表型辅助监督均对最终性能产生独立而互补的贡献;结构消融表明方向性注意力掩码、查询顺序与梯度截断等机制主要承担方向性结构约束的功能,而非单纯的性能放大手段。2类消融共同支持了本文将 WSI 生存预测组织为“组织先验形成—诊断证据路由—表型语义注入—风险读出”两阶段结构化推理的设计思路。
本节定量考察主动证据路由学到的注意力分布与组织语义之间的对齐情况。值得强调的是,PathRoute 中的路由查询$ {\boldsymbol{Q}}_{{\mathrm{route}}} $本身并不接受任何直接的组织类型监督:尽管模型整体保留了组织组成感知的监督,但路由查询的优化信号仅来自下游生存损失。因此,路由分布是否对齐到诊断相关组织,可作为对所学证据语义的间接观察。
具体地,对每张测试切片,将路由 Transformer 输出的$ {Q}_{2} $个查询的注意力权重按查询维度平均,得到每个区块的归一化注意力权重$ {a}_{i} $;同时利用 CONCH 零样本伪标签$ {\ell}_{i}\in \{1,…,6\} $得到每个区块的组织类型。对每种组织类型$ t $,定义注意力集中度为
$ {c o n}{ {c}}_{ {t}}=\dfrac{\displaystyle\sum\limits_{ {i}}{ {a}}_{ {i}}\cdot {\mathbb{I}}[{{{\ell}}}_{ {i}}= {t}]}{\displaystyle\sum\limits_{ {i}}{\mathbb{I}}[{{\ell}}_{ {i}}= {t}]/{ {N}}_{ {{\mathrm{t i l e s}}}}}=\frac{ {a t t e n t i o n}\_ {s h a r}{ {e}}_{ {t}}}{ {t i l e}\_ {s h a r}{ {e}}_{ {t}}}, $
其中attention_sharet为注意力质量在组织类型$ t $上的占比,tile_sharet为该类型区块的数目占比。$ con{c}_{t}=1 $对应“路由对组织$ t $无偏好”,与均匀随机路由相同;$ con{c}_{t}>1 $表明路由主动将注意力集中到该组织。
表3 汇总了5个癌种共2391张切片上的每个切片的集中度中位数统计。所有5个癌种的肿瘤组织集中度中位数均显著大于 1(1.90~2.84),其余5种组织的集中度中位数均小于 1,且超过 90% 的切片其肿瘤集中度中位数大于 1。配对 Wilcoxon 符号秩检验拒绝“$ {{{{conc}}}}_{{\mathrm{tumor}}}\leq 1 $”零假设的$ p $值在所有癌种上均小于10−48。跨癌种合并2391 张切片:肿瘤集中度中位数为 2.30,93.8% 的切片肿瘤集中度大于 1;其余5种组织的集中度中位数均小于 1,路由主动冷落了非诊断性组织(如脂肪、淋巴细胞)。
这一现象在5个解剖部位完全不同的癌种上保持一致,表明主动证据路由在生存监督下学到了与肿瘤分布稳定对齐的注意力偏好,并非某一癌种的偶然产物。结合3.5节的消融结果可以看到,主动路由相对随机路由的 C-index 增益虽然有限,但其在证据选择层面的语义对齐为模型提供了端到端 C-index 之外的可解释价值。需要说明的是,本节分析依赖 CONCH 生成的组织伪标签而非人工精标注的肿瘤区域。主动证据路由与传统注意力池化(如 ABMIL)在信息利用与关键区域筛选上的机制差异及定量对照分析见附录D。CONCH 组织伪标签相对人工标注的可靠性验证见附录E。
3.6节 在 2391 张切片上定量考察了主动证据路由的语义对齐情况。本节进一步以一例 BRCA 切片(TCGA-3C-AALI)为代表,展示PathRoute 中4类语义查询在单次前向传播中分别产出的可读中间输出,从而具体说明“结构化证据组织”在临床推理层面的可解释性体现。
图4 整合了该切片的可解释性快照。图4(a)(b)呈现:每个区块按 CONCH 伪标签上色(颜色映射见右上“组织类型”图例),上下文采样区块以白框标记,路由 top-20 区块以阶梯化标记叠加在叠加图上,反映了不同排名查询所聚焦的区域(标记说明见右上“路由标记”图例)。图4(c)~(f)呈现4类查询的中间输出。
1)组织组成感知。 给出 8 个组织查询各自最确定的组织类型预测及其置信度。在该切片上,多数查询同时识别出肿瘤、坏死、基质、淋巴细胞等典型组织类型,且组织类型预测在查询$ {q}_{0} $~$ {q}_{7} $之间相互分化、未发生坍缩,与多样性正则的设计目标一致。
2)主动证据路由。给出全切片范围内每种组织的“路由注意力占比”与“区块数量占比”的对照:肿瘤组织以 35% 的区块数量贡献了 94% 的路由注意力,相应过表达倍数约 2.7;其余组织(基质、坏死、脂肪等)的注意力占比则普遍低于其区块占比。这与3.6节 在 2391 张切片上的统计结论一致,并具体地展示了路由对该切片诊断证据的聚焦行为。
3)表型语义辅助监督。给出 ER/HER2/PR 三个临床受体状态的预测:在该例上,模型给出 ER+(52%)、HER2+(72%)、PR$ - $(94%)的预测,对应 PAM50 推断标签下的真实表型。表型查询的中间输出与PathRoute 整体判断之间通过方向性掩码保持单向可见性,因此该输出可作为独立可读的临床先验接口供下游审计。
4)风险读出。综合前序信息给出最终的临床判断:病理分期、分子亚型、可能突变靶点(如 TP53)以及生存风险等级。在该例上,风险查询给出 HIGH 风险等级(score =$ -2.18 $),与该患者的临床随访结果一致。
总体来看,PathRoute 的4类查询在单次前向传播中各自产出独立可读、独立可监督的中间表示:组织组成感知确立全局组织学先验;主动证据路由聚焦诊断相关区块;表型语义提供受体状态等临床先验;风险查询完成最终预后判断。这一“设计时可解释”的中间表示链条,使模型的内部信息流不必依赖事后注意力可视化即可被审计。
本文提出PathRoute,一个面向 WSI 生存预测的结构化证据组织框架。其核心由2部分构成:独立的路由 Transformer 通过软注意力主动从全量区块中聚合诊断证据,使下游推理的序列长度与切片大小解耦;推理 Transformer 通过方向性注意力掩码约束组织组成感知、表型监督与风险读出之间的前向可见性方向,使风险预测在架构上成为前序证据表示的下游读出。
