FedCLAM: Client Adaptive Momentum with Foreground Intensity Matching for Federated Medical Image Segmentation

2025-06-27

AI Summary

Federated learning for medical image segmentation suffers from feature distribution shifts caused by inter-institutional device and population heterogeneity, while existing aggregation methods lack sufficient generalization. To address this, we propose FedCLAM—a novel federated framework that jointly optimizes model convergence and domain robustness. Its key contributions are: (1) a client-adaptive momentum mechanism that dynamically modulates local gradient updates and aggregation weights via personalized suppression factors; and (2) a foreground intensity matching loss that explicitly aligns foreground region intensity distributions across multi-center images. Evaluated on two public medical imaging benchmarks, FedCLAM outperforms eight state-of-the-art federated segmentation methods, achieving significant improvements in Dice score and robustness to domain shifts. Ablation studies confirm the efficacy of each component, demonstrating superior generalization across heterogeneous clients and resilience to data non-IIDness.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Computer Vision: Segmentation
• Multiagent Systems: Adversarial Agents

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Federated learning is a decentralized training approach that keeps data under stakeholder control while achieving superior performance over isolated training. While inter-institutional feature discrepancies pose a challenge in all federated settings, medical imaging is particularly affected due to diverse imaging devices and population variances, which can diminish the global model's effectiveness. Existing aggregation methods generally fail to adapt across varied circumstances. To address this, we propose FedCLAM, which integrates extit{client-adaptive momentum} terms derived from each client's loss reduction during local training, as well as a extit{personalized dampening factor} to curb overfitting. We further introduce a novel extit{intensity alignment} loss that matches predicted and ground-truth foreground distributions to handle heterogeneous image intensity profiles across institutions and devices. Extensive evaluations on two datasets show that FedCLAM surpasses eight cutting-edge methods in medical segmentation tasks, underscoring its efficacy. The code is available at https://github.com/siomvas/FedCLAM.

Problem

• Addresses feature discrepancies in federated medical image segmentation
• Improves global model adaptability across diverse imaging devices
• Handles heterogeneous image intensity profiles across institutions

Innovation

• Client-adaptive momentum for local loss reduction
• Personalized dampening factor to prevent overfitting
• Intensity alignment loss for heterogeneous image profiles