Multiple sclerosis lesion segmentation: revisiting weighting mechanisms for federated learning

• DOI: 10.3389/fnins.2023.1167612
• Publication Date: 2023-05-18T07:00:44Z
• Authors: Dongnan Liu; Mariano Cabezas; Dongang Wang; Zihao Tang; Lei Bai; Geng Zhan; Yuling Luo; Kain Kyle; Linda Ly; James Yu; Chun-Chien Shieh; Aria Nguyen; Ettikan Kandasamy...; Ryan Sullivan; Fernando Calamante; Michael Barnett; Wanli Ouyang; Weidong Cai; Chenyu Wang
• Keywords: Dice; Sørensen–Dice coefficient

Background and introduction Federated learning (FL) has been widely employed for medical image analysis to facilitate multi-client collaborative without sharing raw data. Despite great success, FL's applications remain suboptimal in neuroimage tasks such as lesion segmentation multiple sclerosis (MS), due variance characteristics imparted by different scanners acquisition parameters. Methods In this work, we propose the first FL MS framework via two effective re-weighting mechanisms. Specifically, a learnable weight is assigned each local node during aggregation process, based on its performance. addition, loss function client also re-weighted according volume data training. Results The proposed method validated scenarios using public clinical datasets. case-wise voxel-wise Dice score of under dataset 65.20 74.30, respectively. On second in-house dataset, 53.66, 62.31, Discussions conclusions Comparison experiments datasets have demonstrated effectiveness significantly outperforming other methods. Furthermore, performance incorporating our mechanism can achieve comparable that from centralized training with all