Related Experiment Video
Updated: Jan 12, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
Federated nnU-Net for privacy-preserving medical image segmentation
Grzegorz Skorupko1, Fotios Avgoustidis2, Carlos Martín-Isla2
1Artificial Intelligence in Medicine Laboratory (BCN-AIM), Departament de Matemàtiques i Informàtica, Universitat de Barcelona, 08007, Barcelona, Spain. grzegorz.skorupko@ub.edu.
Scientific Reports
|November 3, 2025
Summary
FednnU-Net enables decentralized medical image segmentation using federated learning, enhancing privacy. This framework achieves high performance across various segmentation tasks, democratizing AI in healthcare.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Computer Vision
Background:
- The nnU-Net framework is a leading tool for medical image segmentation but typically requires centralized data, posing privacy risks.
- Centralized training can lead to sensitive patient data leakage and privacy violations.
- Federated learning offers a decentralized approach to train models collaboratively while preserving patient privacy.
Purpose of the Study:
- To introduce FednnU-Net, a federated learning extension for the nnU-Net framework.
- To enable decentralized training of nnU-Net for medical image segmentation.
- To enhance patient privacy in medical AI development.
Main Methods:
- Development of FednnU-Net, a plug-and-play federated learning extension for nnU-Net.
- Introduction of two novel federated methodologies: Federated Fingerprint Extraction (FFE) and Asymmetric Federated Averaging (AsymFedAvg).
- Multi-modal segmentation experiments on breast, cardiac, and fetal datasets from 18 institutions.
Main Results:
- FednnU-Net demonstrated high and consistent performance in breast, cardiac, and fetal segmentation tasks.
- The proposed federated methodologies (FFE and AsymFedAvg) effectively enabled decentralized nnU-Net training.
- Experiments validated the framework's robustness across diverse datasets and institutions.
Conclusions:
- FednnU-Net successfully extends nnU-Net capabilities to decentralized, privacy-preserving medical image segmentation.
- The framework democratizes advanced AI research and clinical deployment by enabling collaborative, decentralized training.
- Public release of the FednnU-Net framework aims to foster wider adoption and development in clinical settings.
