CIN-RiskNet: A Dynamic Feature-Enhanced TabTransformer with Hybrid SMOTE-Noise Augmentation for Contrast-Induced Nephropathy Prediction

CIN-RiskNet: A Dynamic Feature-Enhanced TabTransformer with Hybrid SMOTE-Noise Augmentation for Contrast-Induced Nephropathy Prediction | 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 CIN-RiskNet: A Dynamic Feature-Enhanced TabTransformer with Hybrid SMOTE-Noise Augmentation for Contrast-Induced Nephropathy Prediction Peng Zhang, Keyu Gong, Xue Zhang, Xiaogang Liu, Shicheng Yang, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9161846/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 Contrast-Induced Nephropathy (CIN) is a serious complication following the use of contrast media in cardiovascular interventions, with no effective treatment available. Early prediction is crucial for prevention, but existing models often struggle with class imbalance, feature redundancy, and noise in clinical data. Methods This study proposes CIN-RiskNet, a dynamic feature-enhanced TabTransformer model integrated with a hybrid SMOTE-Noise augmentation strategy. The approach includes adaptive feature gating to suppress noise, synthetic minority oversampling to address class imbalance, and multi-head self-attention to capture complex feature interactions. The model was trained and evaluated using five-fold cross-validation on a clinical dataset from Tianjin University Chest Hospital. Results CIN-RiskNet achieved state-of-the-art performance with an accuracy of 99.0%, recall of 99.0%, and an F1-score of 99.0%, outperforming traditional machine learning models such as XGBoost, Random Forest, and support vector machine. Ablation studies confirmed the contributions of each module, demonstrating improved robustness and generalization. Conclusions The proposed model effectively addresses key challenges in CIN prediction, including class imbalance and feature noise, through an integrated deep learning framework. It shows strong potential for clinical application, though further validation on multi-center datasets is recommended to enhance generalizability. Contrast-Induced Nephropathy Transformer Self-Attention Deep Learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Full Text Additional Declarations No competing interests reported. Tables 1 to 3 are available in the Supplementary Files section. Supplementary Files Table1.docx Table2.docx Table3.docx 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-9161846","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":611605598,"identity":"d04ba9e9-3110-4649-8645-c7ac054f70ec","order_by":0,"name":"Peng Zhang","email":"","orcid":"","institution":"Tianjin University","correspondingAuthor":false,"prefix":"","firstName":"Peng","middleName":"","lastName":"Zhang","suffix":""},{"id":611605600,"identity":"3d0ff2ba-809f-4328-9700-ccfda4d48bca","order_by":1,"name":"Keyu Gong","email":"","orcid":"","institution":"Tianjin University","correspondingAuthor":false,"prefix":"","firstName":"Keyu","middleName":"","lastName":"Gong","suffix":""},{"id":611605601,"identity":"972b17fc-99f4-4ed3-a05d-37b8373dbe3b","order_by":2,"name":"Xue Zhang","email":"","orcid":"","institution":"Tianjin University","correspondingAuthor":false,"prefix":"","firstName":"Xue","middleName":"","lastName":"Zhang","suffix":""},{"id":611605602,"identity":"ecb46755-fd90-4678-afaa-b60cfe12b600","order_by":3,"name":"Xiaogang Liu","email":"","orcid":"","institution":"Tianjin University","correspondingAuthor":false,"prefix":"","firstName":"Xiaogang","middleName":"","lastName":"Liu","suffix":""},{"id":611605603,"identity":"c046149d-9c44-4d35-8839-c974c7e427ac","order_by":4,"name":"Shicheng Yang","email":"","orcid":"","institution":"Tianjin University","correspondingAuthor":false,"prefix":"","firstName":"Shicheng","middleName":"","lastName":"Yang","suffix":""},{"id":611605604,"identity":"53849eac-49ba-44af-a765-3649b6b463f4","order_by":5,"name":"Zhiwei Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9klEQVRIiWNgGAWjYDACCTB5gIeBgbEByLbh4WdvIE1LmoxkzwHitMDYh20Mbjjg1yE/u/nZwy9/7sgYHG9uvPGh5jwPww0Gxg8fc3BrYZxzzNxYhucZj8GZg82WM47d5mGc3cAsOXMbbi3MEglm0hISh3nMbiS2SfM23OZhljnAxsyLRwubRPo3aQkDoJb7D0FazvGwSSTg18IjkWMm+SEBZAsjSMsBHh5CWiQkcsqkGQ4c5rE/kwjySzKPBM/BZrx+kZ+Rvk3yx5/D9pLtxx8CQ8zO3v5488EPH/FoAQcBDyqfsQG/epCSHwSVjIJRMApGwYgGAJBFUWxE3ghEAAAAAElFTkSuQmCC","orcid":"","institution":"Second Hospital