在5个 TCGA 癌种的统一特征输入与相同患者级 5 折交叉验证划分下,PathRoute 仅以 WSI 作为输入即在五癌种C-index上的Overall优于代表性多模态融合方法与代表性纯 WSI 方法。进一步的语义对齐分析表明,主动证据路由在仅由生存损失驱动的条件下,仍在数千张切片上跨5种解剖部位完全不同的癌种稳定地将注意力质量自发集中于肿瘤组织区域,为结构化证据组织的有效性提供了直接的可解释性证据。
本文工作的局限主要体现在2个方面:1)所有实验均在 TCGA 数据上进行,跨机构、跨扫描仪的泛化能力仍有待外部多中心数据集评估。2)PathRoute 当前不消费基因组与转录组信号,在分子层信息占主导的癌种(如 BLCA)上其性能上限受限于纯影像信息的可恢复程度;未来可在保留结构化证据组织的前提下,将基因组与文本临床信息以可控方式接入,扩展框架的多模态可用性。在更广阔的方向上,结构化证据路由可推广至分子标志物预测、治疗响应预测等其他病理任务,推动计算病理方法向辅助诊断工具的转化。

参考文献 引证文献
排序方式:
1
Chen R J, Ding T, Lu M Y, et al. Towards a general-purpose foundation model for computational pathology[J]. Nature Medicine, 2024, 30(3): 850−862
2
MahmoodLab. UNI2-h: An updated pathology foundation model [EB/OL]. 2025 [2026-07-10]. https://huggingface.co/MahmoodLab/UNI2-h
3
Zimmermann E, Vorontsov E, Viret J, et al. Virchow2: Scaling self-supervised mixed magnification models in pathology [EB/OL]. 2024 [2026-07-10]. https://arxiv.org/abs/2408.00738
4
Lu M Y, Chen B, Williamson D F K, et al. A visual-language foundation model for computational pathology[J]. Nature Medicine, 2024, 30(3): 863−874
5
Ma J, Guo Z, Zhou F, et al. A generalizable pathology foundation model using a unified knowledge distillation pretraining framework[J]. Nature Biomedical Engineering, 2026, 10: 545−564
6
Ilse M, Tomczak J M, Welling M. Attention-based deep multiple instance learning [C] //Proc of the 35th Int Conf on Machine Learning. Cambridge, MA: PMLR, 2018: 2127−2136
7
Shao Z, Bian H, Chen Y, et al. TransMIL: Transformer based correlated multiple instance learning for whole slide image classification [C] //Advances in Neural Information Processing Systems. Red Hook, NY: Curran Associates, 2021, 34: 2136−2147
8
Zhang H, Meng Y, Zhao Y, et al. DTFD-MIL: Double-tier feature distillation multiple instance learning for histopathology whole slide image classification [C] //Proc of the IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2022: 18802−18812
9
Chen R J, Lu M Y, Weng W H, et al. Multimodal co-attention transformer for survival prediction in gigapixel whole slide images [C] //Proc of the IEEE/CVF Int Conf on Computer Vision. Piscataway, NJ: IEEE, 2021: 4015−4025
10
Jaume G, Vaidya A, Chen R J, et al. Modeling dense multimodal interactions between biological pathways and histology for survival prediction [C] //Proc of the IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2024: 11579−11590
11
Bu Y, Niu Q, Li Z, et al. Cancer survival prediction by cyclic generation and multi-grained alignment [C] //Proc of the AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2026, 40(24): 19781−19789
12
Hou J, Zhang R, Xie Y, et al. Multimodal deep learning for cancer prognosis prediction with clinical information prompts integration[J]. npj Digital Medicine, 2026, 9: 76
13
