FL-MedSegBench: A Comprehensive Benchmark for Federated Learning on Medical Image Segmentation

Meilu Zhu1,†Zhiwei Wang1,†Axiu Mao2Yuxing Li1Xiaohan Xing3Yixuan Yuan4Edmund Y. Lam1,*

1Department of Electrical and Electronic Engineering, The University of Hong Kong 2School of Communication Engineering, Hangzhou Dianzi University 3Department of Radiation Oncology, Stanford University 4Department of Electronic Engineering, The Chinese University of Hong Kong

Paper Code 🗂 Datasets
FL-MedSegBench overview: datasets, modalities, tasks and federated learning methods

Overview of FL-MedSegBench — a unified pipeline for evaluating global and personalized federated learning across diverse medical segmentation datasets, modalities and tasks.

Overview

FL-MedSegBench is a benchmark for federated learning (FL) in medical image segmentation. It provides a unified experimental framework to systematically evaluate both global FL (gFL) and personalized FL (pFL) methods across diverse medical imaging scenarios.

27+ datasets

Public medical segmentation datasets spanning brain, heart, prostate, pancreas, eye, gland, polyp and tissue-level targets.

10+ modalities

MRI, ultrasound, endoscopy, histology, fundus / retinal imaging, microscopy and other clinical modalities.

13+ methods

Representative gFL and pFL baselines, all trained and evaluated under one standardized, reproducible pipeline.

Implemented Methods

FedAvg
FedProx
FedAWA
FedRDN
FedIWS
MOON
FedNova
PN
Ditto
FedBN
FedPer
FedRoD
SioBN

Citation

@article{zhu2026fl,
  title   = {FL-MedSegBench: A Comprehensive Benchmark for Federated Learning on Medical Image Segmentation},
  author  = {Zhu, Meilu and Wang, Zhiwei and Mao, Axiu and Li, Yuxing and Xing, Xiaohan and Yuan, Yixuan and Lam, Edmund Y},
  journal = {arXiv preprint arXiv:2603.11659},
  year    = {2026}
}