[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82073-en":3,"doc-seo-82073-105":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},82073,13056703019404,"Miles","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"Research & Report","Secure-by-Disguise: A Systematic Evaluation of Image Disguising for Confidential Medical Image Modeling","Cloud-based deep learning enables large-scale medical image analysis but increases privacy risks when sensitive patient images are outsourced for model development. Image disguising is a privacy-enhancing technology that converts images into visually unintelligible forms while keeping learnable information. A unified evaluation framework compares representative methods, measuring predictive utility, computational efficiency, parameter sensitivity, and robustness to reconstruction attacks across multiple medical datasets. Results show utility can be preserved for classification while degrading substantially for dense segmentation.","arXiv :2607 .08867v 1 [ cs .CV] 9 Jul 2026  \nSecure-by-Disguise: A Systematic Evaluation of Image Disguising for Confidential Medical Image Modeling  \nJason Rojasa , Jiajie Hea , Yash Patelb , Yuechun Gua , Zeyun Yuc , Keke  \nChena,1  \na University of Maryland, Baltimore County, 1000 Hilltop  \nCircle, Baltimore, 21250, MD, USA  \nb Lawrence Technological University, 21000 West Ten Mile  \nRoad, Southfield, 48075, MI, USA  \nc University of Wisconsin, Milwaukee, 3203 N. Downer  \nAvenue, Milwaukee, 53211, WI, USA  \nAbstract  \nObjective: Cloud-based deep learning enables large-scale medical image analysis but raises significant privacy concerns when sensitive patient images are outsourced for model development. Image disguising has recently emerged as a promising privacy-enhancing technology (PET) that transforms images into visually unintelligible representations while preserving information for downstream learning. However, its suitability for clinically relevant medical image analysis remains largely unknown.  \nMethods: We established a unified evaluation framework to systematically assess representative image disguising methods for confidential medical image analysis. Two representative frameworks, DisguisedNets and NeuraCrypt, were evaluated on four public medical imaging datasets spanning both image classification and semantic segmentation. Predictive utility, computational efficiency, parameter sensitivity, and robustness against representative reconstruction attacks were evaluated under a common experimental protocol.  \nResults: Image disguising exhibited fundamentally different behavior on image-level and pixel-level medical AI tasks. While representative methods  \n∗ Corresponding Author  \nEmail addresses: [jasonr2@umbc.edu](jasonr2@umbc.edu) (Jason Rojas), [jiajieh1@umbc.edu](jiajieh1@umbc.edu) (Jiajie  \nHe), [ypatel3@ltu.edu](ypatel3@ltu.edu) (Yash Patel), [ygu2@umbc.edu](ygu2@umbc.edu) (Yuechun Gu), [yuz@uwm.edu](yuz@uwm.edu)  \n(Zeyun Yu), [kekechen@umbc.edu](kekechen@umbc.edu) (Keke Chen)  \npreserved practical utility for medical image classification, they incurred substantially greater performance degradation for dense semantic segmentation. Among the evaluated approaches, Randomized Multidimensional Transformation (RMT) consistently achieved the best balance between predictive performance, computational efficiency, and resistance to reconstruction attacks, whereas AES-based disguising resulted in severe utility degradation. Regression-based reconstruction attacks that were previously shown to be partially effective on natural-image benchmarks became considerably less effective on realistic medical images.  \nConclusion: Image disguising is a promising PET for confidential medical AI, particularly for cloud-based medical image classification. However, current approaches inadequately preserve the fine-grained spatial information required for dense medical image segmentation. These findings provide the first systematic evidence regarding the capabilities and limitations of image disguising for medical AI and offer practical guidance for the development of next-generation privacy-enhancing technologies for medical image analysis.  \nKeywords:  \nMedical image analysis, Privacy-preserving machine learning, Image disguising, Cloud computing, Medical AI, Deep learning  \n1. Introduction  \nDeep learning (DL) has become a core technology in medical imaging [1, 2, 3, 4], achieving remarkable performance in a broad range of clinical applications [5, 6], including disease screening, lesion detection, tumor segmentation, and longitudinal disease monitoring. The rapid growth of medical imaging datasets and increasingly large deep learning models has simultaneously driven healthcare institutions toward cloud-based computing infrastructures that provide scalable storage and high-performance computational resources. While cloud platforms facilitate the development and deployment of medical AI systems, they also create an urgent need [7] f","cbCairHopzaKciCn","https://ap.wps.com/l/cbCairHopzaKciCn","pdf",3500394,1,36,"English","en",105,"# Abstract\n# Introduction\n# Methods\n# Results\n# Conclusion","[{\"question\":\"What problem does image disguising address in cloud-based medical AI development?\",\"answer\":\"Image disguising targets the privacy concern that sensitive patient images may be exposed when outsourced for training and model development. It transforms images into visually unintelligible representations while preserving information for downstream learning.\"},{\"question\":\"How was image disguising evaluated in the document?\",\"answer\":\"The work uses a unified evaluation framework to assess representative methods, including DisguisedNets and NeuraCrypt. It measures predictive utility, computational efficiency, parameter sensitivity, and robustness against reconstruction attacks using a common protocol across multiple public medical imaging datasets.\"},{\"question\":\"What performance differences are observed between image-level and pixel-level tasks?\",\"answer\":\"The document reports fundamentally different behavior: practical utility is preserved for medical image classification, but substantially greater degradation occurs for dense semantic segmentation. RMT provides the best overall balance, while AES-based disguising causes severe utility degradation.\"}]",1784178063,91,{"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":86,"head_meta":88,"extra_data":90,"updated_unix":27},"secure-by-disguise-a-systematic-evaluation-of-image-disguising-for-confidential-medical-image-modeling","",{"@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/secure-by-disguise-a-systematic-evaluation-of-image-disguising-for-confidential-medical-image-modeling/82073/",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 problem does image disguising address in cloud-based medical AI development?","Question",{"text":75,"@type":76},"Image disguising targets the privacy concern that sensitive patient images may be exposed when outsourced for training and model development. It transforms images into visually unintelligible representations while preserving information for downstream learning.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How was image disguising evaluated in the document?",{"text":80,"@type":76},"The work uses a unified evaluation framework to assess representative methods, including DisguisedNets and NeuraCrypt. It measures predictive utility, computational efficiency, parameter sensitivity, and robustness against reconstruction attacks using a common protocol across multiple public medical imaging datasets.",{"name":82,"@type":73,"acceptedAnswer":83},"What performance differences are observed between image-level and pixel-level tasks?",{"text":84,"@type":76},"The document reports fundamentally different behavior: practical utility is preserved for medical image classification, but substantially greater degradation occurs for dense semantic segmentation. RMT provides the best overall balance, while AES-based disguising causes severe utility degradation.","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"]