Oquab M, Darcet T, Moutakanni T, et al. DINOv2: Learning robust visual features without supervision[J/OL]. Transactions on Machine Learning Research, 2024 [2026-07-16]. https://openreview.net/forum?id=a68SUt6zFt
14
Mobadersany P, Yousefi S, Amgad M, et al. Predicting cancer outcomes from histology and genomics using convolutional networks[J]. Proceedings of the National Academy of Sciences, 2018, 115(13): E2970−E2979
15
Cox D R. Regression models and life-tables[J]. Journal of the Royal Statistical Society: Series B (Methodological), 1972, 34(2): 187−202
16
Kather J N, Heij L R, Grabsch H I, et al. Pan-cancer image-based detection of clinically actionable genetic alterations[J]. Nature Cancer, 2020, 1(8): 789−799
17
王梅霞, 王慧颖, 黄苗苗, 等. 多示例多标记学习综述[J]. 计算机研究与发展, 2026, 63(1): 66−89
Wang Meixia, Wang Huiying, Huang Miaomiao, et al. Survey on multi-instance multi-label learning[J]. Journal of Computer Research and Development, 2026, 63(1): 66−89 (in Chinese)
18
Campanella G, Hanna M G, Geneslaw L, et al. Clinical-grade computational pathology using weakly supervised deep learning on whole slide images[J]. Nature Medicine, 2019, 25(8): 1301−1309
19
丁世飞, 孙玉婷, 梁志贞, 等. 弱监督场景下的支持向量机算法综述[J]. 计算机学报, 2024, 47(5): 987−1009
Ding Shifei, Sun Yuting, Liang Zhizhen, et al. Survey on support vector machine algorithms in weakly supervised scenarios[J]. Chinese Journal of Computers, 2024, 47(5): 987−1009 (in Chinese)
20
Lu M Y, Williamson D F K, Chen T Y, et al. Data-efficient and weakly supervised computational pathology on whole-slide images[J]. Nature Biomedical Engineering, 2021, 5(6): 555−570
21
Li B, Li Y, Eliceiri K W. Dual-stream multiple instance learning network for whole slide image classification with self-supervised contrastive learning [C] //Proc of the IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2021: 14318−14328
22
Carion N, Massa F, Synnaeve G, et al. End-to-end object detection with transformers [G] //LNCS 12346: Proc of the European Conf on Computer Vision. Cham: Springer, 2020: 213−229
23
李文生, 张菁, 卓力, 等. 基于Transformer的视觉分割技术进展[J]. 计算机学报, 2024, 47(12): 2760−2782
Li Wensheng, Zhang Jing, Zhuo Li, et al. Overview of Transformer-based visual segmentation[J]. Chinese Journal of Computers, 2024, 47(12): 2760−2782 (in Chinese)
24
Jaegle A, Gimeno F, Brock A, et al. Perceiver: General perception with iterative attention [C] //Proc of the Int Conf on Machine Learning. Cambridge, MA: PMLR, 2021: 4651−4664
25
Jaegle A, Borgeaud S, Alayrac J B, et al. Perceiver IO: A general architecture for structured inputs and outputs[C/OL]// Proc of the 10th Int Conf on Learning Representations. Amherst, MA: OpenReview. net, 2022 [2026-07-16]. https://openreview.net/forum?id=fILj7WpI-g
26
Li J, Li D, Savarese S, et al. BLIP-2: Bootstrapping language-image pre-training with frozen image encoders and large language models [C] //Proc of the Int Conf on Machine Learning. Cambridge, MA: PMLR, 2023: 19730−19742
