Deep Generative Medical Image Harmonization for Improving Cross-Site Generalization in Deep Learning Predictors.
StarGAN
deep learning
harmonization
Journal
Journal of magnetic resonance imaging : JMRI
ISSN: 1522-2586
Titre abrégé: J Magn Reson Imaging
Pays: United States
ID NLM: 9105850
Informations de publication
Date de publication:
03 2022
03 2022
Historique:
revised:
22
08
2021
received:
07
06
2021
accepted:
23
08
2021
pubmed:
27
9
2021
medline:
27
4
2022
entrez:
26
9
2021
Statut:
ppublish
Résumé
In the medical imaging domain, deep learning-based methods have yet to see widespread clinical adoption, in part due to limited generalization performance across different imaging devices and acquisition protocols. The deviation between estimated brain age and biological age is an established biomarker of brain health and such models may benefit from increased cross-site generalizability. To develop and evaluate a deep learning-based image harmonization method to improve cross-site generalizability of deep learning age prediction. Retrospective. Eight thousand eight hundred and seventy-six subjects from six sites. Harmonization models were trained using all subjects. Age prediction models were trained using 2739 subjects from a single site and tested using the remaining 6137 subjects from various other sites. Brain imaging with magnetization prepared rapid acquisition with gradient echo or spoiled gradient echo sequences at 1.5 T and 3 T. StarGAN v2, was used to perform a canonical mapping from diverse datasets to a reference domain to reduce site-based variation while preserving semantic information. Generalization performance of deep learning age prediction was evaluated using harmonized, histogram matched, and unharmonized data. Mean absolute error (MAE) and Pearson correlation between estimated age and biological age quantified the performance of the age prediction model. Our results indicated a substantial improvement in age prediction in out-of-sample data, with the overall MAE improving from 15.81 (±0.21) years to 11.86 (±0.11) with histogram matching to 7.21 (±0.22) years with generative adversarial network (GAN)-based harmonization. In the multisite case, across the 5 out-of-sample sites, MAE improved from 9.78 (±6.69) years to 7.74 (±3.03) years with histogram normalization to 5.32 (±4.07) years with GAN-based harmonization. While further research is needed, GAN-based medical image harmonization appears to be a promising tool for improving cross-site deep learning generalization. 4 TECHNICAL EFFICACY: Stage 1.
Sections du résumé
BACKGROUND
In the medical imaging domain, deep learning-based methods have yet to see widespread clinical adoption, in part due to limited generalization performance across different imaging devices and acquisition protocols. The deviation between estimated brain age and biological age is an established biomarker of brain health and such models may benefit from increased cross-site generalizability.
PURPOSE
To develop and evaluate a deep learning-based image harmonization method to improve cross-site generalizability of deep learning age prediction.
STUDY TYPE
Retrospective.
POPULATION
Eight thousand eight hundred and seventy-six subjects from six sites. Harmonization models were trained using all subjects. Age prediction models were trained using 2739 subjects from a single site and tested using the remaining 6137 subjects from various other sites.
FIELD STRENGTH/SEQUENCE
Brain imaging with magnetization prepared rapid acquisition with gradient echo or spoiled gradient echo sequences at 1.5 T and 3 T.
ASSESSMENT
StarGAN v2, was used to perform a canonical mapping from diverse datasets to a reference domain to reduce site-based variation while preserving semantic information. Generalization performance of deep learning age prediction was evaluated using harmonized, histogram matched, and unharmonized data.
STATISTICAL TESTS
Mean absolute error (MAE) and Pearson correlation between estimated age and biological age quantified the performance of the age prediction model.
RESULTS
Our results indicated a substantial improvement in age prediction in out-of-sample data, with the overall MAE improving from 15.81 (±0.21) years to 11.86 (±0.11) with histogram matching to 7.21 (±0.22) years with generative adversarial network (GAN)-based harmonization. In the multisite case, across the 5 out-of-sample sites, MAE improved from 9.78 (±6.69) years to 7.74 (±3.03) years with histogram normalization to 5.32 (±4.07) years with GAN-based harmonization.
DATA CONCLUSION
While further research is needed, GAN-based medical image harmonization appears to be a promising tool for improving cross-site deep learning generalization.
LEVEL OF EVIDENCE
4 TECHNICAL EFFICACY: Stage 1.
Identifiants
pubmed: 34564904
doi: 10.1002/jmri.27908
pmc: PMC8844038
mid: NIHMS1735730
doi:
Types de publication
Journal Article
Research Support, N.I.H., Extramural
Research Support, N.I.H., Intramural
Research Support, Non-U.S. Gov't
Langues
eng
Sous-ensembles de citation
IM
Pagination
908-916Subventions
Organisme : NIBIB NIH HHS
ID : R01 EB022573
Pays : United States
Organisme : NIDA NIH HHS
ID : HHSN271201600059C
Pays : United States
Organisme : NIA NIH HHS
ID : U01 AG068057
Pays : United States
Organisme : NIA NIH HHS
ID : RF1 AG059869
Pays : United States
Organisme : NIA NIH HHS
ID : RF1 AG054409
Pays : United States
Organisme : NIH HHS
ID : S10 OD023495
Pays : United States
Organisme : NIA NIH HHS
ID : P30 AG072979
Pays : United States
Organisme : NIMH NIH HHS
ID : R01 MH113565
Pays : United States
Organisme : NIMH NIH HHS
ID : R01 MH120482
Pays : United States
Organisme : NIMH NIH HHS
ID : R01 MH112070
Pays : United States
Organisme : NIA NIH HHS
ID : U19 AG033655
Pays : United States
Informations de copyright
© 2021 International Society for Magnetic Resonance in Medicine.
Références
Alzheimers Dement. 2010 May;6(3):291-6
pubmed: 20451879
Acad Radiol. 2013 Dec;20(12):1566-76
pubmed: 24200484
Med Image Anal. 2011 Apr;15(2):267-82
pubmed: 21233004
Med Image Anal. 2017 Feb;36:61-78
pubmed: 27865153
Proc SPIE Int Soc Opt Eng. 2019 Mar;10949:
pubmed: 31551645
Neuroimage. 2021 Nov;243:118569
pubmed: 34506916
Magn Reson Med. 1999 Dec;42(6):1072-81
pubmed: 10571928
PLoS Med. 2018 Nov 6;15(11):e1002683
pubmed: 30399157
Front Neurol. 2019 Aug 14;10:789
pubmed: 31474922
Neuroimage. 2002 Oct;17(2):825-41
pubmed: 12377157
Neuroimage. 2020 Mar;208:116450
pubmed: 31821869
Rofo. 2009 Aug;181(8):748-59
pubmed: 19598074
PLoS Med. 2015 Mar 31;12(3):e1001779
pubmed: 25826379
Brain. 2020 Jul 1;143(7):2312-2324
pubmed: 32591831
Med Image Anal. 2018 Jan;43:157-168
pubmed: 29107865
J Magn Reson Imaging. 2011 Aug;34(2):438-44
pubmed: 21692137
IEEE Trans Med Imaging. 2019 Sep;38(9):2059-2069
pubmed: 30676951
Nat Commun. 2019 Nov 27;10(1):5409
pubmed: 31776335
Magn Reson Imaging. 2019 Dec;64:160-170
pubmed: 31301354
Biometrics. 1989 Mar;45(1):255-68
pubmed: 2720055
PLoS Comput Biol. 2018 Sep 14;14(9):e1006376
pubmed: 30216352