Multiple sclerosis lesion segmentation: revisiting weighting mechanisms for federated learning

Frontiers in Neuroscience

Neuroscience

Mariano Cabezas¹  Fernando Calamante²  Lei Bai³  Linda Ly⁴  Yuling Luo⁴  Kain Kyle⁴  Chun-Chien Shieh⁴  Geng Zhan⁴  Michael Barnett⁴  Aria Nguyen⁴  Dongang Wang⁴  James Yu⁴  Chenyu Wang⁴  Ettikan Kandasamy Karuppiah⁵  Ryan Sullivan⁶  Weidong Cai⁷  Dongnan Liu⁸  Zihao Tang⁸  Wanli Ouyang⁹ 

[1] Brain and Mind Centre, The University of Sydney, Sydney, NSW, Australia;Brain and Mind Centre, The University of Sydney, Sydney, NSW, Australia;School of Biomedical Engineering, The University of Sydney, Sydney, NSW, Australia;Sydney Imaging, The University of Sydney, Sydney, NSW, Australia;Brain and Mind Centre, The University of Sydney, Sydney, NSW, Australia;School of Electrical and Information Engineering, The University of Sydney, Sydney, NSW, Australia;Brain and Mind Centre, The University of Sydney, Sydney, NSW, Australia;Sydney Neuroimaging Analysis Centre, Camperdown, NSW, Australia;NVIDIA Corporation, Singapore, Singapore;School of Biomedical Engineering, The University of Sydney, Sydney, NSW, Australia;School of Computer Science, The University of Sydney, Sydney, NSW, Australia;School of Computer Science, The University of Sydney, Sydney, NSW, Australia;Brain and Mind Centre, The University of Sydney, Sydney, NSW, Australia;School of Electrical and Information Engineering, The University of Sydney, Sydney, NSW, Australia;

关键词: deep learning;    federated learning;    multiple sclerosis;    segmentation;    MRI;   

DOI  :  10.3389/fnins.2023.1167612

 received in 2023-02-16, accepted in 2023-04-24,  发布年份 2023

来源: Frontiers

【 摘 要 】

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

【 授权许可】

Unknown   
Copyright © 2023 Liu, Cabezas, Wang, Tang, Bai, Zhan, Luo, Kyle, Ly, Yu, Shieh, Nguyen, Kandasamy Karuppiah, Sullivan, Calamante, Barnett, Ouyang, Cai and Wang.