Federated Task-Adaptive Learning for Personalized Selection of Human IVF-derived Embryos

Federated Task-Adaptive Learning for Personalized Selection of Human IVF-derived Embryos | 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 Federated Task-Adaptive Learning for Personalized Selection of Human IVF-derived Embryos Guangyu Wang, Tianrun Gao, Yuning Yang, Kai Wang, Yuanxu Gao, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4631058/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 18 Nov, 2025 Read the published version in Communications Medicine → Version 1 posted You are reading this latest preprint version Abstract In-vitro fertilization (IVF) offers a solution for couples facing infertility challenges. However, the success of IVF, particularly in achieving live-birth outcomes, heavily depends on embryologists to conduct morphological assessments of fertilized embryos, a process that is both time-consuming and labor-intensive. While artificial intelligence (AI) has gained recognition for its potential to automate embryo selection, the application of deep learning (DL) is constrained by privacy concerns associated with the requirement for centralized training on extensive datasets. In this paper, we have developed a distributed DL system, termed ‘FedEmbryo’, tailored for personalized embryo selection while preserving data privacy. Within FedEmbryo, we introduce a Federated Task-Adaptive Learning (FTAL) approach with a hierarchical dynamic weighting adaption (HDWA) mechanism. This approach first uniquely integrates multi-task learning (MTL) with federated learning (FL) by proposing a unified multitask client architecture that consists of shared layers and task-specific layers to accommodate the single- and multi-task learning within each client. Furthermore, the HDWA mechanism mitigates the skewed model performance attributed to data heterogeneity from FTAL. It considers the learning feedback (loss ratios) from the tasks and clients, facilitating a dynamic balance to task attention and client aggregation. Finally, we refine FedEmbryo to address critical clinical scenarios in the IVF processes, including morphology evaluation and live-birth outcomes. We operate each morphological metric as an individual task within the client's model to perform FTAL in morphology evaluation and incorporate embryo images with corresponding clinical factors as multimodal inputs to predict live-birth outcomes. Experimental results indicate that FedEmbryo outperforms both locally trained models and state-of-the-art (SOTA) FL methods. Our research marks a significant advancement in the development of AI in IVF treatments. Biological sciences/Physiology/Reproductive biology/Reproductive disorders/Infertility Health sciences/Diseases/Reproductive disorders/Infertility Figures Figure 1 Figure 2 Figure 3 Figure 4 Full Text Additional Declarations There is NO Competing Interest. Supplementary Files FedEmbryoSupplementaryInformation.pdf Cite Share Download PDF Status: Published Journal Publication published 18 Nov, 2025 Read the published version in Communications Medicine → 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-4631058","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":333740915,"identity":"4ce236e4-07b7-4bf6-a5fa-974a219ef42a","order_by":0,"name":"Guangyu Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCUlEQVRIiWNgGAWjYNCDDzw2QJKx8QDROhhnyKSBqAbitTDz2BwGM/BqkY9Ifvbwa9thBoPjZw+/4Mk5b7e2/TDQlhqbaFxaDG+kmRvLgrScyUuzkDhzO3nbmUSglmNpuQ24tMxIMJOWBGoxO5BjZmDYczvZ7ABQC2PDYTxa0r9BtJx/Y2aQ+O9cstn5h/i1yEvkmEl+BGm5kWP84ADPATuzGwRsMeB5UybNcC6dx/7GGzPGBp7kBLMbQFsS8PhFvj19m+SPMms5yf4c489/eOzszc6nP3zwocYGty0HgNHBy8bAA2SzSQCJRLDKBBzKwbYAVTD++ANmM38AEvZ4FI+CUTAKRsEIBQDi9WaYbKbkfAAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0003-1312-884X","institution":"Beijing University of Posts and Telecommunications","correspondingAuthor":true,"prefix":"","firstName":"Guangyu","middleName":"","lastName":"Wang","suffix":""},{"id":333740916,"identity":"bf9be4bb-483c-4e36-9d80-a6e141a25c67","order_by":1,"name":"Tianrun Gao","email":"","orcid":"","institution":"Beijing University of Posts and Telecommunications","correspondingAuthor":false,"prefix":"","firstName":"Tianrun","middleName":"","lastName":"Gao","suffix":""},{"id":333740917,"identity":"9f3d3e09-23fc-4c3a-a5e2-2e4189658773","order_by":2,"name":"Yuning Yang","email":"","orcid":"","institution":"Beijing University of Posts and