MeshFL: A Federated MeshNet Framework for 3D Brain MRI Segmentation

Résumé fourni par la source

Advances in federated learning paved the way for privacy-preserving collaborative training of machine learning models on decentralized datasets.This is particularly useful in neuroimaging, where sensitive data, such as brain MRI scans, cannot be easily shared across institutions.MeshFL is an open-source framework designed to facilitate distributed training of deep learning models for 3D brain MRI segmentation while maintaining data privacy.Built upon NVFlare (NVIDIA, 2023), MeshFL employs federated learning principles to train MeshNet models (Fedorov et al., 2017) across multiple data sites, enabling high-accuracy segmentation of white and gray matter regions.With Dice scores of approximately 0.92 for training and approximately 0.90 for validation, MeshFL demonstrates effective 3D brain MRI segmentation under the federated training configuration.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.