Low-Dose Medical Image Reconstruction Using Residual U-Net:A Hybrid Approach Integrating Compressed Sensing and Deep Learning

Low-Dose Medical Image Reconstruction Using Residual U-Net:A Hybrid Approach Integrating Compressed Sensing and Deep Learning | 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 Low-Dose Medical Image Reconstruction Using Residual U-Net:A Hybrid Approach Integrating Compressed Sensing and Deep Learning Nripesh Kumar, AJAY SINGH, Rashid Ali, Vishal Pal This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8906855/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 Low-dose computed tomography (LDCT) and otherlow-dose medical imaging modalities are critically important for reducing patient radiation exposure andminimizing cancer risks. However, image reconstruction from severely undersampled or noisy measurements presents a significant ill-posed inverse problem, leading to degraded image quality characterizedby quantum noise, streaking artifacts, and loss ofstructural detail. This paper proposes a novel hybridframework that synergistically combines the theoretical foundations of Compressed Sensing (CS) with therepresentational power of deep learning, specificallythrough a Residual U-Net architecture. The proposed model operates within an iterative reconstruction paradigm, where a CS-based data fidelity step isalternated with a deep learning-based prior step, implemented by the Residual U-Net, to progressivelyrefine the image. The Residual U-Net, enhancedwith squeeze-and-excitation blocks, is designed tolearn the mapping from artifact-corrupted, low-doseimages to their high-dose counterparts, effectivelymodeling the complex noise and artifact distributions while preserving anatomical structures. We formulate the reconstruction as an optimization problem, solved via the Learned Primal-Dual algorithm.Extensive quantitative and qualitative evaluationsare performed on the publicly available Mayo ClinicLDCT dataset and a simulated undersampled MRIdataset (k-space data). Results demonstrate thatour proposed Residual U-Net integrated iterative reconstruction (RN-IR) method significantly outperforms traditional methods like Filtered Back Projection (FBP), total variation (TV)-based CS, andstandalone deep learning models (such as a vanilla UNet) in terms of Peak Signal-to-Noise Ratio (PSNR),Structural Similarity Index (SSIM), and visual quality. Our model achieves a PSNR of 42.3 dB and SSIMof 0.985 on the LDCT dataset, effectively suppressing noise and preserving fine details, thus providing arobust and efficient solution for high-quality low-dose medical image reconstruction. Low-Dose CT MRI Reconstruction Compressed Sensing Deep Learning U-Net Residual Learning Iterative Reconstruction Inverse Problems 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-8906855","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":593784793,"identity":"34363329-52fa-485a-a1f6-8be077595d4d","order_by":0,"name":"Nripesh Kumar","email":"","orcid":"","institution":"Teerthanker Mahaveer University","correspondingAuthor":false,"prefix":"","firstName":"Nripesh","middleName":"","lastName":"Kumar","suffix":""},{"id":593784794,"identity":"2496cf3a-5e2e-4c5e-a164-2de65468b6ab","order_by":1,"name":"AJAY SINGH","email":"","orcid":"","institution":"Teerthanker Mahaveer University","correspondingAuthor":false,"prefix":"","firstName":"AJAY","middleName":"","lastName":"SINGH","suffix":""},{"id":593784795,"identity":"a50cf709-cf11-4472-9020-b876d2610a93","order_by":2,"name":"Rashid Ali","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8klEQVRIiWNgGAWjYDACZjZmhoQDBxgY5B8fOPABKMDGTrQWhrTEgzNAWpgJWgNSAtaSY3yYB2wIAQ3m7GzJBg/O3JE3ZzhjcNjm1zZ5PmYGxg8fc3BrsWxmO5yQcOOZ4c7GtoLDuX23DduYGZglZ27DrcXgMHvzgYQPhxk3HGbecDi35zYjUAsbMy8RWuw3HAOyLXtu2xOhBeyww4kbzrAYHGb4cTuRGC3JBglnDidvuMGWcLC34XZyGzNjM36/nD9mLPnj2GHbDTeYD3/48ee27fz25oMfPuLRggoY28BkA7HqQeAPKYpHwSgYBaNgpAAAhfVd6lGRMSwAAAAASUVORK5CYII=","orcid":"","institution":"Teerthanker Mahaveer University","correspondingAuthor":true,"prefix":"","firstName":"Rashid","middleName":"","lastName":"Ali","suffix":""},{"id":593784796,"identity":"9dba9ad3-1fd6-4371-ab76-e310c820fcba","order_by":3,"name":"Vishal