Foundation Model-Architected Sparse Dictionaries for Fully Quantitative Interpretability in Whole Slide Image Decoding

Foundation Model-Architected Sparse Dictionaries for Fully Quantitative Interpretability in Whole Slide Image Decoding | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Foundation Model-Architected Sparse Dictionaries for Fully Quantitative Interpretability in Whole Slide Image Decoding Jinhua Yu, GuoQing Wu, Tianyi Pan, Hanning Xu, Ye Tang, Xuan Xie, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8542624/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Artificial intelligence (AI)-based pathological image analysis has achieved remarkable progress. However, a critical limitation persists: beyond attention-based visualization methods—which offer only partial interpretability—existing AI approaches fail to provide quantifiable diagnostic evidence that meets clinically actionable standards and aligns with pathologists' domain expertise. Consequently, the interpretability gap between AI diagnostics and pathologists impedes clinical adoption. To resolve this impasse, we propose PATH-SparseD (Pathology-Aware Transformer–Hybridized Sparse Dictionary Learning), a framework that reformulates AI-driven diagnostic inference as a dictionary-query process. Central to PATH-SparseD is the novel integration of foundation models with sparse representation theory. This synergy empowers the model to deconstruct the highly complex and heterogeneous data of a WSI into a sparse combination of fundamental, clinically meaningful visual "primitives"—each acting as an atomic diagnostic unit. Specifically, for any WSI tile, PATH-SparseD does not merely extract features; it encodes the tile by identifying and activating a minimal set of these primitives from a multi-scale dictionary. This process effectively reformulates the entire WSI as a quantifiable histogram of primitive occurrences, creating a transparent and pattern-based transcript of the tissue that directly mirrors a pathologist's diagnostic reasoning. Extensive experiments on 10 datasets covering 8 tumor types demonstrated that PATH-SparseD not only provides an interpretable and quantitatively verifiable pathological analysis framework recognized by clinical pathologists, but also significantly outperforms state-of-the-art foundation-model approaches, yielding accuracy improvements of 5.6% in glioma grading, 4.4% in IDH1 genotyping, 14.5% in TNM grading, 17.0% in tumor differentiation, 11.8% in cellular origin, and 8.6% in organ origin. Ultimately, PATH-SparseD establishes a novel paradigm for WSI decoding that harnesses the performance advantages of foundation models while providing an effective solution to the interpretability challenge. Biological sciences/Cancer/Cancer imaging Biological sciences/Cancer/Cancer microenvironment Full Text Additional Declarations There is NO Competing Interest. Supplementary Files Supplement.docx Supplementary material Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8542624","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":578684712,"identity":"402d4bb4-3cd2-40e5-b7ce-c8c2766c101b","order_by":0,"name":"Jinhua Yu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyElEQVRIiWNgGAWjYBACxhlA4oMNA2MDiMdDrBbGGWmkaGGQYGBg5iFJC/PsHjNpm4TDsv2zGxgfvG1jkDcn6LA5Z8ykcxIOG8+4c4DZcG4bg+HOBkJaZuSYSef+OJzYcCOBTZq3jSHB4AAxWiwSDifOv5HA/pt4LQxALRuAtjATqSWt2LInId14443EZsk55yQMNxDSYjgjeeONHwnWsvNuJB/88KbMRp6gLYYNHAYwCxsYwNFECMgzsD8grGoUjIJRMApGNgAAsZlBY0N1z7QAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0002-0654-6034","institution":"Fudan University","correspondingAuthor":true,"prefix":"","firstName":"Jinhua","middleName":"","lastName":"Yu","suffix":""},{"id":578684713,"identity":"8819e5c5-0360-459e-b723-8a7de15a0f22","order_by":1,"name":"GuoQing Wu","email":"","orcid":"","institution":"Fudan University","correspondingAuthor":false,"prefix":"","firstName":"GuoQing","middleName":"","lastName":"Wu","suffix":""},{"id":578684714,"identity":"7ace6788-b75c-4fed-9979-ff06556578d8","order_by":2,"name":"Tianyi Pan","email":"","orcid":"","institution":"the College of Biomedical Engineering, FudanUniversity, Shanghai, China.","correspondingAuthor":false,"prefix":"","firstName":"Tianyi","middleName":"","lastName":"Pan","suffix":""},{"id":578684715,"identity":"4cb34020-313e-4601-8b17-d110c1fde88a","order_by":3,"name":"Hanning Xu","email":"","orcid":"","institution":"the College of Biomedical Engineering, FudanUniversity, Shanghai, China.","correspondingAuthor":false,"prefix":"","firstName":"Hanning","middleName":"","lastName":"Xu","suffix":""},{"id":578684716,"identity":"09d43acc-7d6c-4f88-bb09-d1dc38bd7be5","order_by":4,"name":"Ye Tang","email":"","orcid":"","institution":"the College of Biomedical Engineering, FudanUniversity, Shanghai, China.","correspondingAuthor":false,"prefix":"","firstName":"Ye","middleName":"","lastName":"Tang","suffix":""},{"id":578684717,"identity":"9c09e40d-e7af-4fef-8e38-f3ab4b4d0389","order_by":5,"name":"Xuan Xie","email":"","orcid":"","institution":"School of Information Science and Technology, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Xuan","middleName":"","lastName":"Xie","suffix":""},{"id":578684719,"identity":"85d71a42-228c-400b-a641-c03b4df75144","order_by":6,"name":"Chengqian Zhao","email":"","orcid":"","institution":"School of Biomedical Engineering and Technological Innovation, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Chengqian","middleName":"","lastName":"Zhao","suffix":""},{"id":578684722,"identity":"af160379-9daa-4c38-82ba-8f3361ded755","order_by":7,"name":"Feiyu