[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86208-en":3,"doc-seo-86208-105":29,"detail-sidebar-cat-0-en-105":82},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},86208,1374391974564,"Clementine","https://ap-avatar.wpscdn.com/avatar/14000253aa45c000a9e?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779874745381141002",7,"Healthcare","Metadata Supervised MRI Representations for Modelling and Controlling Acquisition Variability","Magnetic resonance imaging shows substantial acquisition variability, where the same anatomy can look different across scanners and protocols. Learned representations often entangle biological structure with acquisition-dependent appearance, weakening interpretability, generalisation, and clinical deployment. The work jointly models MRI images and DICOM metadata to separate anatomical structure from contrast-driven appearance. Using large-scale clinical brain MRI data, it learns contrast representations that organise heterogeneous acquisitions, support sequence understanding, and flag image–metadata inconsistencies, while anatomical representations suppress acquisition-specific variation while preserving biologically relevant information. Building on this, a unified anatomy-preserving harmonisation model enables cross-modality and cross-site adaptation, conditioned on image or acquisition metadata.","arXiv :2607 . 11295v1 [ cs .CV] 13 Jul 2026  \nMetadata Supervised MRI Representations for Modelling and Controlling Acquisition Variability  \nMehmet Yigit Avci 1*, Pedro Borges 1 , Virginia Fernandez 1 , Natalia Glazman 1 , Paul Wright2 , Mehmet Yigitsoy3 , Sebastien Ourselin 1 , M. Jorge Cardoso 1  \n1 School of Biomedical Engineering & Imaging Sciences, King’s College London, London, UK.  \n2 University College London, London, UK.  \n3 deepc GMBH, Munich, Germany.  \n*Corresponding author(s). E-mail(s): [yigit.avci@kcl.ac.uk](yigit.avci@kcl.ac.uk) ; Contributing authors: [pedro.borges@kcl.ac.uk](pedro.borges@kcl.ac.uk) ;  \n[virginia.fernandez@kcl.ac.uk](virginia.fernandez@kcl.ac.uk) ; [natalia.glazman@kcl.ac.uk](natalia.glazman@kcl.ac.uk) ;  \n[paul.w@ucl.ac.uk](paul.w@ucl.ac.uk) ; [mehmet.yigitsoy@gmail.com](mehmet.yigitsoy@gmail.com) ; [sebastien.ourselin@kcl.ac.uk](sebastien.ourselin@kcl.ac.uk) ; [m.jorge.cardoso@kcl.ac.uk](m.jorge.cardoso@kcl.ac.uk) ;  \nAbstract  \nMagnetic resonance imaging exhibits substantial acquisition variability, where identical anatomy can appear markedly different across scanners and imaging protocols. Consequently, learned representations entangle biological structure with acquisition-dependent appearance, limiting interpretability, generalisation, and clinical deployment. We show that these sources of variation can be separated by jointly modelling MRI images and DICOM metadata. Using large-scale clinical brain MRI data, we learn representations that separate anatomical structure from contrast-dependent appearance. Resulting contrast representations organise heterogeneous acquisitions, support sequence understanding, and detect image–metadata inconsistencies, whereas anatomical representations suppress acquisition-specific variation while preserving biologically relevant information. Building on these disentangled representations, we introduce a unified anatomypreserving harmonisation model for cross-modality and cross-site adaptation, conditioned on image or acquisition metadata. Our findings suggest that acquisition variability is a structured component of the imaging process that can be modelled, audited, and controlled, providing a foundation for acquisition-aware representation learning in large-scale medical imaging.  \n1  \nKeywords: Magnetic resonance imaging, DICOM metadata, Representation learning, Domain shift, Image harmonisation, Contrastive learning, Quality control  \n1 Introduction  \nMagnetic resonance imaging (MRI) is a cornerstone of neuroimaging, enabling noninvasive visualisation and quantitative assessment of brain structure and pathology in both clinical practice and neuroscience research [1] . However, despite rapid progress in machine learning for medical image analysis, robust deployment of automated MRI models across institutions remains challenging. A key limitation is acquisition shift, where differences in scanner hardware, field strength, pulse sequence design, reconstruction pipelines, and protocol settings induce substantial changes in image appearance without corresponding changes in underlying anatomy [2] . Unlike natural images, MRI data are not solely determined by the object being imaged; instead, image appearance reflects a complex interaction between biological structure and the acquisition process. As a result, the same brain can exhibit markedly different intensity distributions, textures, and contrast under different acquisition conditions, while visually similar scans may originate from distinct protocols rather than shared biology [3–5] . When learned representations entangle acquisition-specific factors with anatomical content, downstream models become sensitive to domain shifts, limiting their generalisability and hindering reliable deployment across scanners, institutions, and patient populations [6, 7] .  \nThis challenge reflects a broader limitation of current MRI representation learning. Existing approaches address acquisition variability through two ma","cbCaij2eEFIY7IDQ","https://ap.wps.com/l/cbCaij2eEFIY7IDQ","pdf",36255546,1,36,"English","en",105,"# Abstract\n# Keywords\n# Introduction\n## Acquisition shift and representation entanglement\n## Existing strategies: domain generalisation and harmonisation\n## Role of DICOM metadata","[{\"question\":\"What are the main outcomes of the learned contrast and anatomical representations?\",\"answer\":\"Contrast representations organise heterogeneous acquisitions, enable sequence understanding, and detect inconsistencies between images and metadata. Anatomical representations suppress acquisition-specific variation while preserving biologically relevant information, supporting controlled harmonisation and cross-site adaptation.\"}]",1784209471,91,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":77,"head_meta":79,"extra_data":81,"updated_unix":27},"metadata-supervised-mri-representations-for-modelling-and-controlling-acquisition-variability","",{"@graph":35,"@context":76},[36,53,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/healthcare/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/metadata-supervised-mri-representations-for-modelling-and-controlling-acquisition-variability/86208/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70],{"name":71,"@type":72,"acceptedAnswer":73},"What are the main outcomes of the learned contrast and anatomical representations?","Question",{"text":74,"@type":75},"Contrast representations organise heterogeneous acquisitions, enable sequence understanding, and detect inconsistencies between images and metadata. Anatomical representations suppress acquisition-specific variation while preserving biologically relevant information, supporting controlled harmonisation and cross-site adaptation.","Answer","https://schema.org",{"og:url":51,"og:type":78,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":80,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":83},[84,88,92,96,101,106,109,114,119,122,126],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":45,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":107,"slug":108},40,"healthcare",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},8,"Research & Report",30,"research-report",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},9,"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":45,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general"]