Federated Learning for Thyroid Ultrasound Image Analysis to Protect Personal Information: Validation Study in a Real Health Care Environment.
deep learning
federated learning
thyroid nodules
ultrasound image
Journal
JMIR medical informatics
ISSN: 2291-9694
Titre abrégé: JMIR Med Inform
Pays: Canada
ID NLM: 101645109
Informations de publication
Date de publication:
18 May 2021
18 May 2021
Historique:
received:
19
11
2020
accepted:
03
04
2021
revised:
02
02
2021
pubmed:
17
4
2021
medline:
17
4
2021
entrez:
16
4
2021
Statut:
epublish
Résumé
Federated learning is a decentralized approach to machine learning; it is a training strategy that overcomes medical data privacy regulations and generalizes deep learning algorithms. Federated learning mitigates many systemic privacy risks by sharing only the model and parameters for training, without the need to export existing medical data sets. In this study, we performed ultrasound image analysis using federated learning to predict whether thyroid nodules were benign or malignant. The goal of this study was to evaluate whether the performance of federated learning was comparable with that of conventional deep learning. A total of 8457 (5375 malignant, 3082 benign) ultrasound images were collected from 6 institutions and used for federated learning and conventional deep learning. Five deep learning networks (VGG19, ResNet50, ResNext50, SE-ResNet50, and SE-ResNext50) were used. Using stratified random sampling, we selected 20% (1075 malignant, 616 benign) of the total images for internal validation. For external validation, we used 100 ultrasound images (50 malignant, 50 benign) from another institution. For internal validation, the area under the receiver operating characteristic (AUROC) curve for federated learning was between 78.88% and 87.56%, and the AUROC for conventional deep learning was between 82.61% and 91.57%. For external validation, the AUROC for federated learning was between 75.20% and 86.72%, and the AUROC curve for conventional deep learning was between 73.04% and 91.04%. We demonstrated that the performance of federated learning using decentralized data was comparable to that of conventional deep learning using pooled data. Federated learning might be potentially useful for analyzing medical images while protecting patients' personal information.
Sections du résumé
BACKGROUND
BACKGROUND
Federated learning is a decentralized approach to machine learning; it is a training strategy that overcomes medical data privacy regulations and generalizes deep learning algorithms. Federated learning mitigates many systemic privacy risks by sharing only the model and parameters for training, without the need to export existing medical data sets. In this study, we performed ultrasound image analysis using federated learning to predict whether thyroid nodules were benign or malignant.
OBJECTIVE
OBJECTIVE
The goal of this study was to evaluate whether the performance of federated learning was comparable with that of conventional deep learning.
METHODS
METHODS
A total of 8457 (5375 malignant, 3082 benign) ultrasound images were collected from 6 institutions and used for federated learning and conventional deep learning. Five deep learning networks (VGG19, ResNet50, ResNext50, SE-ResNet50, and SE-ResNext50) were used. Using stratified random sampling, we selected 20% (1075 malignant, 616 benign) of the total images for internal validation. For external validation, we used 100 ultrasound images (50 malignant, 50 benign) from another institution.
RESULTS
RESULTS
For internal validation, the area under the receiver operating characteristic (AUROC) curve for federated learning was between 78.88% and 87.56%, and the AUROC for conventional deep learning was between 82.61% and 91.57%. For external validation, the AUROC for federated learning was between 75.20% and 86.72%, and the AUROC curve for conventional deep learning was between 73.04% and 91.04%.
CONCLUSIONS
CONCLUSIONS
We demonstrated that the performance of federated learning using decentralized data was comparable to that of conventional deep learning using pooled data. Federated learning might be potentially useful for analyzing medical images while protecting patients' personal information.
Identifiants
pubmed: 33858817
pii: v9i5e25869
doi: 10.2196/25869
pmc: PMC8170555
doi:
Types de publication
Journal Article
Langues
eng
Pagination
e25869Informations de copyright
©Haeyun Lee, Young Jun Chai, Hyunjin Joo, Kyungsu Lee, Jae Youn Hwang, Seok-Mo Kim, Kwangsoon Kim, Inn-Chul Nam, June Young Choi, Hyeong Won Yu, Myung-Chul Lee, Hiroo Masuoka, Akira Miyauchi, Kyu Eun Lee, Sungwan Kim, Hyoun-Joong Kong. Originally published in JMIR Medical Informatics (https://medinform.jmir.org), 18.05.2021.
Références
Healthc Inform Res. 2019 Jul;25(3):201-211
pubmed: 31406612
Nature. 2017 Feb 2;542(7639):115-118
pubmed: 28117445
Brainlesion. 2019;11383:92-104
pubmed: 31231720
Ultrasonics. 2017 Jan;73:221-230
pubmed: 27668999
J Appl Clin Med Phys. 2019 Mar;20(3):115-124
pubmed: 30861278
Ann Intern Med. 1994 Jan 15;120(2):135-42
pubmed: 8256973
Healthc Inform Res. 2020 Jan;26(1):13-19
pubmed: 32082696
BMC Health Serv Res. 2012 Sep 08;12:309
pubmed: 22958365
Medicine (Baltimore). 2019 Apr;98(15):e15133
pubmed: 30985680
J Clin Med. 2019 Nov 14;8(11):
pubmed: 31739517
Cell. 2018 Feb 22;172(5):1122-1131.e9
pubmed: 29474911
Biomed Opt Express. 2020 May 11;11(6):2976-2995
pubmed: 32637236
IEEE Trans Ultrason Ferroelectr Freq Control. 2020 Jul;67(7):1344-1353
pubmed: 32054578
IEEE Trans Pattern Anal Mach Intell. 2020 Aug;42(8):2011-2023
pubmed: 31034408