Cut Instance Mixing: a domain-specific data augmentation method applied to gastrointestinal lesion detection

Cut Instance Mixing: a domain-specific data augmentation method applied to gastrointestinal lesion detection | 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 Cut Instance Mixing: a domain-specific data augmentation method applied to gastrointestinal lesion detection Alexandre Neto, Eduarda Almeida, Diogo Libânio, Mário Dinis-Ribeiro, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6401606/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 04 Mar, 2026 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract We propose Cut Instance Mixing (CIM), a domain-specific data augmentation method for improving deep learning (DL) models in detecting gastrointestinal lesions during endoscopy. CIM overcomes the limitations of state-of-the-art techniques like MixUp and CutMix by adapting them to the unique challenges of endoscopy, including localized, irregular lesions such as intestinal metaplasia (IM), dysplasia, and polyps. CIM promotes biologically relevant augmentations by identifying regions of interest and blending lesion features seamlessly using Poisson image editing and gradient mixing. Our experiments utilized ResNet50, trained on datasets for IM, dysplasia, and polyps, with extensive evaluation of internal and external test sets. Results demonstrate that CIM with optimized blending (α=0.8) significantly outperforms MixUp and CutMix across key metrics, achieving the highest AUC (0.879) and accuracy (0.823) for IM detection and near-perfect AUC (0.997) for dysplasia classification. Additionally, CIM exhibits superior generalization capabilities, maintaining robust performance on external polyp datasets under diverse conditions. CIM enhances model sensitivity and precision by producing realistic, lesion-focused training samples, as confirmed by Grad-CAM heatmap analyses. These results highlight its potential in improving DL-based endoscopic diagnosis or other specific domain contexts, particularly for underrepresented lesion classes. Our findings underscore the importance of domain-specific augmentations, especially in limited and unbalanced datasets. Health sciences/Gastroenterology/Colonoscopy Health sciences/Gastroenterology/Gastrointestinal system Health sciences/Gastroenterology/Oesophagogastroscopy Physical sciences/Engineering/Biomedical engineering Health sciences/Gastroenterology Health sciences/Health care Physical sciences/Engineering Biological sciences/Computational biology and bioinformatics Biological sciences/Computational biology and bioinformatics/Computational models Biological sciences/Computational biology and bioinformatics/Data processing Biological sciences/Computational biology and bioinformatics/Machine learning cut-and-paste data augmentation deep learning gastrointestinal endoscopy multiple-image combination Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 04 Mar, 2026 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 24 Nov, 2025 Reviews received at journal 12 Nov, 2025 Reviews received at journal 22 Oct, 2025 Reviewers agreed at journal 19 Oct, 2025 Reviewers agreed at journal 13 Oct, 2025 Reviewers invited by journal 13 Oct, 2025 Editor assigned by journal 27 Jul, 2025 Editor invited by journal 23 Apr, 2025 Submission checks completed at journal 23 Apr, 2025 First submitted to journal 08 Apr, 2025 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-6401606","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":533550305,"identity":"23922499-5949-47d1-9421-59f162fa6f55","order_by":0,"name":"Alexandre Neto","email":"data:image/png;base64,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","orcid":"","institution":"INESC TEC – Instituto de Engenharia de Sistemas e Computadores, Tecnologia e Ciência, 3200-465 Porto, Portugal","correspondingAuthor":true,"prefix":"","firstName":"Alexandre","middleName":"","lastName":"Neto","suffix":""},{"id":533550307,"identity":"8a1c7475-9154-430a-a9b8-4e38cbcf4adb","order_by":1,"name":"Eduarda Almeida","email":"","orcid":"","institution":"INESC TEC – Instituto de Engenharia de Sistemas e Computadores, Tecnologia e Ciência, 3200-465 Porto, Portugal","correspondingAuthor":false,"prefix":"","firstName":"Eduarda","middleName":"","lastName":"Almeida","suffix":""},{"id":533550314,"identity":"f27687a0-f82b-48dc-9847-7280b7df3fcb","order_by":2,"name":"Diogo Libânio","email":"","orcid":"","institution":"Faculdade de Medicina, Universidade do Porto, 4200-319 Porto, Portugal","correspondingAuthor":false,"prefix":"","firstName":"Diogo","middleName":"","lastName":"Libânio","suffix":""},{"id":533550315,"identity":"ed44f769-fa6b-49de-b5ea-c619b1366450","order_by":3,"name":"Mário Dinis-Ribeiro","email":"","orcid":"","institution":"Faculdade de Medicina, Universidade do Porto, 4200-319 Porto, Portugal","correspondingAuthor":false,"prefix":"","firstName":"Mário","middleName":"","lastName":"Dinis-Ribeiro","suffix":""},{"id":533550317,"identity":"48fa37d2-3847-49f5-8225-0bebd73cdf71","order_by":4,"name":"Miguel Coimbra","email":"","orcid":"","institution":"INESC TEC – Instituto de Engenharia de Sistemas e Computadores, Tecnologia e Ciência, 3200-465 Porto, Portugal","correspondingAuthor":false,"prefix":"","firstName":"Miguel","middleName":"","lastName":"Coimbra","suffix":""},{"id":533550320,"identity":"27f06b1c-2fd2-4e73-80c7-748563e3b491","order_by":5,"name":"António Cunha","email":"","orcid":"","institution":"Escola de Ciências e Tecnologia, Universidade de Trás-os-Montes e Alto Douro, 5001-801 Vila Real, Portugal","correspondingAuthor":false,"prefix":"","firstName":"António","middleName":"","lastName":"Cunha","suffix":""}],"badges":[],"createdAt":"2025-04-08 09:23:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6401606/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6401606/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-026-42138-2","type":"published","date":"2026-03-04T15:59:53+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":104250906,"identity":"0ed3a30c-b644-4dd6-a825-6dcf8d546f3c","added_by":"auto","created_at":"2026-03-09 16:11:24","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1204584,"visible":true,"origin":"","legend":"","description":"","filename":"20250331CIMSRAN.