[{"data":1,"prerenderedAt":396},["ShallowReactive",2],{"publication-2026\u002Funiultra-interactive-parameter-efficient-sam2-for-universal-ultrasound-segmentat-en":3,"publication-members":83},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"_hidden":6,"Segmentation\" authors":10,"authors_orcid":22,"year":33,"doi":34,"openalex_id":35,"venue":36,"abstract_screenshot":37,"keywords":38,"body":46,"_type":76,"_id":77,"_source":78,"_file":79,"_stem":80,"_extension":81,"locale":82},"\u002Fpublications\u002F2026\u002Funiultra-interactive-parameter-efficient-sam2-for-universal-ultrasound-segmentat","2026",false,"","UniUltra: Interactive Parameter-Efficient SAM2 for Universal Ultrasoun","The Segment Anything Model 2 (SAM2) demonstrates remarkable universal segmentation capabilities on natural images. However, its performance on ultrasound images is significantly degraded due to domain disparities. This limitation raises two critical challenges: how to efficiently adapt SAM2 to ultrasound imaging while maintaining parameter efficiency, and how to deploy the adapted model effectively in resource-constrained clinical environments. To address these issues, we propose UniUltra for universal ultrasound segmentation. Specifically, we first introduce a novel context-edge hybrid adapter (CH-Adapter) that enhances fine-grained perception across diverse ultrasound imaging modalities while achieving parameter-efficient fine-tuning. To further improve clinical applicability, we develop a deep-supervised knowledge distillation (DSKD) technique that transfers knowledge from the large image encoder of the fine-tuned SAM2 to a super lightweight encoder, substantially reducing computational requirements without compromising performance. Extensive experiments demonstrate that UniUltra outperforms state-of-the-arts with superior generalization capabilities. Notably, our framework achieves competitive performance using only 8.91% of SAM2's parameters during fine-tuning, and the final compressed model reduces the parameter count by 94.08% compared to the original SAM2, making it highly suitable for practical clinical deployment. The source code is available at https:\u002F\u002Fgithub.com\u002Fxq141839\u002FUniUltra.",[11,12,13,14,15,16,17,18,19,20,21],"Li, Yue","Xu, Qingsong","Zhang, Yixuan","He, Xueqiu","Zhang, Qian","Yao, Yuan","Tesem, Fiseha B.","Chen, Xin","Wang, Ruili","Chen, Zhen","Chen, Chang Wen",[23,24,25,26,27,28,29,30,31,32,29],"0009-0006-5923-6044","0000-0002-1700-322X","0000-0002-1924-9848","0000-0003-2760-582X","0000-0001-6074-2354","0000-0002-2705-6245",null,"0000-0002-3685-0854","0000-0001-7117-2772","0000-0002-3263-3043",2026,"https:\u002F\u002Fdoi.org\u002F10.1109\u002Ftmm.2026.3687007","W4416551711","IEEE Transactions on Multimedia","uniultra-interactive-parameter-efficient-sam2-for-universal-ultrasound-segmentat.png",[39,40,41,42,43,44,45],"Segmentation","Encoder","Ultrasound","Generalization","Pattern recognition (psychology)","Ultrasound imaging","Domain (mathematical analysis)",{"type":47,"children":48,"toc":73},"root",[49],{"type":50,"tag":51,"props":52,"children":53},"element","p",{},[54,57,71],{"type":55,"value":56},"text","The Segment Anything Model 2 (SAM2) demonstrates remarkable universal segmentation capabilities on natural images. However, its performance on ultrasound images is significantly degraded due to domain disparities. This limitation raises two critical challenges: how to efficiently adapt SAM2 to ultrasound imaging while maintaining parameter efficiency, and how to deploy the adapted model effectively in resource-constrained clinical environments. To address these issues, we propose UniUltra for universal ultrasound segmentation. Specifically, we first introduce a novel context-edge hybrid adapter (CH-Adapter) that enhances fine-grained perception across