Graph-Based Learning and Multimodal Learning for Colon Disease Classification: An Interpretable Study using CNN-GNN Pipelines and Vision-Language Models

Graph-Based Learning and Multimodal Learning for Colon Disease Classification: An Interpretable Study using CNN-GNN Pipelines and Vision-Language Models | 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 Research Article Graph-Based Learning and Multimodal Learning for Colon Disease Classification: An Interpretable Study using CNN-GNN Pipelines and Vision-Language Models Shahriar Sultan. Ramit, Alaya Parven. Alo, Md. Sadekur Rahman, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9038592/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Colorectal cancer (CRC) is a significant health issue in the world that requires the use of improved diagnostic instruments to detect it at an early and precise stage. In this paper, the researcher proposes an interpretable classification of colon diseases based on endoscopic images of the Kvasir V2 data set. Each image was subjected to a systematic preprocessing pipeline prior to being model trained to be consistent and better represent features. The size of the images was reduced to 224 224 pixels to match the specifications of deep learning inputs. Pixel intensities were brought to a stable value to enable convergence and contrast enhancement was used to improve the visibility of the mucosal textures. The edge sharpening methods included unsharp masking and Laplacian filter as a way of emphasizing structural boundaries and highlighted the edges of lesions and polyp margins. To augment the data, random rotation and flips, zoom scaling, and light articulations were introduced to enhance the diversity of the data as well as reduce overfitting and enhance resistance to real-world variation in colonoscopy imaging. In this paper, a hybrid pipeline consisting of CNNs and GNNs is suggested to extract visual features and model relational dependencies and Vision-Language Models (VLMs), which combines Vision Transformer (ViT) with BERT to learn across multiple modalities. It tested various methods of creating graphs (cosine similarity, ε-radius, k-nearest neighbors) and GNN models (GCN, GAT, GraphSAGE, GIN) and reached the highest accuracy of 91 percent with ViT + Epsilon + GIN. Tuned ViT-BERT performed the best with 95.17% accuracy and 0.95 F1-score. Grad-CAM visualizations have further improvements in interpretability as they demonstrate clinically-relevant areas of pictures, which place the framework as a strong, interpretable, and transparent instrument of automated CRC diagnostics in various clinical environments. Colorectal cancer Graph Neural Networks (GNN) Vision-Language Models (VLM) Vision Transformer (ViT) BERT Kvasir V2 dataset Grad-CAM interpretability endoscopy Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 09 Apr, 2026 Reviews received at journal 08 Apr, 2026 Reviewers agreed at journal 08 Apr, 2026 Reviewers invited by journal 06 Apr, 2026 Editor assigned by journal 11 Mar, 2026 Submission checks completed at journal 11 Mar, 2026 First submitted to journal 05 Mar, 2026 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-9038592","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":620364461,"identity":"548d6939-12db-416c-9308-85c007c780d2","order_by":0,"name":"Shahriar Sultan. Ramit","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABFUlEQVRIie3Qv0vDQBTA8RcC1+XZOcWS/AsvFCzF4N+SIiRLSutSBAcPhLoE5xT/ibo4nzywS6lrBoeC4FwolA5VvHSUJOrmcF+45XEf7geAyfQfc0AAEIDwFNAaAxe4GB+GYT1pOgAXWTvq/JLoXE02GHBfqh+Id3/z/nY5OhuKY2nNMIniuzkSrMcM3UY5odfnrr+g896krWxyFsFgykhWtmTopaty4oQnLUk2CRhuyU+jwUwT+2jCQHnFxbJ4q8m1JmFj1//kmAryUUMgT4pTmIQTAink8ECsGkJ5MtZkTgIV+BIjf8pi9JQuY6x6i77YY0vur8i7ldDZY+A1X/hhtRufulU/Vp7SC/+w32QymUzf+gLUf1yZmWRgXgAAAABJRU5ErkJggg==","orcid":"","institution":"Daffodil International University","correspondingAuthor":true,"prefix":"","firstName":"Shahriar","middleName":"Sultan.","lastName":"Ramit","suffix":""},{"id":620364463,"identity":"8071278b-21bb-4bd6-8616-a5a926877662","order_by":1,"name":"Alaya Parven. Alo","email":"","orcid":"","institution":"Daffodil International University","correspondingAuthor":false,"prefix":"","firstName":"Alaya","middleName":"Parven.","lastName":"Alo","suffix":""},{"id":620364465,"identity":"48953922-5808-4fac-8d29-e865321ec0b6","order_by":2,"name":"Md. Sadekur Rahman","email":"","orcid":"","institution":"Daffodil International University","correspondingAuthor":false,"prefix":"","firstName":"Md.","middleName":"Sadekur","lastName":"Rahman","suffix":""},{"id":620364466,"identity":"45fad7f5-0ce6-4dcf-8693-f0601fb9add1","order_by":3,"name":"Masud Rana Rashel","email":"","orcid":"","institution":"Daffodil International