Federated Medical Image Classification under Class and Domain Imbalance exploiting Synthetic Sample Generation

arXiv cs.CV / 4/30/2026

📰 NewsDeveloper Stack & InfrastructureModels & Research

共有:

Key Points

The paper proposes FedSSG, a new federated learning framework for medical image classification that targets privacy constraints and domain shifts from heterogeneous imaging devices.
It introduces a synthetic sample generation strategy that is distributed across federated clients to reduce class imbalance, especially improving representation of rare pathologies.
Experimental results indicate improved model performance and better generalization across different institutions and imaging setups compared with baseline approaches.
The method is designed to add minimal computational overhead on clients, making it more practical for real-world federated deployments in healthcare.
Overall, the work combines federated learning with synthetic data coverage to jointly address both device-related domain imbalance and pathology-related class imbalance.

Abstract

Exploiting deep learning in medical imaging faces critical challenges, including strict privacy constraints, heterogeneous imaging devices with varying acquisition properties, and class imbalance due to the uneven prevalence of pathologies. In this work, we propose FedSSG, a novel Federated Learning framework that addresses domain shifts caused by diverse imaging devices while mitigating the under-representation of rare pathologies. The key contribution is a strategy for generating synthetic samples and distributing them across clients to improve coverage of both underrepresented pathologies and imaging devices. Experimental results demonstrate that our approach significantly enhances model performance and generalization across heterogeneous institutions, with minimal computational overhead at the client side.