Improving multiple sclerosis lesion segmentation across clinical sites: A federated learning approach with noise-resilient training

Lei Bai, Dongang Wang, Hengrui Wang, Michael Barnett, Mariano Cabezas, Weidong Cai, Fernando Calamante, Kain Kyle, Dongnan Liu, Linda Ly, Aria Nguyen, Chun-Chien Shieh, Ryan Sullivan, Geng Zhan, Wanli Ouyang, Chenyu Wang

Abstract: Highlights•The first MS lesion segmentation with noisy labels in a federated learning paradigm.•A Decoupled Hard Label Correction strategy is proposed to correct lesion lables.•A Centrally Enhanced Label Correction strategy is introduced to improve DHLC.•Comparable performance to centralized training with clean labels on real-world data.