University","correspondingAuthor":false,"prefix":"","firstName":"Masud","middleName":"Rana","lastName":"Rashel","suffix":""},{"id":620364468,"identity":"c02b8c9a-8e85-490f-984a-74e0211f20a3","order_by":4,"name":"A. K.M. Kamrul Islam","email":"","orcid":"","institution":"Tigerlabs","correspondingAuthor":false,"prefix":"","firstName":"A.","middleName":"K.M. Kamrul","lastName":"Islam","suffix":""}],"badges":[],"createdAt":"2026-03-05 09:39:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9038592/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9038592/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106727793,"identity":"0fcd41b3-df3c-413c-ac0a-80d1fde5f90d","added_by":"auto","created_at":"2026-04-12 18:40:53","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1445137,"visible":true,"origin":"","legend":"","description":"","filename":"colonpaper5.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9038592/v1_covered_2fa86607-adb0-4988-8a3e-c12d6dd679be.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Graph-Based Learning and Multimodal Learning for Colon Disease Classification: An Interpretable Study using CNN-GNN Pipelines and Vision-Language Models","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Colorectal cancer, Graph Neural Networks (GNN), Vision-Language Models (VLM), Vision Transformer (ViT), BERT, Kvasir V2 dataset, Grad-CAM, interpretability, endoscopy","lastPublishedDoi":"10.21203/rs.3.rs-9038592/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9038592/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eColorectal cancer (CRC) is a significant health issue in the world that requires the use of improved diagnostic instruments to detect it at an early and precise stage. In this paper, the researcher proposes an interpretable classification of colon diseases based on endoscopic images of the Kvasir V2 data set. Each image was subjected to a systematic preprocessing pipeline prior to being model trained to be consistent and better represent features. The size of the images was reduced to 224 224 pixels to match the specifications of deep learning inputs. Pixel intensities were brought to a stable value to enable convergence and contrast enhancement was used to improve the visibility of the mucosal textures. The edge sharpening methods included unsharp masking and Laplacian filter as a way of emphasizing structural boundaries and highlighted the edges of lesions and polyp margins. To augment the data, random rotation and flips, zoom scaling, and light articulations were introduced to enhance the diversity of the data as well as reduce overfitting and enhance resistance to real-world variation in colonoscopy imaging. In this paper, a hybrid pipeline consisting of CNNs and GNNs is suggested to extract visual features and model relational dependencies and Vision-Language Models (VLMs), which combines Vision Transformer (ViT) with BERT to learn across multiple modalities. It tested various methods of creating graphs (cosine similarity, ε-radius, k-nearest neighbors) and GNN models (GCN, GAT, GraphSAGE, GIN) and reached the highest accuracy of 91 percent with ViT\u0026thinsp;+\u0026thinsp;Epsilon\u0026thinsp;+\u0026thinsp;GIN. Tuned ViT-BERT performed the best with 95.17% accuracy and 0.95 F1-score. Grad-CAM visualizations have further improvements in interpretability as they demonstrate clinically-relevant areas of pictures, which place the framework as a strong, interpretable, and transparent instrument of automated CRC diagnostics in various clinical environments.\u003c/p\u003e","manuscriptTitle":"Graph-Based Learning and Multimodal Learning for Colon Disease Classification: An Interpretable Study using CNN-GNN Pipelines and Vision-Language Models","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-10 20:48:17","doi":"10.21203/rs.3.rs-9038592/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"282178628109173486861620173838153089166","date":"2026-04-09T13:30:45+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-09T02:56:53+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"259251580947532640674160644841766186311","date":"2026-04-09T02:45:26+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-06T11:33:15+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-12T02:11:18+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-12T02:10:40+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Artificial Intelligence","date":"2026-03-05T09:29:14+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2586d42f-65ba-4610-82aa-5afb77eff76d","owner":[],"postedDate":"April 10th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-10T20:48:17+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-10 20:48:17","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9038592","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9038592","identity":"rs-9038592","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}