Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Federated nnU-Net for Privacy-Preserving Medical Image Segmentation

Created by
  • Haebom

Author

Grzegorz Skorupko, Fotios Avgoustidis, Carlos Mart in-Isla, Lidia Garrucho, Dimitri A. Kessler, Esmeralda Ruiz Pujadas, Oliver D iaz, Maciej Bobowicz, Katarzyna Gwo zdziewicz, Xavier Bargall o, Paulius Jaru\v{s}evi\v{c}ius, Richard Osuala, Kaisar Kushibar, Karim Lekadir

Outline

This paper proposes FednnU-Net, a distributed learning-based framework, to overcome the limitations of the centralized approach of the nnU-Net framework, which has established itself as the gold standard in medical image segmentation (risk of sensitive patient information leakage and privacy violations). FednnU-Net can be integrated into nnU-Net in a plug-and-play manner and presents two distributed learning methodologies: Federated Fingerprint Extraction (FFE) and Asymmetric Federated Averaging (AsymFedAvg). Experimental results demonstrate high and consistent performance in multimodal experiments using six datasets from 18 institutions for breast, heart, and fetal segmentation tasks. The framework is publicly available ( https://github.com/faildeny/FednnUNet ).

Takeaways, Limitations

•
Takeaways:
◦
Enhancing patient privacy by enabling distributed learning of nnU-Net.
◦
We present two effective distributed learning methodologies, FFE and AsymFedAvg.
◦
High performance verified in various medical image segmentation tasks.
◦
Democratizing research and practical application through open source disclosure.
•
Limitations:
◦
Further comparative analysis of the proposed methodology with other distributed learning methodologies may be needed.
◦
Additional experiments with different datasets and clinical settings may be needed.
◦
Additional validation and evaluation may be required for practical clinical application.
👍
Made with Slashpage