of Tianjin Medical University","correspondingAuthor":true,"prefix":"","firstName":"Zhiwei","middleName":"","lastName":"Zhang","suffix":""},{"id":611605605,"identity":"c59ecfcd-b704-4f06-9f19-2a2384fd1009","order_by":6,"name":"Ximing Li","email":"","orcid":"","institution":"Tianjin University","correspondingAuthor":false,"prefix":"","firstName":"Ximing","middleName":"","lastName":"Li","suffix":""},{"id":611605606,"identity":"f643b31f-0554-4d2c-8dfa-5cad7aee6a27","order_by":7,"name":"Runnan He","email":"","orcid":"","institution":"Tianjin University","correspondingAuthor":false,"prefix":"","firstName":"Runnan","middleName":"","lastName":"He","suffix":""}],"badges":[],"createdAt":"2026-03-18 17:08:45","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9161846/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9161846/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105358211,"identity":"146fa68d-03a4-45e1-80ad-800421b208c2","added_by":"auto","created_at":"2026-03-25 07:17:17","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":268273,"visible":true,"origin":"","legend":"\u003cp\u003eOverview of workflow. (A) Acquisition of multidimensional physiological and biochemical indicators from patient records. (B) Hybrid SMOTE-Noise data augmentation module: applying Synthetic Minority Oversampling Technique (SMOTE) to mitigate class imbalance and injecting calibrated noise to simulate clinical measurement variability, enhancing robustness. (C) Core predictive architecture: Gated feature selection dynamically filters informative features; a Transformer encoder with multi-head self-attention captures complex feature interactions; and a Multi-Layer Perceptron (MLP) head generates the final CIN risk prediction.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-9161846/v1/fe3ddb617f21eaa47c6745c3.png"},{"id":105564706,"identity":"dcb76e84-b11c-4de2-98e2-062c0f847d61","added_by":"auto","created_at":"2026-03-27 12:50:34","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":364918,"visible":true,"origin":"","legend":"\u003cp\u003eArchitecture and workflow of the CIN-RiskNet framework. Input layer: Tabular data of n patients structured as a n × 27 matrix of clinical measurements with corresponding binary labels. Processing pipeline: ① Hybrid Data Augmentation: SMOTE: Synthetic minority oversampling to balance class distribution. Gaussian Noise Injection: Simulates clinical measurement errors through feature-wise noise perturbation. ② Gated Feature Selection: Learns adaptive weights to suppress redundant/noisy features. ③ Transformer Encoder: Multi-head self-attention captures cross-feature interactions. Layer Normalization (Norm) stabilizes training. Multilayer Perceptron (MLP) refines representations. ④ Classification Head: Sigmoid-activated MLP outputting CIN risk probability. Output: Binary classification output: Negative (CIN risk \u0026lt;0.5) / Positive (CIN risk ≥0.5).\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-9161846/v1/0cb443dc9f4f468eb1f58c87.png"},{"id":105358208,"identity":"0f8cd47c-ae06-4bd0-93a7-2d8b46317abe","added_by":"auto","created_at":"2026-03-25 07:17:16","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":545381,"visible":true,"origin":"","legend":"\u003cp\u003eSMOTE algorithm flow diagram. (a) Neighbor Selection: For each CIN-positive patient xi, identify k = 3 nearest neighbors based on Euclidean distance. (b) Synthetic Generation: Randomly interpolate between xi and neighbor xj to create new samples xnew. (c) Output: Augmented training set with balanced class distribution.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-9161846/v1/d6de618058c8368f4bf9a7fd.png"},{"id":105728011,"identity":"c7aeeed3-2578-41b4-a626-f4914799949c","added_by":"auto","created_at":"2026-03-30 11:08:02","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":108317,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance analysis of CIN-RiskNet for contrast-induced nephropathy prediction. (a) Baseline model comparison: Performance of CIN-RiskNet versus clinical prediction models. Asterisk (*) denotes p \u0026lt; 0.001 statistical significance. (b) Component ablation study: Contribution of each innovation to the final performance. Data source: Tianjin University Chest Hospital Cardiology Department | Metric: F1-score | Dashed line: Clinical excellence threshold (0.98).