Telecommunications","correspondingAuthor":false,"prefix":"","firstName":"Yuning","middleName":"","lastName":"Yang","suffix":""},{"id":333740918,"identity":"14947f40-26e7-4cf4-b703-bb62ece4ec61","order_by":3,"name":"Kai Wang","email":"","orcid":"","institution":"Peking University and Peking-Tsinghua Center for Life Sciences","correspondingAuthor":false,"prefix":"","firstName":"Kai","middleName":"","lastName":"Wang","suffix":""},{"id":333740919,"identity":"b2a531ec-5692-4932-ab79-042cc0e70ad9","order_by":4,"name":"Yuanxu Gao","email":"","orcid":"https://orcid.org/0000-0001-5314-0195","institution":"Macau University of Science and Technology and University Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yuanxu","middleName":"","lastName":"Gao","suffix":""},{"id":333740920,"identity":"66f9f59f-8803-42f0-87c5-b36ed6a05b8c","order_by":5,"name":"Li-Shuang Ma","email":"","orcid":"","institution":"Capital Institute of Pediatrics, Affiliated Children's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Li-Shuang","middleName":"","lastName":"Ma","suffix":""},{"id":333740921,"identity":"8baf78b4-6504-4d58-8478-6c15f86d8820","order_by":6,"name":"Lei Chen","email":"","orcid":"","institution":"Department of Gynaecology and Obstetrics，The Sixth Medical Center of the General Hospital of the People's Liberation Army","correspondingAuthor":false,"prefix":"","firstName":"Lei","middleName":"","lastName":"Chen","suffix":""},{"id":333740922,"identity":"90ea3746-b149-4ce9-bb88-20961d674e4c","order_by":7,"name":"Guangdong Liu","email":"","orcid":"","institution":"Department of Gynaecology and Obstetrics，The Sixth Medical Center of the General Hospital of the People's Liberation Army","correspondingAuthor":false,"prefix":"","firstName":"Guangdong","middleName":"","lastName":"Liu","suffix":""},{"id":333740923,"identity":"7bef38df-892b-42af-9b10-1a92cec23fc8","order_by":8,"name":"Ping Zhang","email":"","orcid":"","institution":"Beijing University of Posts and Telecommunications","correspondingAuthor":false,"prefix":"","firstName":"Ping","middleName":"","lastName":"Zhang","suffix":""},{"id":333740924,"identity":"a34b943e-e224-457c-8caf-9d94541a6b1e","order_by":9,"name":"Xiaohong Liu","email":"","orcid":"","institution":"UCL Cancer Institute, University College London","correspondingAuthor":false,"prefix":"","firstName":"Xiaohong","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2024-06-24 15:00:32","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4631058/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4631058/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s43856-025-01182-1","type":"published","date":"2025-11-18T05:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":64004640,"identity":"18f2a125-6023-4e5a-8653-955be6b1a6e7","added_by":"auto","created_at":"2024-09-04 21:27:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":64581,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOverview of FedEmbryo workflow, architecture, development and evaluation.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea\u003c/strong\u003e, Schematic of the FedEmbryo. FedEmbryo is a Federated Task-Adaptive Learning (FTAL), which focuses on the client processing multiple tasks simultaneously. We build four private clients collaboratively training the model without any data sharing. For each communication round, every client transmits its local model and the corresponding loss ratio to the server. The server then aggregates these local models and redistributes the updated model to clients. \u003cstrong\u003eb\u003c/strong\u003e, In FedEmbryo, we introduce the hierarchical dynamic weighting adaptation (HDWA) mechanism to dynamic balance weight coefficient at both client and task levels. The server (Upper) assigns the aggregated weight based on the loss ratio λt, derived from previous t-1 \u0026nbsp;to t-2 communication round. Unlike traditional approaches where client weight remains fixed, the HDWA mechanism dynamically balances the weight based on client performance in each training rounds. At client level (Bottom), the framework manages complex clinical practices, such as morphology assessment (Bottom Left) and prediction of live-birth outcomes (Bottom Right). We utilize the HDWA mechanism to balance the weights to various tasks—such as pronuclear features, symmetry, cell count, fragmentation rate, and blastocyst formation—based on the loss ratios from the two previous local epochs. We integrate images and clinical factors as multimodal input to improve the prediction of the live-birth outcomes. \u003cstrong\u003ec\u003c/strong\u003e, Datasets Description (N is the number of patients).