diverse ultrasound imaging modalities while achieving parameter-efficient fine-tuning. To further improve clinical applicability, we develop a deep-supervised knowledge distillation (DSKD) technique that transfers knowledge from the large image encoder of the fine-tuned SAM2 to a super lightweight encoder, substantially reducing computational requirements without compromising performance. Extensive experiments demonstrate that UniUltra outperforms state-of-the-arts with superior generalization capabilities. Notably, our framework achieves competitive performance using only 8.91% of SAM2's parameters during fine-tuning, and the final compressed model reduces the parameter count by 94.08% compared to the original SAM2, making it highly suitable for practical clinical deployment. The source code is available at ",{"type":50,"tag":58,"props":59,"children":62},"uri",{"xmlns:mml":60,"xmlnsXLink":61},"http:\u002F\u002Fwww.w3.org\u002F1998\u002FMath\u002FMathML","http:\u002F\u002Fwww.w3.org\u002F1999\u002Fxlink",[63],{"type":50,"tag":64,"props":65,"children":69},"a",{"href":66,"rel":67},"https:\u002F\u002Fgithub.com\u002Fxq141839\u002FUniUltra",[68],"nofollow",[70],{"type":55,"value":66},{"type":55,"value":72},".",{"title":7,"searchDepth":74,"depth":74,"links":75},2,[],"markdown","content:publications:2026:uniultra-interactive-parameter-efficient-sam2-for-universal-ultrasound-segmentat.md","content","publications\u002F2026\u002Funiultra-interactive-parameter-efficient-sam2-for-universal-ultrasound-segmentat.md","publications\u002F2026\u002Funiultra-interactive-parameter-efficient-sam2-for-universal-ultrasound-segmentat","md","en",[84,97,102,113,120,128,134,143,150,156,161,171,178,187,193,205,214,223,229,237,242,250,256,264,268,278,285,293,298,307,313,321,327,335,340,346,356,364,370,378,383,391],{"_path":85,"title":86,"name":87,"role":88,"email":29,"image":89,"category":90,"interests":91,"order":74,"_id":96},"\u002Fmembers\u002Fstaff\u002Falain-chong","Vice President for Global Affairs and Partnerships · Professor of Information Systems and Digital Innovation","Alain Chong","Deputy Director of Lab","assets\u002F8.png","staff",[92,93,94,95],"信息系统与运作管理","计算机科学与运筹学","Information Systems and Operations Management","Computer Science and Operations Research","content:members:staff:alain-chong.md",{"_path":85,"title":98,"role":99,"interests":100,"_id":101},"全球事务与合作副校长 · 信息系统与数字创新教授","实验室副主任",[92,93],"content:members:staff:alain-chong.zh-CN.md",{"_path":103,"title":104,"name":105,"role":106,"email":29,"image":107,"category":90,"interests":108,"order":111,"_id":112},"\u002Fmembers\u002Fstaff\u002Fanthony-belloti","Professor","Anthony Belloti","Core Member","assets\u002F41.png",[109,110],"Machine Learning and Credit Risk Model","Model Risks",9,"content:members:staff:anthony-belloti.md",{"_path":103,"title":114,"role":115,"interests":116,"_id":119},"计算机科学系教授","核心成员",[117,118],"机器学习与信用风险模型","模型风险","content:members:staff:anthony-belloti.zh-CN.md",{"_path":121,"title":122,"name":122,"role":106,"email":29,"image":123,"category":90,"interests":124,"order":126,"_id":127},"\u002Fmembers\u002Fstaff\u002Fboon-giin-lee","Boon Giin Lee","assets\u002F31.jpg",[125],"Intelligent Sensor and Extended Reality",11,"content:members:staff:boon-giin-lee.md",{"_path":121,"title":129,"role":115,"interests":130,"_id":133},"人机交互实验室负责人 · 计算机科学系副教授",[131,132],"人机交互 HCI","智能传感与扩展现实技术","content:members:staff:boon-giin-lee.zh-CN.md",{"_path":135,"title":136,"name":136,"role":137,"email":29,"image":138,"category":90,"interests":139,"order":141,"_id":142},"\u002Fmembers\u002Fstaff\u002Fcong-cao","Cong Cao","Direction Leader","assets\u002FCC.png",[140],"Science and technology policy and institutional