FusionNeXt-XtremeNet: A Deep Ensemble Model with LLM-Aided Clinical Report Generation for Dermoscopic Image Classification

FusionNeXt-XtremeNet: A Deep Ensemble Model with LLM-Aided Clinical Report Generation for Dermoscopic Image Classification | 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 FusionNeXt-XtremeNet: A Deep Ensemble Model with LLM-Aided Clinical Report Generation for Dermoscopic Image Classification Saroj Ghadle, Jitendra Tembhurne This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7784164/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 This paper presents FusionNeXt-XtremeNet, a novel deep ensemble architecture that combines ConvNeXt, Vision Transformer (ViT), and EfficientNetV2 for classifying dermoscopic images based on acquisition types. To improve clinical interpretability, a GPT-2-based Large Language Model (LLM) enhanced by the Language-augmented Multimodal Attention (LeMMA) mechanism is integrated to generate structured diagnostic reports. Model evaluated on the ISIC 2020--2022 dataset of 1,767 images, and achieves state-of-the-art performance in binary classification (94.1% accuracy, 94.1% F1-score, 0.969 ROC-AUC), three-class classification (90.6% accuracy, 90.8% F1-score), and four-class classification (87.6% accuracy, 87.8% F1-score). The LeMMA-augmented GPT-2 generates clinically relevant reports with a BLEU score of 0.85, reducing generation time by 15.2% compared to baseline, and achieves high dermatologist evaluation scores (accuracy: 4.3/5, relevance: 4.4/5). Grad-CAM visualisations demonstrate strong alignment with clinical features (r=0.82, p<0.001), with 85% of attention regions corresponding to dermatologically significant patterns. This dual framework not only enhances prediction reliability but also bridges the gap between black-box AI models and clinical usability through explainable, text-based outputs. Dermoscopy Deep Learning Vision Transformer ConvNeXt LLM Clinical Report Generation 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-7784164","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":600019851,"identity":"7d49b3a0-7c7e-4fac-b2a9-357cf847e2d9","order_by":0,"name":"Saroj Ghadle","email":"","orcid":"","institution":"Indian Institute of Information Technology, Nagpur","correspondingAuthor":false,"prefix":"","firstName":"Saroj","middleName":"","lastName":"Ghadle","suffix":""},{"id":600019852,"identity":"ff63d724-84c3-44d4-964a-22661d71a539","order_by":1,"name":"Jitendra Tembhurne","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6ElEQVRIiWNgGAWjYBACxgYGBmYQAaQOHEgwgAon4NPSBtfClgDUYkBYCwMbXAsPSLkBPrUQwDy/+ejmwh2H883Zez4eeFDwJ4+B/fADhoc78DmMLe32zDOHLXf2nN0AclgxA0+aAUPiGXxaeMxu87YdNjC4kQvWktjAkMPAkNiGTwv/N4iW+28eQLTwvyGkhYcNagsPA0SLBEFb0sxuz2xLNzA4k2YA1GKc2CbxzOAAPi2GzYef3S5sszYwOH748ccff+QS+/mTHz78iU9LA7oIMKIYDuDWwMAgj09yFIyCUTAKRgEYAACoo1bDCSSkaAAAAABJRU5ErkJggg==","orcid":"","institution":"Indian Institute of Information Technology, Nagpur","correspondingAuthor":true,"prefix":"","firstName":"Jitendra","middleName":"","lastName":"Tembhurne","suffix":""}],"badges":[],"createdAt":"2025-10-05 10:23:23","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7784164/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7784164/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107490289,"identity":"2a578b26-7ff6-4c87-8658-744f885747a7","added_by":"auto","created_at":"2026-04-22 02:51:37","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":945337,"visible":true,"origin":"","legend":"","description":"","filename":"paper.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7784164/v1_covered_d99f7585-7e3b-4d40-8ac6-8d5f6870a5b4.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"FusionNeXt-XtremeNet: A Deep Ensemble Model with LLM-Aided Clinical Report Generation for Dermoscopic Image Classification","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"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":"Dermoscopy, Deep Learning, Vision Transformer, ConvNeXt, LLM, Clinical Report Generation","lastPublishedDoi":"10.21203/rs.3.rs-7784164/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7784164/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"This paper presents FusionNeXt-XtremeNet, a novel deep ensemble architecture that combines ConvNeXt, Vision Transformer (ViT), and EfficientNetV2 for classifying dermoscopic images based on acquisition types. To improve clinical interpretability, a GPT-2-based Large Language Model (LLM) enhanced by the Language-augmented Multimodal Attention (LeMMA) mechanism is integrated to generate structured diagnostic reports. Model evaluated on the ISIC 2020--2022 dataset of 1,767 images, and achieves state-of-the-art performance in binary classification (94.1% accuracy, 94.1% F1-score, 0.969 ROC-AUC), three-class classification (90.6% accuracy, 90.8% F1-score), and four-class classification (87.6% accuracy, 87.8% F1-score). The LeMMA-augmented GPT-2 generates clinically relevant reports with a BLEU score of 0.85, reducing generation time by 15.2% compared to baseline, and achieves high dermatologist evaluation scores (accuracy: 4.3/5, relevance: 4.4/5). Grad-CAM visualisations demonstrate strong alignment with clinical features (r=0.82, p\u003c0.001), with 85% of attention regions corresponding to dermatologically significant patterns. This dual framework not only enhances prediction reliability but also bridges the gap between black-box AI models and clinical usability through explainable, text-based outputs.","manuscriptTitle":"FusionNeXt-XtremeNet: A Deep Ensemble Model with LLM-Aided Clinical Report Generation for Dermoscopic Image Classification","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-05 16:16:20","doi":"10.21203/rs.3.rs-7784164/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":"716fd270-59bf-48a1-8bac-83223b900bbd","owner":[],"postedDate":"March 5th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-21T16:25:42+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-05 16:16:20","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7784164","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7784164","identity":"rs-7784164","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}