\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-9161846/v1/84c8fbf621457c561a678321.png"},{"id":107417486,"identity":"77f07cdb-c383-40f1-bc05-a17b877015c1","added_by":"auto","created_at":"2026-04-21 09:58:35","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":833968,"visible":true,"origin":"","legend":"","description":"","filename":"CINArticleBMCCD.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9161846/v1_covered_169b27b5-6697-41cb-9d17-df9a4ebe5997.pdf"},{"id":105358205,"identity":"9c484fec-445a-4d6b-9fea-85d2b7304c5e","added_by":"auto","created_at":"2026-03-25 07:17:16","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":20472,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-9161846/v1/fc7cbeffcaa9df1dfe7d0052.docx"},{"id":105358209,"identity":"6c2c7e7a-b60e-4f79-8d3a-879cc082bd5c","added_by":"auto","created_at":"2026-03-25 07:17:17","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":19839,"visible":true,"origin":"","legend":"","description":"","filename":"Table2.docx","url":"https://assets-eu.researchsquare.com/files/rs-9161846/v1/42c0561457e57750f37a1a19.docx"},{"id":105358210,"identity":"c7ef933a-00ff-4a29-9523-0b71e914a278","added_by":"auto","created_at":"2026-03-25 07:17:17","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":19729,"visible":true,"origin":"","legend":"","description":"","filename":"Table3.docx","url":"https://assets-eu.researchsquare.com/files/rs-9161846/v1/ad0b94528b6ff53ae7daa2d8.docx"}],"financialInterests":"\u003cp\u003eNo competing interests reported.\u003c/p\u003e\n\u003cp\u003eTables 1 to 3 are available in the Supplementary Files section.\u003c/p\u003e","formattedTitle":"CIN-RiskNet: A Dynamic Feature-Enhanced TabTransformer with Hybrid SMOTE-Noise Augmentation for Contrast-Induced Nephropathy Prediction","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":"Contrast-Induced Nephropathy, Transformer, Self-Attention, Deep Learning","lastPublishedDoi":"10.21203/rs.3.rs-9161846/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9161846/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eContrast-Induced Nephropathy (CIN) is a serious complication following the use of contrast media in cardiovascular interventions, with no effective treatment available. Early prediction is crucial for prevention, but existing models often struggle with class imbalance, feature redundancy, and noise in clinical data.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis study proposes CIN-RiskNet, a dynamic feature-enhanced TabTransformer model integrated with a hybrid SMOTE-Noise augmentation strategy. The approach includes adaptive feature gating to suppress noise, synthetic minority oversampling to address class imbalance, and multi-head self-attention to capture complex feature interactions. The model was trained and evaluated using five-fold cross-validation on a clinical dataset from Tianjin University Chest Hospital.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eCIN-RiskNet achieved state-of-the-art performance with an accuracy of 99.0%, recall of 99.0%, and an F1-score of 99.0%, outperforming traditional machine learning models such as XGBoost, Random Forest, and support vector machine. Ablation studies confirmed the contributions of each module, demonstrating improved robustness and generalization.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe proposed model effectively addresses key challenges in CIN prediction, including class imbalance and feature noise, through an integrated deep learning framework. It shows strong potential for clinical application, though further validation on multi-center datasets is recommended to enhance generalizability.\u003c/p\u003e","manuscriptTitle":"CIN-RiskNet: A Dynamic Feature-Enhanced TabTransformer with Hybrid SMOTE-Noise Augmentation for Contrast-Induced Nephropathy Prediction","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-25 07:17:11","doi":"10.21203/rs.3.rs-9161846/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":"9b1c1ee2-6740-4235-80cc-4cf65772c1c9","owner":[],"postedDate":"March 25th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-21T09:57:17+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-25 07:17:11","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9161846","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9161846","identity":"rs-9161846","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}