Deep learning-based tumor segmentation on digital images of histopathology slides for microdosimetry applications

Deep learning-based tumor segmentation on digital images of histopathology slides for microdosimetry applications | 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 Deep learning-based tumor segmentation on digital images of histopathology slides for microdosimetry applications Luca L. Weishaupt, Jose Torres, Sophie Camilleri-Broët, Roni F. Rayes, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-225323/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 The goal of this study was (i) to use artificial intelligence to automate the traditionally labor-intensive process of manual segmentation of tumor regions in pathology slides performed by a pathologist and (ii) to validate the use of a deep learning architecture. Automation will reduce the human error involved in the manual process, increase efficiency, and result in more accurate and reproducible segmentation. This advancement will alleviate the bottleneck in the workflow in clinical and research applications due to a lack of pathologist time. Our application is patient-specific microdosimetry and radiobiological modeling, which builds on the contoured pathology slides. A deep neural network named UNet was used to segment tumor regions in pathology core biopsies of lung tissue with adenocarcinoma stained using hematoxylin and eosin. A pathologist manually contoured the tumor regions in 56 images with binary masks for training. To overcome memory limitations overlapping and non-overlapping patch extraction with various patch sizes and image downsampling were investigated individually. Data augmentation was used to reduce overfitting and artificially create more data for training. Using this deep learning approach, the UNet achieved accuracy of 0.91±0.06, specificity of 0.90±0.08, sensitivity of 0.92±0.07, and precision of 0.8±0.1. The F1/DICE score was 0.85±0.07, with a segmentation time of 3.24±0.03 seconds per image, thus achieving a 370±3 times increased efficiency over manual segmentation, which took 20 minutes per image on average. In some cases, the neural network correctly delineated the tumor's stroma from its epithelial component in tumor regions that were classified as tumor by the pathologist. The UNet architecture can segment images with a level of efficiency and accuracy that makes it suitable for tumor segmentation of histopathological images in fields such as radiotherapy dosimetry, specifically in the subfields of microdosimetry. Nuclear Medicine & Medical Imaging Theoretical Computer Science Oncology deep learning tumor segmentation histopathology Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Full Text Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterials.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-225323","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":13304758,"identity":"dd15eebf-aa84-4619-8ded-334bf779d8e1","order_by":0,"name":"Luca L. Weishaupt","email":"data:image/png;base64,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","orcid":"","institution":"Medical Physics Unit, Department of Oncology, Faculty of Medicine, McGill University, Montréal, Québec","correspondingAuthor":true,"prefix":"","firstName":"Luca","middleName":"L.","lastName":"Weishaupt","suffix":""},{"id":13304759,"identity":"b904eeea-fb9f-41a9-b38e-a1925a3717c9","order_by":1,"name":"Jose Torres","email":"","orcid":"","institution":"Department of Pathology, Faculty of Medicine, McGill University, Montréal, Québec","correspondingAuthor":false,"prefix":"","firstName":"Jose","middleName":"","lastName":"Torres","suffix":""},{"id":13304760,"identity":"591644c7-5769-4d7e-80dd-ac8a456cf073","order_by":2,"name":"Sophie Camilleri-Broët","email":"","orcid":"","institution":"Department of Pathology, Faculty of Medicine, McGill University, Montréal, Québec","correspondingAuthor":false,"prefix":"","firstName":"Sophie","middleName":"","lastName":"Camilleri-Broët","suffix":""},{"id":13304761,"identity":"a4537745-9205-436a-9ffc-537970739333","order_by":3,"name":"Roni F. Rayes","email":"","orcid":"","institution":"Cancer Research Program and the LD MacLean Surgical Research Laboratories, Department of Surgery, Division of Upper GI and Thoracic Surgery, Research Institute of the McGill University Health Center","correspondingAuthor":false,"prefix":"","firstName":"Roni","middleName":"F.","lastName":"Rayes","suffix":""},{"id":13304762,"identity":"751ceb86-5038-4e0d-a79d-9bcf731349ea","order_by":4,"name":"Jonathan D. Spicer","email":"","orcid":"","institution":"Cancer Research Program and the LD MacLean Surgical Research Laboratories, Department of Surgery, Division of Upper GI and Thoracic Surgery, Research Institute of the McGill University Health