[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86248-en":3,"doc-seo-86248-105":30,"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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},86248,687197207919,"Theodora","https://ap-avatar.wpscdn.com/avatar/a000253d6f5f7c60be?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779446848396160552",8,"Research & Report","CFR-Net Collaborative Feature Refinement Network for Medical Image Anomaly Detection","Medical image anomaly detection remains challenging because networks pretrained on natural images adapt poorly to medical data, where abnormalities often appear as fine-grained local shifts, multi-scale contextual mismatches, and orientation-sensitive structural deviations. CFR-Net introduces Collaborative Feature Refinement Network to refine shared teacher–student features before decoding and enforce cross-space consistency after decoding. It uses a Multi-Path Feature Refinement Module to reduce domain discrepancy while modeling local, multi-scale, and orientation-sensitive characteristics, optimized by a variance-sensitive objective and dynamic data reorganization. Experiments on medical benchmarks show competitive classification and strong anomaly localization under normal-only training.","CFR-Net: Collaborative Feature Refininement Network for Medical Image Anomaly Detection  \n1 School of Mechanical Engineering Shandong University  \nJinan, China  \n2 Key Laboratory of High Efficiency and Clean Mechanical Manufacture Shandong University, Ministry of Education  \nJinan, China  \n3 Monash University Clayton, VIC 3800, Australia  \n4 Airdoc-Monash Research, Monash University  \nClayton, VIC 3800, Australia  \n5 Shandong Key Laboratory of Ubiquitous Intelligent Computing University of Jinan  \nJinan, China  \n6 Institute of Medical Technology and Cancer Hospital, Peking University Institute of Advanced Clinical Medicine and Biomedical Engineering Department,  \nPeking University  \nPeking University International Cancer Institute  \nBeijing, China  \nAbstract  \nMedical image anomaly detection remains challenging because networks pretrained on natural images often exhibit limited adaptability to medical images, where abnormal patterns appear as fine-grained local shifts, multi-scale contextual mismatches, and orientation-sensitive structural deviations. To address this, we propose the Collaborative Feature Refinement Network (CFR-Net), which combines shared teacher-student feature refinement before decoding with cross-space consistency after decoding. CFR-Net refines frozen teacher features and trainable student features using a Multi-Path Feature Refinement Module (MPFRM) with shared parameters, imposing common multi-path refinement rules on generic visual references and representations adapted to the medical domain, thereby mitigating domain discrepancy while modeling local, multi-scale, and orientation-sensitive feature characteristics. A variance-sensitive objective and dynamic  \narXiv :2607 .  \n“homework set” reorganization further support layer-adaptive consistency learning. Experiments on medical benchmarks show that CFR-Net achieves competitive anomaly classification and strong anomaly localization performance when trained on normal data.  \n1 Introduction  \nIn medical imaging, precise anomaly detection plays a pivotal role in enabling early diagnosis and timely treatment of diseases [1, 14] . Medical images provide critical information about internal body structures, helping physicians identify abnormalities such as tumors, microcalcifications, and vascular distortions [10, 29] . However, the prohibitive cost of medical image acquisition and scarcity of annotated abnormal data pose significant challenges to traditional supervised learning approaches [41] .  \nUnsupervised anomaly detection has emerged as a promising alternative, relying solely on normal images for training [23, 47] . Recent advances have explored different strategies to improve medical anomaly detection, including segmentation guided by normal images, multimodal priors, and uncertainty-aware anomaly modeling [28, 33, 51] . Nevertheless, the indistinct boundary between normal and abnormal patterns in medical images, combined with their multi-scale characteristics and complex backgrounds, still limits the effectiveness of existing methods in challenging medical imaging scenarios [22] .  \nKnowledge distillation shows substantial promise for medical image processing [15], with recent successes in industrial anomaly detection [24, 46] inspiring medical applications. However, existing distillation approaches face significant limitations when applied to medical images. Commonly used pretrained encoders are learned from natural-image distributions [49], and their feature priors are therefore biased toward natural-scene semantics and texture statistics. Although such priors can still provide stable visual primitives at low and intermediate levels, including edges, textures, and local structures, they are not directly optimized for normal anatomical patterns in medical images. Consequently, existing methods may struggle to process fine local variances, multi-scale context mismatches [44], and orientation-sensitive structural changes inherent in medical data [19, 30","cbCaipfHlG8wKK20","https://ap.wps.com/l/cbCaipfHlG8wKK20","pdf",2254460,2,1,19,"English","en",105,"# Introduction\n## Challenges in Medical Anomaly Detection\n## Unsupervised Learning and Distillation Limits\n## CFR-Net Framework Overview\n## Multi-Path Feature Refinement Module (MPFRM)\n## Variance-Sensitive Objective and Dynamic Learning Strategy\n# Experimental Validation and Contributions","[{\"question\":\"Why do pretrained natural-image networks struggle with medical anomaly detection?\",\"answer\":\"Their feature priors are biased toward natural-scene semantics and texture statistics, so they are not directly optimized for normal anatomical patterns. This leads to difficulties with fine local variations, multi-scale context mismatches, and orientation-sensitive structural changes.\"},{\"question\":\"How does CFR-Net refine features differently before and after decoding?\",\"answer\":\"CFR-Net performs shared feature refinement before decoding using a multi-path refinement module. After decoding, it applies teacher–student cross-space consistency to constrain decoded features using complementary encoder feature spaces and promote coherent reconstruction of normal patterns.\"},{\"question\":\"What training setup does CFR-Net support and what results does it achieve?\",\"answer\":\"CFR-Net is trained on normal data only. Experiments on medical benchmarks demonstrate competitive anomaly classification performance and strong anomaly localization performance across diverse modalities.\"}]",1784209799,48,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"cfr-net-collaborative-feature-refinement-network-for-medical-image-anomaly-detection","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/cfr-net-collaborative-feature-refinement-network-for-medical-image-anomaly-detection/86248/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-25","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},"Why do pretrained natural-image networks struggle with medical anomaly detection?","Question",{"text":75,"@type":76},"Their feature priors are biased toward natural-scene semantics and texture statistics, so they are not directly optimized for normal anatomical patterns. This leads to difficulties with fine local variations, multi-scale context mismatches, and orientation-sensitive structural changes.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does CFR-Net refine features differently before and after decoding?",{"text":80,"@type":76},"CFR-Net performs shared feature refinement before decoding using a multi-path refinement module. After decoding, it applies teacher–student cross-space consistency to constrain decoded features using complementary encoder feature spaces and promote coherent reconstruction of normal patterns.",{"name":82,"@type":73,"acceptedAnswer":83},"What training setup does CFR-Net support and what results does it achieve?",{"text":84,"@type":76},"CFR-Net is trained on normal data only. Experiments on medical benchmarks demonstrate competitive anomaly classification performance and strong anomaly localization performance across diverse modalities.","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":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":22,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":106,"slug":137},"General","general"]