reform",7,"content:members:staff:cong-cao.md",{"_path":135,"title":144,"name":145,"role":146,"interests":147,"_id":149},"宁波诺丁汉大学商学院创新学教授","曹聪","方向带头人",[148],"科技政策与体制改革","content:members:staff:cong-cao.zh-CN.md",{"_path":151,"title":152,"name":152,"role":106,"email":29,"image":153,"category":90,"order":154,"_id":155},"\u002Fmembers\u002Fstaff\u002Fdave-towey","Dave Towey","assets\u002F32.jpg",8,"content:members:staff:dave-towey.md",{"_path":151,"title":157,"role":115,"interests":158,"_id":160},"计算机科学系教授 · 计算机科学系主任",[159],"计算机科学与语言学","content:members:staff:dave-towey.zh-CN.md",{"_path":162,"title":163,"name":163,"role":106,"email":29,"image":164,"category":90,"interests":165,"order":169,"_id":170},"\u002Fmembers\u002Fstaff\u002Ffazl-ullah-khan","Fazl Ullah Khan","assets\u002F44.png",[166,167,168],"Computer Network","Computer Architecture and Network Security","Software Engineering",12,"content:members:staff:fazl-ullah-khan.md",{"_path":162,"title":172,"role":115,"interests":173,"_id":177},"计算机科学系助理教授 · IEEE 高级会员",[174,175,176],"计算机网络","计算机和网络安全","软件工程","content:members:staff:fazl-ullah-khan.zh-CN.md",{"_path":179,"title":180,"name":181,"role":106,"email":29,"image":182,"category":90,"interests":183,"order":185,"_id":186},"\u002Fmembers\u002Fstaff\u002Fheng-yu","Associate Professor","Heng Yu","assets\u002FHENGYU.png",[184],"Embedded Systems Design",17,"content:members:staff:heng-yu.md",{"_path":179,"title":188,"name":189,"role":115,"interests":190,"_id":192},"计算机科学系副教授","于恒",[191],"嵌入式系统设计","content:members:staff:heng-yu.zh-CN.md",{"_path":194,"title":180,"name":195,"role":106,"email":29,"image":196,"category":90,"interests":197,"order":203,"_id":204},"\u002Fmembers\u002Fstaff\u002Fheshan-du","Heshan Du","assets\u002Fhesahndu.png",[198,199,200,201,202],"Logic, Knowledge Representation and Reasoning","Geographic Information Systems","Operations Research","Machine Learning","Reinforcement Learning",20,"content:members:staff:heshan-du.md",{"_path":194,"title":188,"name":206,"role":115,"interests":207,"_id":213},"杜何珊",[208,209,210,211,212],"逻辑与知识表示","地理信息系统","运筹学","机器学习","强化学习","content:members:staff:heshan-du.zh-CN.md",{"_path":215,"title":216,"name":217,"role":106,"email":29,"image":218,"category":90,"interests":219,"order":221,"_id":222},"\u002Fmembers\u002Fstaff\u002Fhuan-jin","Assistant Professor","Huan Jin","assets\u002Fhuanjin.png",[220,201],"Optimisation",21,"content:members:staff:huan-jin.md",{"_path":215,"title":224,"name":225,"role":115,"interests":226,"_id":228},"计算机科学系助理教授","靳欢",[227,211],"优化","content:members:staff:huan-jin.zh-CN.md",{"_path":230,"title":180,"name":231,"role":137,"email":29,"image":232,"category":90,"interests":233,"order":235,"_id":236},"\u002Fmembers\u002Fstaff\u002Fjianfeng-ren","Jianfeng Ren","assets\u002F42.jpg",[201,234],"Computer Vision",3,"content:members:staff:jianfeng-ren.md",{"_path":230,"title":188,"name":238,"role":146,"interests":239,"_id":241},"任剑锋",[211,240],"计算机视觉","content:members:staff:jianfeng-ren.zh-CN.md",{"_path":243,"title":244,"name":244,"role":137,"email":29,"image":245,"category":90,"interests":246,"order":248,"_id":249},"\u002Fmembers\u002Fstaff\u002Fjiawei-li","Jiawei Li","assets\u002F11.png",[247],"Computer Science and Artificial Intelligence",15,"content:members:staff:jiawei-li.md",{"_path":243,"title":251,"name":252,"role":146,"interests":253,"_id":255},"计算机科学系助理教授 · 英国诺丁汉大学博士后","李家炜",[254],"计算机与人工智能","content:members:staff:jiawei-li.zh-CN.md",{"_path":257,"title":180,"name":258,"role":106,"email":29,"image":259,"category":90,"interests":260,"order":262,"_id":263},"\u002Fmembers\u002Fstaff\u002Fmatthew-pike","Matthew Pike","assets\u002F43.jpg",[261],"Digitalised Learning",16,"content:members:staff:matthew-pike.md",{"_path":257,"title":188,"role":115,"interests":265,"_id":267},[266],"数字化学习","content:members:staff:matthew-pike.zh-CN.md",{"_path":269,"title":216,"name":270,"role":106,"email":29,"image":271,"category":90,"interests":272,"order":276,"_id":277},"\u002Fmembers\u002Fstaff\u002Fning-xue","Ning