Federated brain tumor segmentation: an extensive benchmark

TL;DR

This paper conducts a comprehensive benchmark of federated learning algorithms for brain tumor segmentation on the 2022 dataset, analyzing their performance and behavior across different data distribution scenarios.

Contribution

It provides the first extensive comparison of federated learning methods on brain tumor segmentation, highlighting their relative performance and effects of data distribution strategies.

Findings

Standard FedAvg performs very well.

Some methods improve performance slightly.

Federated methods can reduce bias towards dominant data distributions.

Abstract

Recently, federated learning has raised increasing interest in the medical image analysis field due to its ability to aggregate multi-center data with privacy-preserving properties. A large amount of federated training schemes have been published, which we categorize into global (one final model), personalized (one model per institution) or hybrid (one model per cluster of institutions) methods. However, their applicability on the recently published Federated Brain Tumor Segmentation 2022 dataset has not been explored yet. We propose an extensive benchmark of federated learning algorithms from all three classes on this task. While standard FedAvg already performs very well, we show that some methods from each category can bring a slight performance improvement and potentially limit the final model(s) bias toward the predominant data distribution of the federation. Moreover, we provide a deeper understanding of the behaviour of federated learning on this task through alternative ways of distributing the pooled dataset among institutions, namely an Independent and Identical Distributed (IID) setup, and a limited data setup.