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6401606/v1_covered_eb09f582-80b5-407e-8537-61aede427674.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Cut Instance Mixing: a domain-specific data augmentation method applied to gastrointestinal lesion detection","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"cut-and-paste, data augmentation, deep learning, gastrointestinal endoscopy, multiple-image combination","lastPublishedDoi":"10.21203/rs.3.rs-6401606/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6401606/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"We propose Cut Instance Mixing (CIM), a domain-specific data augmentation method for improving deep learning (DL) models in detecting gastrointestinal lesions during endoscopy. CIM overcomes the limitations of state-of-the-art techniques like MixUp and CutMix by adapting them to the unique challenges of endoscopy, including localized, irregular lesions such as intestinal metaplasia (IM), dysplasia, and polyps. CIM promotes biologically relevant augmentations by identifying regions of interest and blending lesion features seamlessly using Poisson image editing and gradient mixing.\nOur experiments utilized ResNet50, trained on datasets for IM, dysplasia, and polyps, with extensive evaluation of internal and external test sets. Results demonstrate that CIM with optimized blending (α=0.8) significantly outperforms MixUp and CutMix across key metrics, achieving the highest AUC (0.879) and accuracy (0.823) for IM detection and near-perfect AUC (0.997) for dysplasia classification. Additionally, CIM exhibits superior generalization capabilities, maintaining robust performance on external polyp datasets under diverse conditions.\nCIM enhances model sensitivity and precision by producing realistic, lesion-focused training samples, as confirmed by Grad-CAM heatmap analyses. These results highlight its potential in improving DL-based endoscopic diagnosis or other specific domain contexts, particularly for underrepresented lesion classes. Our findings underscore the importance of domain-specific augmentations, especially in limited and unbalanced datasets.","manuscriptTitle":"Cut Instance Mixing: a domain-specific data augmentation method applied to gastrointestinal lesion detection","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-27 11:31:22","doi":"10.21203/rs.3.rs-6401606/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-25T03:57:26+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-12T15:59:13+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-22T17:43:20+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"276879427077940626344551376280838296671","date":"2025-10-20T01:09:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"293574218175111649052750926292083540162","date":"2025-10-13T14:47:38+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-13T11:27:59+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-07-27T19:06:50+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-04-23T11:08:25+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-23T10:13:46+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-04-08T09:15:25+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"8242f382-6a00-4dd2-9633-5d4d3b456fb4","owner":[],"postedDate":"October 27th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":56728687,"name":"Health sciences/Gastroenterology/Colonoscopy"},{"id":56728688,"name":"Health sciences/Gastroenterology/Gastrointestinal system"},{"id":56728689,"name":"Health sciences/Gastroenterology/Oesophagogastroscopy"},{"id":56728690,"name":"Physical sciences/Engineering/Biomedical engineering"},{"id":56728691,"name":"Health sciences/Gastroenterology"},{"id":56728692,"name":"Health sciences/Health care"},{"id":56728693,"name":"Physical sciences/Engineering"},{"id":56728694,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":56728695,"name":"Biological sciences/Computational biology and bioinformatics/Computational models"},{"id":56728696,"name":"Biological sciences/Computational biology and bioinformatics/Data processing"},{"id":56728697,"name":"Biological sciences/Computational biology and bioinformatics/Machine learning"}],"tags":[],"updatedAt":"2026-03-09T16:07:39+00:00","versionOfRecord":{"articleIdentity":"rs-6401606","link":"https://doi.org/10.1038/s41598-026-42138-2","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2026-03-04 15:59:53","publishedOnDateReadable":"March 4th, 2026"},"versionCreatedAt":"2025-10-27 11:31:22","video":"","vorDoi":"10.1038/s41598-026-42138-2","vorDoiUrl":"https://doi.org/10.1038/s41598-026-42138-2","workflowStages":[]},"version":"v1","identity":"rs-6401606","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6401606","identity":"rs-6401606","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}