Center","correspondingAuthor":false,"prefix":"","firstName":"Jonathan","middleName":"D.","lastName":"Spicer","suffix":""},{"id":13304763,"identity":"9b2f232a-adfe-4406-903f-1421825b8353","order_by":5,"name":"Sabrina Côté Maldonado","email":"","orcid":"","institution":"Medical Physics Unit, Department of Oncology, Faculty of Medicine, McGill University, Montréal, Québec","correspondingAuthor":false,"prefix":"","firstName":"Sabrina","middleName":"Côté","lastName":"Maldonado","suffix":""},{"id":13304764,"identity":"877222f3-cca0-470c-b6a9-d219235246d2","order_by":6,"name":"Shirin A. Enger","email":"","orcid":"","institution":"Medical Physics Unit, Department of Oncology, Faculty of Medicine, McGill University, Montréal, Québec","correspondingAuthor":false,"prefix":"","firstName":"Shirin","middleName":"A.","lastName":"Enger","suffix":""}],"badges":[],"createdAt":"2021-02-09 04:29:03","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-225323/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-225323/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":6309380,"identity":"6f9c31df-f0f4-485e-aaca-6e65f664d787","added_by":"auto","created_at":"2021-02-24 15:03:05","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":93358,"visible":true,"origin":"","legend":"The UNet architecture employed for image segmentation of 1872x1872 pixel patches. The\nimage patch with its normalized RGB channels undergoes four convolutions, max-pooling, and\ndropout iterations for feature extraction. The resulting feature maps are then up-convoluted and\nconcatenated with copies of the feature maps from the contracting path to retain location information\nfour times. A convolutional block consists of a convolution with a 3x3 kernel, batch normalization,\nand a ReLU activation function. In the final step, the feature maps are convoluted with a 1x1 filter\nand a sigmoid activation function to create a normalized mask for the input image patch.","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-225323/v1/0ab37de876399601683a2417.jpg"},{"id":6309060,"identity":"813f72dc-df23-4b19-acda-31b6eb7ad379","added_by":"auto","created_at":"2021-02-24 15:00:05","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":257505,"visible":true,"origin":"","legend":"Overlapping patch extraction with 1872-pixel patches. The image was converted to black\nand white for this illustration. Patch extraction begins at the top left corner. The black square\nindicates the first patch. A horizontal stride is then taken before the next patch is extracted. The\nhorizontal stride and the subsequent patch are illustrated in yellow. After all patches in one row have\n been extracted, a vertical stride is taken. The vertical stride and the first patch in the second row are\nillustrated in red. This process is repeated until the entire image area has been extracted. Nine\npatches were extracted for this image. Dashed lines are used to indicate overlapping patch edges.","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-225323/v1/15cdd59fffc32a86bf058362.jpg"},{"id":6308359,"identity":"313e9b58-5c6f-4a8c-8d6c-2c40c3654eeb","added_by":"auto","created_at":"2021-02-24 14:57:05","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":165237,"visible":true,"origin":"","legend":"Representative example of a pathology core. A – The original image. B – The contour\nmade by the pathologist. Tumor regions are contoured in red. C – Prediction map generated by the\nUNet algorithm with 1872x1872 pixel patches and 939-pixel strides (case 2). D – The prediction\nmap after applying a 50% confidence threshold. High values indicate a high probability of tumor.","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-225323/v1/d0cc1b42af3ad6f02c0b5ee4.jpg"},{"id":6308362,"identity":"2db3c37a-07ae-48e5-af59-d6265cb60ace","added_by":"auto","created_at":"2021-02-24 14:57:05","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":205780,"visible":true,"origin":"","legend":"Segmented tumor regions from each of the 7 different test cases. GT is the ground truth\ncontour created by the pathologist. The rest of the indices indicated in each pane correspond with a\ncase index from Table 2.","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-225323/v1/201d3deecd1ae7b5ec4ac41d.jpg"},{"id":6309062,"identity":"9cbb9187-6a10-49c8-b19c-c91b5a509eed","added_by":"auto","created_at":"2021-02-24 15:00:05","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":158812,"visible":true,"origin":"","legend":"Zoomed in example of the stroma segmented in the image presented in Figure 3. A – The\nimage used as input for the UNet. B – The contour made by the pathologist. Tumor regions are\ncontoured in dark red. All other regions are contoured in dark blue. Note, the pathologist contoured\nthe stroma as part of a tumor region. C – Prediction map generated by the test case with 1872x1872 pixel patches and 939-pixel strides (case 