\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-4631058/v1/f153fb16377fdde9abcb1ed1.png"},{"id":64004973,"identity":"9655c8fe-ba5f-4641-aca3-1f8c7a69a110","added_by":"auto","created_at":"2024-09-04 21:35:07","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":153853,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEffectiveness of the hierarchical dynamic weighting adaption (HDWA) on local Clients.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIllustration of the label distribution across four clients and the comparative performance metrics for two models: Local (multitask) and FedEmbryo. \u003cstrong\u003ea\u003c/strong\u003e, The distribution of labels by client, highlighting the variance in dataset sizes among Client A, B, C, and D.\u003cstrong\u003e \u003c/strong\u003eEach client possesses five types of labels including day1 pronuclear (abnormal), day3 symmetry, fragmentation, number of cells, and day5 blastocyst formation. \u003cstrong\u003eb-e\u003c/strong\u003e, Performance metrics for the Local and FedEmbryo models across four clients: (\u003cstrong\u003eb\u003c/strong\u003e) Client A; (\u003cstrong\u003ec\u003c/strong\u003e) Client B; (\u003cstrong\u003ed\u003c/strong\u003e) Client C; (\u003cstrong\u003ee\u003c/strong\u003e) Client D. Error bar indicates 95% confidence intervals. (++) extremely asymmetry.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-4631058/v1/36dc409e00726f12f4911f3f.png"},{"id":64004641,"identity":"de484480-5e24-4793-b869-25d4a743d25c","added_by":"auto","created_at":"2024-09-04 21:27:07","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":135189,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePerformance of the live-birth outcomes.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea\u003c/strong\u003e, AUC score shows performance on live-birth outcomes with the image-only models.\u003cstrong\u003e \u003c/strong\u003eThe columns list cohort and different approaches. Cohort includes internal and external test sets. Approaches are: Local scenario, federated baselines (FedAvg, FedProx), our approach (FedEmbryo), Centralized scenarios. The rows list the average performance on internal test set and external test sets, separately. \u003cstrong\u003eb-c\u003c/strong\u003e, ROC curve shows the performance of live-birth outcomes on multimodality including: FedAvg (image), FedProx (image), FedEmbryo (image), FedEmbryo (metadata), and FedEmbryo (combined). (\u003cstrong\u003eb\u003c/strong\u003e) external test set (Cohort E); (\u003cstrong\u003ec\u003c/strong\u003e) external test set (Cohort F).\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-4631058/v1/cc51db5e132a1d04f0a0922a.png"},{"id":64004643,"identity":"4b2afb37-eeb8-4d34-bbe5-403a994538de","added_by":"auto","created_at":"2024-09-04 21:27:07","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":987473,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGradient visualization and interpretability of the FedEmbryo.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea\u003c/strong\u003e, Visual explanations highlight the areas of an image that are most important for a model's prediction. The row lists the representative embryo images of the day 1 and day 3 during the embryo development. The first column on the left is the original embryo images. The rest columns represent the heatmap overlaying the original image generated by Client A; Client B; Client C; Client D; FedEmbryo. \u003cstrong\u003eb-c\u003c/strong\u003e, Illustration of the contribution of each factor to the model's prediction of live-birth outcomes by SHAP visualization analysis on (\u003cstrong\u003ea\u003c/strong\u003e) mean absolute SHAP value of each factor; (\u003cstrong\u003eb\u003c/strong\u003e) detailed SHAP value of each factor.\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-4631058/v1/4271013abf883108e2ad536a.png"},{"id":96262871,"identity":"180658be-2ebb-4097-a217-5fd8072e2ecb","added_by":"auto","created_at":"2025-11-19 08:11:40","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1023679,"visible":true,"origin":"","legend":"","description":"","filename":"FedEmbryoManuscirpt.