Yin","email":"","orcid":"","institution":"School of Biomedical Engineering and Technological Innovation, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Feiyu","middleName":"","lastName":"Yin","suffix":""},{"id":578684723,"identity":"8ba3f7d8-d53f-44e2-9a6b-ca4383809e43","order_by":8,"name":"Pengfei Song","email":"","orcid":"","institution":"School of Information Science and Technology, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Pengfei","middleName":"","lastName":"Song","suffix":""},{"id":578684726,"identity":"dc9c2750-eee2-475e-bc57-dd096ba8a7ad","order_by":9,"name":"Ji Xiong","email":"","orcid":"","institution":"Huashan Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ji","middleName":"","lastName":"Xiong","suffix":""},{"id":578684730,"identity":"1f2c93a2-6da9-4f92-903a-c29512ed3946","order_by":10,"name":"Zhifeng Shi","email":"","orcid":"","institution":"Department of Neurosurgery, Huashan Hospital of Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Zhifeng","middleName":"","lastName":"Shi","suffix":""},{"id":578684734,"identity":"e40cb9d3-b88c-4e2f-9601-0c6b40099a2a","order_by":11,"name":"Ying Liu","email":"","orcid":"","institution":"Department of Pathology, School of Basic Medical Sciences, Fudan University.","correspondingAuthor":false,"prefix":"","firstName":"Ying","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2026-01-07 14:26:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8542624/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8542624/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102801563,"identity":"d3b4a286-e40b-40dc-9270-aaf29bb21c40","added_by":"auto","created_at":"2026-02-16 21:10:48","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2104307,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8542624/v1_covered_703149e8-777d-4654-8444-5dabb416eb91.pdf"},{"id":102424007,"identity":"3742bcad-b4e1-4baa-99dd-6b5e25f49c1a","added_by":"auto","created_at":"2026-02-11 14:19:32","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":14493581,"visible":true,"origin":"","legend":"Supplementary material","description":"","filename":"Supplement.docx","url":"https://assets-eu.researchsquare.com/files/rs-8542624/v1/ca6582c83170abbeb75554af.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Foundation Model-Architected Sparse Dictionaries for Fully Quantitative Interpretability in Whole Slide Image Decoding","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8542624/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8542624/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Artificial intelligence (AI)-based pathological image analysis has achieved remarkable progress. However, a critical limitation persists: beyond attention-based visualization methods—which offer only partial interpretability—existing AI approaches fail to provide quantifiable diagnostic evidence that meets clinically actionable standards and aligns with pathologists' domain expertise. Consequently, the interpretability gap between AI diagnostics and pathologists impedes clinical adoption. To resolve this impasse, we propose PATH-SparseD (Pathology-Aware Transformer–Hybridized Sparse Dictionary Learning), a framework that reformulates AI-driven diagnostic inference as a dictionary-query process. Central to PATH-SparseD is the novel integration of foundation models with sparse representation theory. This synergy empowers the model to deconstruct the highly complex and heterogeneous data of a WSI into a sparse combination of fundamental, clinically meaningful visual \"primitives\"—each acting as an atomic diagnostic unit. Specifically, for any WSI tile, PATH-SparseD does not merely extract features; it encodes the tile by identifying and activating a minimal set of these primitives from a multi-scale dictionary. This process effectively reformulates the entire WSI as a quantifiable histogram of primitive occurrences, creating a transparent and pattern-based transcript of the tissue that directly mirrors a pathologist's diagnostic reasoning. Extensive experiments on 10 datasets covering 8 tumor types demonstrated that PATH-SparseD not only provides an interpretable and quantitatively verifiable pathological analysis framework recognized by clinical pathologists, but also significantly outperforms state-of-the-art foundation-model approaches, yielding accuracy improvements of 5.6% in glioma grading, 4.4% in IDH1 genotyping, 14.5% in TNM grading, 17.0% in tumor differentiation, 11.8% in cellular origin, and 8.6% in organ origin. Ultimately, PATH-SparseD establishes a novel paradigm for WSI decoding that harnesses the performance advantages of foundation models while providing an effective solution to the interpretability challenge.","manuscriptTitle":"Foundation Model-Architected Sparse Dictionaries for Fully Quantitative Interpretability in Whole Slide Image Decoding","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-11 14:19:22","doi":"10.21203/rs.3.rs-8542624/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"79d497b9-9def-4f53-8fca-34aa2c9dbbbc","owner":[],"postedDate":"February 11th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":61572859,"name":"Biological sciences/Cancer/Cancer imaging"},{"id":61572860,"name":"Biological sciences/Cancer/Cancer microenvironment"}],"tags":[],"updatedAt":"2026-02-16T21:10:35+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-11 14:19:22","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8542624","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8542624","identity":"rs-8542624","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}