[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82490-en":3,"doc-seo-82490-105":29,"detail-sidebar-cat-0-en-105":91},{"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":20,"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},82490,1099513958762,"Logic","https://ap-avatar.wpscdn.com/avatar/1000023916a998db790?x-image-process=image/resize,m_fixed,w_180,h_180&k=1784791008015729253",8,"Research & Report","AEGIS Multi-Task Joint-Embedding Predictive Architecture for Mammography","AEGIS presents a joint-embedding predictive architecture for breast cancer detection and mammographic density assessment. The method trains Vision Transformer variants (Small/Base/Large) using self-supervised JEPA pre-training on 71,103 studies from 14 clinical sites, then applies supervised fine-tuning with progressive resolution scaling to 2048×1536. On a curated 785-study test set, the largest model reaches AUC 0.949 for triage (93% sensitivity, 75% specificity), and AUC 0.952 with an ensemble using an FDA-cleared baseline. For density classification it achieves 0.953 AUC for dense vs non-dense and 62.6% exact accuracy across BI-RADS categories, with 98.8% adjacent accuracy; external VinDr-Mammo validation shows 0.871 AUC in zero-shot triage.","AEGIS: A Multi-Task Joint-Embedding Predictive Architecture for Mammography  \nScott Chase Waggener, Sai Karthik Navuluru, and Lakshman Tamil  \nDepartment of Electrical and Computer Engineering  \nUniversity of Texas at Dallas  \nRichardson, TX  \n[scott.waggener@utdallas.edu](scott.waggener@utdallas.edu) ; [SaiKarthik.Navuluru@UTDallas.edu](SaiKarthik.Navuluru@UTDallas.edu) ; [laxman@utdallas.edu](laxman@utdallas.edu)  \narXiv :2607 .00277v1 [ cs .CV] 30 Jun 2026  \nAbstract—We present Aegis, a joint-embedding predictive architecture for breast cancer detection and density assessment in mammography. We train three Vision Transformer variants (Small/Base/Large) using self-supervised joint-embedding predictive architecture (JEPA) pre-training on 71,103 studies from 14 clinical sites, followed by supervised fine-tuning with progressive resolution scaling up to 2048 ×1536. On a curated 785-study test set, our largest model achieves area under the receiver operating characteristic curve (AUC) 0.949 for breast cancer triage with 93% sensitivity and 75% specificity at the optimal operating point. An ensemble combining our model with a U.S. Food and Drug Administration-cleared baseline further improves discrimination to 0.952 AUC. For breast density classification, the model achieves 0.953 AUC for binary (dense vs. non-dense) classification and 62.6% exact accuracy across four Breast Imaging Reporting and Data System (BI-RADS) categories, with 98.8% adjacent accuracy comparable to reported human inter-reader agreement. External validation on the public VinDr-Mammo dataset provides evidence of cross-population transfer under a different reference standard, with the largest model achieving 0.871 AUC for triagein a zero-shot setting.  \nIndex Terms—Deep Learning, Computer Vision, SelfSupervised Learning, JEPA, mammography.  \nI. INTRODUCTION  \nBreast cancer remains one of the most prevalent and deadly malignancies worldwide, posing a significant public health challenge. In 2022, approximately 2.3 million women were diagnosed with breast cancer globally, resulting in an estimated 670,000 deaths [1] . This disease disproportionately affects women in lower-resource settings, where access to early detection and treatment is limited, leading to higher mortality rates; notably, the World Health Organization (WHO) reports that approximately 80% of breast cancers occur in women with no specific risk factors other than sex and age [2], underscoring the limits of risk-factor-based prevention alone. Incidence rates vary by region and socioeconomic development, with projections indicating a 38% increase in cases anda 68% rise in deaths by 2050 if current trends persist [3] . The harm extends beyond mortality, encompassing treatmentrelated complications and long-term survivorship burdens for patients, families, and healthcare systems.  \nArtificial intelligence (AI) offers a promising path to address these challenges by augmenting radiologist capabilities in breast cancer screening. Recent studies have demonstrated that AI systems can improve cancer detection rates while  \nreducing false positives in clinical mammography workflows. AI also enables automated breast density assessment and risk stratification, tasks traditionally subject to substantial interreader variability among radiologists. Section II reviews prior work on AI for mammography and the self-supervised learning methods that motivate our approach.  \nJoint-Embedding Predictive Architecture (JEPA) represents an advanced self-supervised learning paradigm designed to learn semantic representations from data without relying on hand-crafted augmentations or generative reconstruction. In the image-based JEPA (I-JEPA) variant [4], the model predicts latent representations of target image blocks from a context encoder, promoting abstract, predictive understanding of visual structures while avoiding pitfalls like collapsed representations common in contrastive methods. In this paper, we leverage JEPA-pre","cbCaijoEFv34OSNK","https://ap.wps.com/l/cbCaijoEFv34OSNK","pdf",487247,1,11,"English","en",105,"# Abstract\n# Introduction\n# Related Work\n## Self-Supervised Learning in Computer Vision","[{\"question\":\"What is AEGIS and what tasks does it address in mammography?\",\"answer\":\"AEGIS is a joint-embedding predictive architecture for breast cancer detection and mammographic density assessment. It is designed to support triage, detection-related objectives, and density classification using Vision Transformers.\"},{\"question\":\"How is the model trained before fine-tuning?\",\"answer\":\"The approach uses self-supervised JEPA pre-training with three Vision Transformer variants trained on 71,103 studies from 14 clinical sites. It is then followed by supervised fine-tuning with progressive resolution scaling up to 2048×1536.\"},{\"question\":\"What performance does the largest model achieve on the test set and in external validation?\",\"answer\":\"On the curated 785-study test set, the largest model achieves AUC 0.949 for breast cancer triage (93% sensitivity, 75% specificity). For external validation on VinDr-Mammo in a zero-shot setting, the largest model reaches 0.871 AUC, and density-related results report strong BI-RADS-adjacent performance.\"}]",1784180890,28,{"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":86,"head_meta":88,"extra_data":90,"updated_unix":27},"aegis-multi-task-joint-embedding-predictive-architecture-for-mammography","",{"@graph":35,"@context":85},[36,53,68],{"@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/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/aegis-multi-task-joint-embedding-predictive-architecture-for-mammography/82490/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What is AEGIS and what tasks does it address in mammography?","Question",{"text":75,"@type":76},"AEGIS is a joint-embedding predictive architecture for breast cancer detection and mammographic density assessment. It is designed to support triage, detection-related objectives, and density classification using Vision Transformers.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How is the model trained before fine-tuning?",{"text":80,"@type":76},"The approach uses self-supervised JEPA pre-training with three Vision Transformer variants trained on 71,103 studies from 14 clinical sites. It is then followed by supervised fine-tuning with progressive resolution scaling up to 2048×1536.",{"name":82,"@type":73,"acceptedAnswer":83},"What performance does the largest model achieve on the test set and in external validation?",{"text":84,"@type":76},"On the curated 785-study test set, the largest model achieves AUC 0.949 for breast cancer triage (93% sensitivity, 75% specificity). For external validation on VinDr-Mammo in a zero-shot setting, the largest model reaches 0.871 AUC, and density-related results report strong BI-RADS-adjacent performance.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]