FedDAG: Federated Domain Adversarial Generation Towards Generalizable Medical Image Analysis

TL;DR

FedDAG addresses unseen-domain generalization in federated medical imaging by locally generating novel-domain, semantically-preserving images through instance-level adversarial learning. It combines dual-level semantic constraints, a domain-invariant representation learner, and a sharpness-aware hierarchical aggregation to balance contributions across clients and rounds. The method yields consistent gains over strong baselines across four medical benchmarks, demonstrating robust generalization to unseen domains while respecting privacy constraints. This framework provides a practical, scalable approach to improving federated medical diagnosis under domain shifts, with clear mechanisms to mitigate data isolation and heterogeneity effects.

Abstract

Federated domain generalization aims to train a global model from multiple source domains and ensure its generalization ability to unseen target domains. Due to the target domain being with unknown domain shifts, attempting to approximate these gaps by source domains may be the key to improving model generalization capability. Existing works mainly focus on sharing and recombining local domain-specific attributes to increase data diversity and simulate potential domain shifts. However, these methods may be insufficient since only the local attribute recombination can be hard to touch the out-of-distribution of global data. In this paper, we propose a simple-yet-efficient framework named Federated Domain Adversarial Generation (FedDAG). It aims to simulate the domain shift and improve the model generalization by adversarially generating novel domains different from local and global source domains. Specifically, it generates novel-style images by maximizing the instance-level feature discrepancy between original and generated images and trains a generalizable task model by minimizing their feature discrepancy. Further, we observed that FedDAG could cause different performance improvements for local models. It may be due to inherent data isolation and heterogeneity among clients, exacerbating the imbalance in their generalization contributions to the global model. Ignoring this imbalance can lead the global model's generalization ability to be sub-optimal, further limiting the novel domain generation procedure. Thus, to mitigate this imbalance, FedDAG hierarchically aggregates local models at the within-client and across-client levels by using the sharpness concept to evaluate client model generalization contributions. Extensive experiments across four medical benchmarks demonstrate FedDAG's ability to enhance generalization in federated medical scenarios.

Figures (14)

• Figure 1: The potential solutions of approximating domain shifts in federated scenarios: (a) sharing and recombining local domain-specific attributes and (b) adversarial novel domain generation.
• Figure 2: An overview of FedDAG. It enhances the model's generalization by locally performing NDAG. Subsequently, SHA is employed to mitigate the NDAG-exacerbated imbalance of generalization contributions and further promote the novel domain generation.
• Figure 3: Examples from WILDS-Camelyon17, MIDOG2022, GDRBench and ISIC2019. These Benchmarks visually display subtle domain gaps.
• Figure 4: Left: visualizations of generated images and perturbations. Right: original and generated samples from different benchmarks.
• Figure 5: Left: effects of $\alpha$, $\rho$, $k$, and distribution strategies. The upper x-axis represents values of $\rho$ and the bottom represents values of $k$ and ($\alpha$). Right: the effects of different distribution strategies of $\mathcal{F}_{\theta}$.