Xue","\u002Fimages\u002Fuon-logo.png",[273,274,275],"Artificial Intelligence","Computational Intelligence","Combinatorial Optimization",13,"content:members:staff:ning-xue.md",{"_path":269,"title":224,"name":279,"role":115,"interests":280,"_id":284},"薛宁",[281,282,283],"人工智能","计算智能","组合优化","content:members:staff:ning-xue.zh-CN.md",{"_path":286,"title":216,"name":287,"role":106,"email":29,"image":288,"category":90,"interests":289,"order":291,"_id":292},"\u002Fmembers\u002Fstaff\u002Fqian-zhang","Qian Zhang","assets\u002Fqz.png",[290,234,201],"Image Processing",14,"content:members:staff:qian-zhang.md",{"_path":286,"title":224,"name":294,"role":115,"interests":295,"_id":297},"张茜",[296,240,211],"图像处理","content:members:staff:qian-zhang.zh-CN.md",{"_path":299,"title":104,"name":300,"role":301,"email":29,"image":302,"category":90,"interests":303,"orcid":304,"order":305,"_id":306},"\u002Fmembers\u002Fstaff\u002Fruibin-bai","Ruibin Bai","Director of Lab","assets\u002F38.png",[95],"0000-0003-1722-568X",1,"content:members:staff:ruibin-bai.md",{"_path":299,"title":308,"name":309,"role":310,"interests":311,"_id":312},"教授","白瑞斌","实验室主任",[93],"content:members:staff:ruibin-bai.zh-CN.md",{"_path":314,"title":315,"name":315,"role":137,"email":29,"image":316,"category":90,"interests":317,"order":319,"_id":320},"\u002Fmembers\u002Fstaff\u002Fsean-he","Sean He","assets\u002F39.png",[234,318,201],"Data Analytics",5,"content:members:staff:sean-he.md",{"_path":314,"title":322,"name":323,"role":146,"interests":324,"_id":326},"计算机科学系教授 · 国家级讲席学者","何祥健",[240,325,211],"数据分析","content:members:staff:sean-he.zh-CN.md",{"_path":328,"title":329,"name":329,"role":106,"email":29,"image":330,"category":90,"interests":331,"order":333,"_id":334},"\u002Fmembers\u002Fstaff\u002Ftianxiang-cui","Tianxiang Cui","assets\u002Ftianxiangcui.png",[274,332,201,202],"Operation Research",19,"content:members:staff:tianxiang-cui.md",{"_path":328,"title":172,"name":336,"role":115,"interests":337,"_id":339},"崔天翔",[282,338,211,212],"运筹研究","content:members:staff:tianxiang-cui.zh-CN.md",{"_path":341,"title":104,"name":342,"role":106,"email":29,"image":343,"category":90,"order":344,"_id":345},"\u002Fmembers\u002Fstaff\u002Fxiuping-hua","Xiuping Hua","assets\u002FxiupignHua.png",10,"content:members:staff:xiuping-hua.md",{"_path":341,"title":347,"name":348,"role":115,"interests":349,"_id":355},"金融、会计与经济系教授","华秀萍",[350,351,352,353,354,117],"资产定价","公司金融","衍生品","金融科技","创新金融和普惠金融","content:members:staff:xiuping-hua.zh-CN.md",{"_path":357,"title":104,"name":358,"role":106,"email":29,"image":359,"category":90,"interests":360,"order":362,"_id":363},"\u002Fmembers\u002Fstaff\u002Fying-weng","Ying Weng","assets\u002Fyingweng.png",[234,290,361],"IoT",4,"content:members:staff:ying-weng.md",{"_path":357,"title":114,"name":365,"role":115,"interests":366,"_id":369},"翁莹",[240,296,367,368],"物联网 IoT","无线网络安全与服务质量","content:members:staff:ying-weng.zh-CN.md",{"_path":371,"title":216,"name":372,"role":106,"email":29,"image":373,"category":90,"interests":374,"order":376,"_id":377},"\u002Fmembers\u002Fstaff\u002Fyuan-yao","Yuan Yao","assets\u002Fyuanyao.png",[375],"Autonomous Agents and Multi-Agent Systems",18,"content:members:staff:yuan-yao.md",{"_path":371,"title":224,"name":379,"role":115,"interests":380,"_id":382},"姚远",[381],"自主智能体与多智能体系统","content:members:staff:yuan-yao.zh-CN.md",{"_path":384,"title":216,"name":385,"role":137,"email":29,"image":386,"category":90,"interests":387,"order":389,"_id":390},"\u002Fmembers\u002Fstaff\u002Fzheng-lu","Zheng Lu","assets\u002F13.png",[388],"Computer Science",6,"content:members:staff:zheng-lu.md",{"_path":384,"title":224,"name":392,"role":146,"interests":393,"_id":395},"卢正",[394],"计算机科学","content:members:staff:zheng-lu.zh-CN.md",1782639441098]