Decouple, Purify and Unite: Semantic-Structural Prototype Learning for Federated Medical Segmentation

2026-10-03

AI Summary

This study addresses the challenges of feature heterogeneity, missing contextual representations, and aggregation bias arising from device discrepancies in federated medical image segmentation by proposing the FedBCS+ framework. This method introduces a novel decoupled context-prototype alignment mechanism that purifies prototypes via frequency-domain style recalibration, decouples semantic and structural features for independent alignment, and incorporates distribution-aware adaptive weighted aggregation to effectively mitigate consensus bias, all supported by rigorous convergence guarantees. Evaluated across five heterogeneous medical segmentation benchmarks, the proposed framework achieves state-of-the-art average Dice scores, substantially enhancing model robustness and segmentation accuracy in multi-center scenarios.

Abstract

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.

Problem

• Federated Learning
• Medical Image Segmentation
• Feature Heterogeneity
• Prototype Learning
• Domain Shift

Innovation

• Federated Learning
• Medical Image Segmentation
• Prototype Learning
• Style Purification
• Contextual Alignment