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4631058/v1_covered_cf9190f2-1dce-4f73-abf5-9c6b807ef515.pdf"},{"id":64004644,"identity":"b3f036de-4774-4b6c-8104-d32dd270a22d","added_by":"auto","created_at":"2024-09-04 21:27:07","extension":"pdf","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":1626028,"visible":true,"origin":"","legend":"","description":"","filename":"FedEmbryoSupplementaryInformation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4631058/v1/885e1bb8dd9176163bc22981.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Federated Task-Adaptive Learning for Personalized Selection of Human IVF-derived Embryos","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-4631058/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4631058/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn-vitro fertilization (IVF) offers a solution for couples facing infertility challenges. However, the success of IVF, particularly in achieving live-birth outcomes, heavily depends on embryologists to conduct morphological assessments of fertilized embryos, a process that is both time-consuming and labor-intensive. While artificial intelligence (AI) has gained recognition for its potential to automate embryo selection, the application of deep learning (DL) is constrained by privacy concerns associated with the requirement for centralized training on extensive datasets. In this paper, we have developed a distributed DL system, termed \u0026lsquo;FedEmbryo\u0026rsquo;, tailored for personalized embryo selection while preserving data privacy. Within FedEmbryo, we introduce a Federated Task-Adaptive Learning (FTAL) approach with a hierarchical dynamic weighting adaption (HDWA) mechanism. This approach first uniquely integrates multi-task learning (MTL) with federated learning (FL) by proposing a unified multitask client architecture that consists of shared layers and task-specific layers to accommodate the single- and multi-task learning within each client. Furthermore, the HDWA mechanism mitigates the skewed model performance attributed to data heterogeneity from FTAL. It considers the learning feedback (loss ratios) from the tasks and clients, facilitating a dynamic balance to task attention and client aggregation. Finally, we refine FedEmbryo to address critical clinical scenarios in the IVF processes, including morphology evaluation and live-birth outcomes. We operate each morphological metric as an individual task within the client's model to perform FTAL in morphology evaluation and incorporate embryo images with corresponding clinical factors as multimodal inputs to predict live-birth outcomes. Experimental results indicate that FedEmbryo outperforms both locally trained models and state-of-the-art (SOTA) FL methods. Our research marks a significant advancement in the development of AI in IVF treatments.\u003c/p\u003e","manuscriptTitle":"Federated Task-Adaptive Learning for Personalized Selection of Human IVF-derived Embryos","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-09-04 21:27:02","doi":"10.21203/rs.3.rs-4631058/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"communications-medicine","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"commsmed","sideBox":"Learn more about [Communications Medicine](http://www.nature.com/commsmed)","snPcode":"43856","submissionUrl":"https://mts-commsmed.nature.com/cgi-bin/main.plex","title":"Communications Medicine","twitterHandle":"@commsmedicine","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Communications Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"035bf646-b834-48ff-a1ff-5f1e9ec188e6","owner":[],"postedDate":"September 4th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":35341471,"name":"Biological sciences/Physiology/Reproductive biology/Reproductive disorders/Infertility"},{"id":35341472,"name":"Health sciences/Diseases/Reproductive disorders/Infertility"}],"tags":[],"updatedAt":"2025-11-19T08:11:31+00:00","versionOfRecord":{"articleIdentity":"rs-4631058","link":"https://doi.org/10.1038/s43856-025-01182-1","journal":{"identity":"communications-medicine","isVorOnly":false,"title":"Communications Medicine"},"publishedOn":"2025-11-18 05:00:00","publishedOnDateReadable":"November 18th, 2025"},"versionCreatedAt":"2024-09-04 21:27:02","video":"","vorDoi":"10.1038/s43856-025-01182-1","vorDoiUrl":"https://doi.org/10.1038/s43856-025-01182-1","workflowStages":[]},"version":"v1","identity":"rs-4631058","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4631058","identity":"rs-4631058","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}