Pal","email":"","orcid":"","institution":"Teerthanker Mahaveer University","correspondingAuthor":false,"prefix":"","firstName":"Vishal","middleName":"","lastName":"Pal","suffix":""}],"badges":[],"createdAt":"2026-02-18 07:38:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8906855/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8906855/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103050245,"identity":"59f27504-aad1-4234-9aed-2139e7238c8d","added_by":"auto","created_at":"2026-02-20 07:48:57","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":537525,"visible":true,"origin":"","legend":"","description":"","filename":"final.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8906855/v1_covered_09152228-b1f9-430a-bef4-1703cdfa3d40.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eLow-Dose Medical Image Reconstruction Using Residual U-Net:A Hybrid Approach Integrating Compressed Sensing and Deep Learning\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"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":"Low-Dose CT, MRI Reconstruction, Compressed Sensing, Deep Learning, U-Net, Residual Learning, Iterative Reconstruction, Inverse Problems","lastPublishedDoi":"10.21203/rs.3.rs-8906855/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8906855/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eLow-dose computed tomography (LDCT) and otherlow-dose medical imaging modalities are critically important for reducing patient radiation exposure andminimizing cancer risks. However, image reconstruction from severely undersampled or noisy measurements presents a significant ill-posed inverse problem, leading to degraded image quality characterizedby quantum noise, streaking artifacts, and loss ofstructural detail. This paper proposes a novel hybridframework that synergistically combines the theoretical foundations of Compressed Sensing (CS) with therepresentational power of deep learning, specificallythrough a Residual U-Net architecture. The proposed model operates within an iterative reconstruction paradigm, where a CS-based data fidelity step isalternated with a deep learning-based prior step, implemented by the Residual U-Net, to progressivelyrefine the image. The Residual U-Net, enhancedwith squeeze-and-excitation blocks, is designed tolearn the mapping from artifact-corrupted, low-doseimages to their high-dose counterparts, effectivelymodeling the complex noise and artifact distributions while preserving anatomical structures. We formulate the reconstruction as an optimization problem, solved via the Learned Primal-Dual algorithm.Extensive quantitative and qualitative evaluationsare performed on the publicly available Mayo ClinicLDCT dataset and a simulated undersampled MRIdataset (k-space data). Results demonstrate thatour proposed Residual U-Net integrated iterative reconstruction (RN-IR) method significantly outperforms traditional methods like Filtered Back Projection (FBP), total variation (TV)-based CS, andstandalone deep learning models (such as a vanilla UNet) in terms of Peak Signal-to-Noise Ratio (PSNR),Structural Similarity Index (SSIM), and visual quality. Our model achieves a PSNR of 42.3 dB and SSIMof 0.985 on the LDCT dataset, effectively suppressing noise and preserving fine details, thus providing arobust and efficient solution for\u0026nbsp; high-quality low-dose medical image reconstruction.\u003c/p\u003e","manuscriptTitle":"Low-Dose Medical Image Reconstruction Using Residual U-Net:A Hybrid Approach Integrating Compressed Sensing and Deep Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-20 02:05:53","doi":"10.21203/rs.3.rs-8906855/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":"c1c6e32c-4557-41e1-a280-84c314b5ff59","owner":[],"postedDate":"February 20th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-18T11:54:31+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-20 02:05:53","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8906855","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8906855","identity":"rs-8906855","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}