Multi-Level Attention and Boundary Refinement Network for Polyp Segmentation

Multi-Level Attention and Boundary Refinement Network for Polyp Segmentation | 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 Multi-Level Attention and Boundary Refinement Network for Polyp Segmentation Kaihui Dong, Peng Duan, Jinjiang Li This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6435376/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 In recent years, deep neural networks have made significant progress in automatic segmentation of colorectal polyps, overcoming the limitations of visual inspection. However, there are still two major challenges: first, it is difficult to achieve high-precision segmentation in uncertain regions for polyps with irregular and fuzzy boundaries in colonoscopy images; second, lesions and surrounding tissues in medical images show complex co-occurrence phenomena, which leads to fuzzy boundaries and structural confusion, thus affecting the segmentation accuracy. To cope with this problem, this paper proposes a deep segmentation network, MABR-Net, based on multilevel attention and boundary optimization. Specifically, we first use it to enhance multi-scale perception through the Modified Receptive Field Block (MRFB); subsequently, in order to optimize the uncertain boundary regions, we propose the Dynamic Boundary Refinement Module (DBR) to refine and enhance it for the fuzzy or uncertain edge regions. Meanwhile, we design the Multi-Level Attention Decoder (MLAD) to further enhance the attention to the target region through horizontal-vertical viewgraph decoupling and orientation-aware self-attention mechanism. Finally, the optimized features at all levels are integrated through the Multi-Scale Feature Fusion Module (MSFM) to gradually achieve segmentation prediction from coarse to fine. The MABR-Net model is trained using deep learning techniques on high-resolution colonoscopy images, with the training process conducted in a multi-GPU server environment to accelerate the learning process. Through parallel processing and distributed computation, the model is trained across multiple GPUs, handling large-scale data and ensuring real-time segmentation performance. The experimental results show that MABR-Net performs excellently in segmenting polyps and other lesions in endoscopic images, which not only significantly improves the edge segmentation accuracy, but also enhances the robustness and generalization ability of the model to complex structures under co-occurrence. This provides a novel and efficient technological path to solve the co-occurrence problem in medical imaging, and provides strong support for early clinical diagnosis and treatment planning. Code:https://github.com/User-deepll/MABR-Net. Polyp segmentation Deep learning Dynamic Boundary Refinement Multi-level attention Full Text Additional Declarations No competing interests reported. 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-6435376","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":553889720,"identity":"26afdb26-538d-4b37-9ca6-df27476929f6","order_by":0,"name":"Kaihui Dong","email":"","orcid":"","institution":"Shandong Technology and Business University","correspondingAuthor":false,"prefix":"","firstName":"Kaihui","middleName":"","lastName":"Dong","suffix":""},{"id":553889721,"identity":"e0a2db9f-5d0b-4e16-8d74-42fe8e1f6090","order_by":1,"name":"Peng Duan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6UlEQVRIiWNgGAWjYDCCA1Can4HBgBlFhKAWyQYGw2bStBgcIFYL343kZ48LKu7YbT5/ePvjwjYGOb4bCYyfC/BokbyRZm4848yz5G030gqbZ7YxGEveSGCWnoFHi8GNBDNp3rbDyWY3eAybedsYEjfcSGBj5sGrJf2bNO+/w8nG/WfAWuqJ0JIDtKXhsJ0BQw5YS4IBIS2SZ96USc84djhBAuiX2TPOSRjOPPOwWRqfFr7j6dukC2oO2/P3H97wuaDMRp7vePLBz/i0gAAo0hMbIGwJIGZsIKABosWeoKpRMApGwSgYuQAANe9RNjlaLkkAAAAASUVORK5CYII=","orcid":"","institution":"Shandong Technology and Business University","correspondingAuthor":true,"prefix":"","firstName":"Peng","middleName":"","lastName":"Duan","suffix":""},{"id":553889722,"identity":"474c5b43-88b6-473d-ae0b-54c582631568","order_by":2,"name":"Jinjiang Li","email":"","orcid":"","institution":"Shandong Technology and Business University","correspondingAuthor":false,"prefix":"","firstName":"Jinjiang","middleName":"","lastName":"Li","suffix":""}],"badges":[],"createdAt":"2025-04-12 15:23:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6435376/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6435376/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":97369614,"identity":"486a5e72-c400-48b2-86d3-f7df08dc5025","added_by":"auto","created_at":"2025-12-03 16:25:19","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2897669,"visible":true,"origin":"","legend":"","description":"","filename":"snarticletemplate.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6435376/v1_covered_b3d7ec40-2774-4c5d-920b-23e2a579ae48.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Multi-Level Attention and Boundary Refinement Network for Polyp Segmentation","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":"Polyp segmentation, Deep learning, Dynamic Boundary Refinement, Multi-level attention","lastPublishedDoi":"10.21203/rs.3.rs-6435376/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6435376/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn recent years, deep neural networks have made significant progress in automatic segmentation of colorectal polyps, overcoming the limitations of visual inspection. However, there are still two major challenges: first, it is difficult to achieve high-precision segmentation in uncertain regions for polyps with irregular and fuzzy boundaries in colonoscopy images; second, lesions and surrounding tissues in medical images show complex co-occurrence phenomena, which leads to fuzzy boundaries and structural confusion, thus affecting the segmentation accuracy. To cope with this problem, this paper proposes a deep segmentation network, MABR-Net, based on multilevel attention and boundary optimization. Specifically, we first use it to enhance multi-scale perception through the Modified Receptive Field Block (MRFB); subsequently, in order to optimize the uncertain boundary regions, we propose the Dynamic Boundary Refinement Module (DBR) to refine and enhance it for the fuzzy or uncertain edge regions. Meanwhile, we design the Multi-Level Attention Decoder (MLAD) to further enhance the attention to the target region through horizontal-vertical viewgraph decoupling and orientation-aware self-attention mechanism. Finally, the optimized features at all levels are integrated through the Multi-Scale Feature Fusion Module (MSFM) to gradually achieve segmentation prediction from coarse to fine. The MABR-Net model is trained using deep learning techniques on high-resolution colonoscopy images, with the training process conducted in a multi-GPU server environment to accelerate the learning process. Through parallel processing and distributed computation, the model is trained across multiple GPUs, handling large-scale data and ensuring real-time segmentation performance. The experimental results show that MABR-Net performs excellently in segmenting polyps and other lesions in endoscopic images, which not only significantly improves the edge segmentation accuracy, but also enhances the robustness and generalization ability of the model to complex structures under co-occurrence. This provides a novel and efficient technological path to solve the co-occurrence problem in medical imaging, and provides strong support for early clinical diagnosis and treatment planning. Code:https://github.com/User-deepll/MABR-Net.\u003c/p\u003e","manuscriptTitle":"Multi-Level Attention and Boundary Refinement Network for Polyp Segmentation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-03 02:35:03","doi":"10.21203/rs.3.rs-6435376/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":"cca08076-5594-44d4-b2d1-a2715ee3242c","owner":[],"postedDate":"December 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-12-03T02:35:03+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-03 02:35:03","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6435376","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6435376","identity":"rs-6435376","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}