27
曾诚, 葛云洁, 赵令辰, 等. 多模态视觉语言表征学习模型及其对抗样本攻防技术综述[J]. 计算机研究与发展, 2025, 62(9): 2208−2232
Zeng Cheng, Ge Yunjie, Zhao Lingchen, et al. Survey of multimodal vision-language representation learning models and their adversarial examples attack and defense techniques[J]. Journal of Computer Research and Development, 2025, 62(9): 2208−2232 (in Chinese)
28
Zhang J, Kapse S, Ma K, et al. Prompt-MIL: Boosting multi-instance learning schemes via task-specific prompt tuning [C] //Proc of the Int Conf on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer, 2023: 624−634
29
Selvaraju R R, Cogswell M, Das A, et al. Grad-CAM: Visual explanations from deep networks via gradient-based localization [C] //Proc of the IEEE Int Conf on Computer Vision. Piscataway, NJ: IEEE, 2017: 618−626
30
Hashimoto N, Fukushima D, Koga R, et al. Multi-scale domain-adversarial multiple-instance CNN for cancer subtype classification with unannotated histopathological images [C] //Proc of the IEEE/CVF Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 3852−3861
31
Koh P W, Nguyen T, Tang Y S, et al. Concept bottleneck models [C] //Proc of the Int Conf on Machine Learning. Cambridge, MA: PMLR, 2020: 5338−5348
32
Kuhn H W. The Hungarian method for the assignment problem[J]. Naval Research Logistics Quarterly, 1955, 2(1-2): 83−97
33
Zadeh S G, Schmid M. Bias in cross-entropy-based training of deep survival networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(9): 3126−3137
34
Hutter C, Zenklusen J C. The cancer genome atlas: Creating lasting value beyond its data[J]. Cell, 2018, 173(2): 283−285
35
Harrell F E, Califf R M, Pryor D B, et al. Evaluating the yield of medical tests[J]. JAMA, 1982, 247(18): 2543−2546
36
Klambauer G, Unterthiner T, Mayr A, et al. Self-normalizing neural networks [C] //Advances in Neural Information Processing Systems. Red Hook, NY: Curran Associates, 2017, 30: 971−980
37
Yao J, Zhu X, Jonnagaddala J, et al. Whole slide images based cancer survival prediction using attention guided deep multiple instance learning networks[J]. Medical Image Analysis, 2020, 65: 101789
2026年第63卷第9期
91
45
0
引用本文
BibTeX
文章信息
doi: 10.7544/issn1000-1239.202660468
  • 接收时间:2026-06-05
  • 首发时间:2026-11-17
  • 出版时间:2026-09-01
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2026-06-05
  • 修回日期:2026-07-17
基金
作者信息
    1山东大学软件学院 济南 250101
    2山东大学-南洋理工大学人工智能国际联合研究院 济南 250101
    3山东大学齐鲁医院 济南 250012

通讯作者:

余国先()
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/10.7544/issn1000-1239.202660468
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
科
Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)
属
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