发表机构
Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College; Beijing University of Posts and Telecommunications; Nanyang Technological University; Ant Group; Shanghai Innovation Institute; Institute of Automation, Chinese Academy of Sciences; DAMO Academy, Alibaba Group; Fourth Military Medical University; Macao Polytechnic University; Wuhan University; The Hong Kong Polytechnic University（北京协和医院，中国医学科学院北京协和医学院; 北京邮电大学; 南洋理工大学; 蚂蚁集团; 上海创新研究院; 中国科学院自动化研究所; 阿里巴巴达摩院; 中国人民解放军空军军医大学; 澳门理工大学; 武汉大学; 香港理工大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对联邦医学分割中的特征异质性，提出FedBCS+，通过频域风格重校准解耦内容与风格，并分别对齐语义与结构原型，自适应聚合减少偏差，在五个基准上取得最优平均Dice。
AI 中文摘要
联邦学习使医疗机构能够在无需共享数据的情况下训练全局模型，然而来自不同扫描仪或协议的特征异质性仍然具有挑战性。现有的基于表示的方法面临两个局限：1）不完整的上下文表示学习：单层或耦合的表示忽视了多层级的结构线索，并将区域语义与边界细节纠缠在一起。2）逐层风格与聚合偏差：中间层跨域的特定风格差异会降低原型质量，而忽略客户端分布偏移的聚合可能进一步放大偏差。我们提出FedBCS+，一种联邦解耦上下文对齐与风格净化聚合方法。我们在原型构建中采用频域风格重校准（FSR）来解耦内容-风格表示并提取风格净化的原型。基于这些净化特征，解耦上下文原型对齐（DCPA）明确地将多层级特征解耦为语义原型和结构原型，并分别对齐区域语义和细粒度解剖结构。风格净化语义原型聚合（S2PA）衡量每个客户端的净化原型与全局共识的差异，并自适应地重新加权聚合，偏向代表性不足的客户端以减少共识偏差。在涵盖组织病理学、MRI、超声和结肠镜检查的五个异构医学分割基准上，FedBCS+在比较方法中取得了最高的平均Dice。收敛性分析进一步刻画了聚合和对齐如何影响优化界。
英文摘要
Federated learning enables medical institutions to train a global model without sharing data, yet feature heterogeneity from diverse scanners or protocols remains challenging. Existing representation-based methods face two limitations: 1) Incomplete Contextual Representation Learning: single-layer or coupled representations overlook multi-level structural cues and entangle regional semantics with boundary details. 2) Layerwise Style and Aggregation Biases: domain-specific style discrepancies across intermediate layers degrade prototypes, while aggregation that overlooks client distribution shifts can further amplify bias. We propose FedBCS+, federated decoupled contextual alignment with style-purified aggregation. We employ Frequency-domain Style Recalibration (FSR) in prototype construction to decouple content-style representations and extract style-purified prototypes. Built upon these purified features, Decoupled Contextual Prototype Alignment (DCPA) explicitly decouples multi-level features into semantic and structural prototypes and aligns regional semantics and fine-grained anatomical structures separately. Style-purified Semantic Prototype Aggregation (S2PA) measures each client's purified prototype divergence from the global consensus and adaptively reweights aggregation toward under-represented clients to reduce consensus bias. On five heterogeneous medical segmentation benchmarks spanning histopathology, MRI, ultrasound, and colonoscopy, FedBCS+ achieves the highest mean Dice among the compared methods. A convergence analysis further characterizes how aggregation and alignment affect the optimization bound.
Comments17 pages, 9 figures, 7 tables