2). D – The prediction map after applying a 50% confidence\nthreshold.","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-225323/v1/655499af8185900629083684.jpg"},{"id":13592350,"identity":"13d78d3a-c76d-459e-b688-3bae4315f1b2","added_by":"auto","created_at":"2021-09-17 05:12:35","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2017490,"visible":true,"origin":"","legend":"","description":"","filename":"ManuscriptDeepLearningForTumorSegmentation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-225323/v1_covered.pdf"},{"id":6309759,"identity":"f34f233e-359f-4c23-9c0c-8039d22f7bca","added_by":"auto","created_at":"2021-02-24 15:06:16","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2412366,"visible":true,"origin":"","legend":"","description":"","filename":"ManuscriptDeepLearningForTumorSegmentation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-225323/v1_stamped.pdf"},{"id":6308364,"identity":"c774bc38-f49d-4813-8480-32ac43bf569a","added_by":"auto","created_at":"2021-02-24 14:57:05","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":817278,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterials.pdf","url":"https://assets-eu.researchsquare.com/files/rs-225323/v1/5ff57ad7d61fd3f1ae7e8411.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Deep learning-based tumor segmentation on digital images of histopathology slides for microdosimetry applications","fulltext":[{"header":"Full Text","content":"\u003cp\u003eThis preprint is available for \u003ca href='/article/rs-225323/latest.pdf' target='_blank'\u003edownload as a PDF\u003c/a\u003e.\u003c/p\u003e"}],"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":false,"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":"deep learning, tumor segmentation, histopathology","lastPublishedDoi":"10.21203/rs.3.rs-225323/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-225323/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"The goal of this study was (i) to use artificial intelligence to automate the traditionally labor-intensive process of manual segmentation of tumor regions in pathology slides performed by a pathologist and (ii) to validate the use of a deep learning architecture. Automation will reduce the human error involved in the manual process, increase efficiency, and result in more accurate and reproducible segmentation. This advancement will alleviate the bottleneck in the workflow in clinical and research applications due to a lack of pathologist time. Our application is patient-specific microdosimetry and radiobiological modeling, which builds on the contoured pathology slides. A deep neural network named UNet was used to segment tumor regions in pathology core biopsies of lung tissue with adenocarcinoma stained using hematoxylin and eosin. A pathologist manually contoured the tumor regions in 56 images with binary masks for training. To overcome memory limitations overlapping and non-overlapping patch extraction with various patch sizes and image downsampling were investigated individually. Data augmentation was used to reduce overfitting and artificially create more data for training. Using this deep learning approach, the UNet achieved accuracy of 0.91±0.06, specificity of 0.90±0.08, sensitivity of 0.92±0.07, and precision of 0.8±0.1. The F1/DICE score was 0.85±0.07, with a segmentation time of 3.24±0.03 seconds per image, thus achieving a 370±3 times increased efficiency over manual segmentation, which took 20 minutes per image on average. In some cases, the neural network correctly delineated the tumor's stroma from its epithelial component in tumor regions that were classified as tumor by the pathologist. The UNet architecture can segment images with a level of efficiency and accuracy that makes it suitable for tumor segmentation of histopathological images in fields such as radiotherapy dosimetry, specifically in the subfields of microdosimetry.","manuscriptTitle":"Deep learning-based tumor segmentation on digital images of histopathology slides for microdosimetry applications","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2021-02-24 14:57:03","doi":"10.21203/rs.3.rs-225323/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":"c3d16fe1-0767-4de4-aff5-4bc276b02e1b","owner":[],"postedDate":"February 24th, 2021","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":2604915,"name":"Nuclear Medicine \u0026 Medical Imaging"},{"id":2604916,"name":"Theoretical Computer Science"},{"id":2604917,"name":"Oncology"}],"tags":[],"updatedAt":"2021-04-16T11:44:13+00:00","versionOfRecord":[],"versionCreatedAt":"2021-02-24 14:57:03","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-225323","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-225323","identity":"rs-225323","version":["v1"]},"buildId":"_2-kVJe1T_tPrBINL-cwx","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}