SkinGuardian: On-Device AI for Private, Fair, Robust, and Explainable Skin Cancer Detection

SkinGuardian: On-Device AI for Private, Fair, Robust, and Explainable Skin Cancer 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 SkinGuardian: On-Device AI for Private, Fair, Robust, and Explainable Skin Cancer Detection Aayush Kumar, Fahad Salim Dalwai This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8484276/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 Background: Early skin cancer detection improves outcomes, but access to dermatology screening remains limited. Many AI systems rely on cloud inference, raising privacy concerns and restricting use in low-connectivity settings. Methods: We present SkinGuardian, an on-device benign–malignant skin lesion classifier that integrates four trustworthiness dimensions: fairnessaware learning, adversarial robustness, differential privacy, and explainability. We fine-tune a BEiT vision transformer on ISIC 2019 and Fitzpatrick17k (train/validation only; test held out for subgroup evaluation), and deploy via ONNX Runtime with INT8 weights-only quantization. Results: SkinGuardian-Clean achieves AUROC 0.956 on ISIC 2019, and generalizes to the SIIM-ISIC 2020 melanoma setting with AUROC 0.927; at the ISIC-2019-tuned operating threshold, accuracy is 85.4%. Fairness mitigation reduces demographic parity difference on Fitzpatrick17k from 0.12 to 0.04 and equalized odds difference from 0.15 to 0.05. SkinGuardian-Robust attains 74.8% robust accuracy against PGD-10 (ϵ = 8/255; clean 87.1%). With DPSGD, accuracy remains 86.1% at ϵ = 1 (δ = 1/N) on ISIC 2019. On-device inference achieves p95 ≤160 ms with INT8. Conclusion: SkinGuardian demonstrates a practical, privacy-preserving and equitable on-device screening research prototype and is not a standalone diagnostic device. Biological sciences/Cancer Biological sciences/Computational biology and bioinformatics Health sciences/Health care Physical sciences/Mathematics and computing Algorithmic fairness Differential privacy Edge computing Explainable artificial intelligence (XAI) Vision transformers Adversarial robustness Full Text Additional Declarations No competing interests reported. Supplementary Files OnlineResourceSupplement.pdf 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-8484276","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":570196217,"identity":"2b33464b-a5bd-4e63-bbba-fc5ee3243619","order_by":0,"name":"Aayush Kumar","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxElEQVRIiWNgGAWjYDACHhBhICHHz8DYAGQxE6ulwMZYsoE0LR/SEjccAHOJ0MLfc8Z0ww+Dw4ybbyQ3fmCosE5sIKRF4myP2c0eg8PMZjcSmyUYzqQT1sJwnsfsBo/BYTagljYGxrbDhLXIA7Xc/GNwmMd4BkjLPyK0GAAddpvHIE3CQAKkpYEILYZnjpXdljGwMZA487BZIuFYujFBLXJnkrfdfPNHor6/Pf3hhw811rIEtaCCBNKUj4JRMApGwSjABQBCnUFXVMaWYAAAAABJRU5ErkJggg==","orcid":"","institution":"University of Illinois at Chicago","correspondingAuthor":true,"prefix":"","firstName":"Aayush","middleName":"","lastName":"Kumar","suffix":""},{"id":570196220,"identity":"95286f59-cb58-42e6-90e0-0128b99f2810","order_by":1,"name":"Fahad Salim Dalwai","email":"","orcid":"","institution":"University of Illinois at Chicago","correspondingAuthor":false,"prefix":"","firstName":"Fahad","middleName":"Salim","lastName":"Dalwai","suffix":""}],"badges":[],"createdAt":"2025-12-30 21:23:09","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8484276/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8484276/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":99717177,"identity":"64c3de26-59ff-40dd-9847-1f52e0e54f5e","added_by":"auto","created_at":"2026-01-07 14:47:23","extension":"json","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4522,"visible":true,"origin":"","legend":"","description":"","filename":"c2481a2170e741189051b30e5a1f285b.json","url":"https://assets-eu.researchsquare.com/files/rs-8484276/v1/6b6b79b0bcc9c6d173e14b21.json"},{"id":102746135,"identity":"06547868-13a2-4a69-baea-d852d3e64fd2","added_by":"auto","created_at":"2026-02-16 08:55:49","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":900650,"visible":true,"origin":"","legend":"","description":"","filename":"SkinGuardianSubmission.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8484276/v1_covered_8c053b4c-3950-4674-af19-6c9e9aeda81f.pdf"},{"id":99717178,"identity":"8b79f078-2b03-42a5-ac18-12e72babeb49","added_by":"auto","created_at":"2026-01-07 14:47:23","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":1679182,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineResourceSupplement.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8484276/v1/1cd9eba3448387fad605bd51.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"SkinGuardian: On-Device AI for Private, Fair, Robust, and Explainable Skin Cancer Detection","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":"Algorithmic fairness, Differential privacy, Edge computing, Explainable artificial intelligence (XAI), Vision transformers, Adversarial robustness","lastPublishedDoi":"10.21203/rs.3.rs-8484276/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8484276/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e Early skin cancer detection improves outcomes, but access to dermatology screening remains limited. Many AI systems rely on cloud inference, raising privacy concerns and restricting use in low-connectivity settings.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e We present SkinGuardian, an on-device benign–malignant skin lesion classifier that integrates four trustworthiness dimensions: fairnessaware learning, adversarial robustness, differential privacy, and explainability. We fine-tune a BEiT vision transformer on ISIC 2019 and Fitzpatrick17k (train/validation only; test held out for subgroup evaluation), and deploy via ONNX Runtime with INT8 weights-only quantization.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e SkinGuardian-Clean achieves AUROC 0.956 on ISIC 2019, and generalizes to the SIIM-ISIC 2020 melanoma setting with AUROC 0.927; at the ISIC-2019-tuned operating threshold, accuracy is 85.4%. Fairness mitigation reduces demographic parity difference on Fitzpatrick17k from 0.12 to 0.04 and equalized odds difference from 0.15 to 0.05. SkinGuardian-Robust attains 74.8% robust accuracy against PGD-10 (ϵ = 8/255; clean 87.1%). With DPSGD, accuracy remains 86.1% at ϵ = 1 (δ = 1/N) on ISIC 2019. On-device inference achieves p95 ≤160 ms with INT8.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003e SkinGuardian demonstrates a practical, privacy-preserving and equitable on-device screening research prototype and is not a standalone diagnostic device.\u003c/p\u003e","manuscriptTitle":"SkinGuardian: On-Device AI for Private, Fair, Robust, and Explainable Skin Cancer Detection","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-07 14:47:15","doi":"10.21203/rs.3.rs-8484276/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":"02768be5-9780-46c6-afd8-40bfabc6dd6a","owner":[],"postedDate":"January 7th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":60672435,"name":"Biological sciences/Cancer"},{"id":60672436,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":60672437,"name":"Health sciences/Health care"},{"id":60672438,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2026-02-12T00:09:00+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-07 14:47:15","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8484276","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8484276","identity":"rs-8484276","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}