Multiple sclerosis lesion segmentation: revisiting weighting mechanisms for federated learning.
MRI
deep learning
federated learning
multiple sclerosis
segmentation
Journal
Frontiers in neuroscience
ISSN: 1662-4548
Titre abrégé: Front Neurosci
Pays: Switzerland
ID NLM: 101478481
Informations de publication
Date de publication:
2023
2023
Historique:
received:
16
02
2023
accepted:
24
04
2023
medline:
5
6
2023
pubmed:
5
6
2023
entrez:
5
6
2023
Statut:
epublish
Résumé
Federated 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. In 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. The 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. The 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.
Sections du résumé
Background and introduction
UNASSIGNED
Federated 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.
Methods
UNASSIGNED
In 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.
Results
UNASSIGNED
The 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 conclusions
UNASSIGNED
The 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.
Identifiants
pubmed: 37274196
doi: 10.3389/fnins.2023.1167612
pmc: PMC10232857
doi:
Types de publication
Journal Article
Langues
eng
Pagination
1167612Informations de copyright
Copyright © 2023 Liu, Cabezas, Wang, Tang, Bai, Zhan, Luo, Kyle, Ly, Yu, Shieh, Nguyen, Kandasamy Karuppiah, Sullivan, Calamante, Barnett, Ouyang, Cai and Wang.
Déclaration de conflit d'intérêts
EK is employed by NVIDIA Corporation, Singapore. DW, GZ, YL, KK, LL, JY, C-CS, AN, and CW are employees at Sydney Neuroimaging Analysis Centre. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Références
IEEE Trans Med Imaging. 2002 Oct;21(10):1280-91
pubmed: 12585710
Neuroimage. 2006 Feb 1;29(3):859-67
pubmed: 16203159
Med Image Comput Comput Assist Interv. 2021;12902:549-559
pubmed: 34734216
IEEE Trans Med Imaging. 2020 Jul;39(7):2494-2505
pubmed: 32054572
AJNR Am J Neuroradiol. 2021 Nov;42(11):1927-1933
pubmed: 34531195
Neuroimage. 2021 Dec 1;244:118589
pubmed: 34563682
Front Neurol. 2015 Mar 04;6:40
pubmed: 25788892
Neuroimage. 2017 Jul 15;155:159-168
pubmed: 28435096
Magn Reson Imaging. 2019 Dec;64:160-170
pubmed: 31301354
IEEE Trans Med Imaging. 2021 Jan;40(1):154-165
pubmed: 32915732
Comput Med Imaging Graph. 2018 Dec;70:83-100
pubmed: 30326367
Neuroimage. 2019 Aug 1;196:1-15
pubmed: 30953833
Neuroimage Clin. 2017 Feb 04;14:391-399
pubmed: 28271039
Neuroimage Clin. 2019;21:101638
pubmed: 30555005
IEEE Trans Med Imaging. 2016 May;35(5):1229-1239
pubmed: 26886978
N Engl J Med. 2008 Oct 23;359(17):1786-801
pubmed: 18946064
Front Comput Neurosci. 2020 Mar 09;14:19
pubmed: 32210780
Ann Neurol. 2011 Feb;69(2):292-302
pubmed: 21387374
Neuroinformatics. 2018 Jan;16(1):51-63
pubmed: 29103086
Neuroimage. 2021 Jan 15;225:117471
pubmed: 33099007
IEEE J Biomed Health Inform. 2022 Jun;26(6):2680-2692
pubmed: 35171783
Front Neurol. 2019 Mar 15;10:188
pubmed: 30930829
Nat Methods. 2021 Feb;18(2):203-211
pubmed: 33288961
Neuroimage Clin. 2020;25:102104
pubmed: 31927500
Front Neurol. 2018 Jan 23;9:5
pubmed: 29410647
Med Image Anal. 2020 Oct;65:101765
pubmed: 32679533
Front Neurosci. 2019 Aug 16;13:810
pubmed: 31474816
Proc IEEE Comput Soc Conf Comput Vis Pattern Recognit. 2021 Jun;2021:2423-2432
pubmed: 35444379
Neuroimage. 2017 Mar 1;148:77-102
pubmed: 28087490
Front Neurosci. 2018 Nov 08;12:818
pubmed: 30467462
Med Image Anal. 2020 Jan;59:101557
pubmed: 31677438
Neuroimage. 2019 Apr 15;190:32-45
pubmed: 28917696
Front Neurosci. 2019 Feb 28;13:97
pubmed: 30872986
Front Neurosci. 2014 Aug 20;8:229
pubmed: 25191215
Sci Rep. 2018 Sep 12;8(1):13650
pubmed: 30209345