Adopting transfer learning for neuroimaging: a comparative analysis with a custom 3D convolution neural network model.
Brain Neurodegenerative Disorders
Convolution Neural Networks
Medical Image Classification
Transfer Learning
Journal
BMC medical informatics and decision making
ISSN: 1472-6947
Titre abrégé: BMC Med Inform Decis Mak
Pays: England
ID NLM: 101088682
Informations de publication
Date de publication:
07 12 2022
07 12 2022
Historique:
received:
10
11
2022
accepted:
15
11
2022
entrez:
8
12
2022
pubmed:
9
12
2022
medline:
15
12
2022
Statut:
epublish
Résumé
In recent years, neuroimaging with deep learning (DL) algorithms have made remarkable advances in the diagnosis of neurodegenerative disorders. However, applying DL in different medical domains is usually challenged by lack of labeled data. To address this challenge, transfer learning (TL) has been applied to use state-of-the-art convolution neural networks pre-trained on natural images. Yet, there are differences in characteristics between medical and natural images, also image classification and targeted medical diagnosis tasks. The purpose of this study is to investigate the performance of specialized and TL in the classification of neurodegenerative disorders using 3D volumes of 18F-FDG-PET brain scans. Results show that TL models are suboptimal for classification of neurodegenerative disorders, especially when the objective is to separate more than two disorders. Additionally, specialized CNN model provides better interpretations of predicted diagnosis. TL can indeed lead to superior performance on binary classification in timely and data efficient manner, yet for detecting more than a single disorder, TL models do not perform well. Additionally, custom 3D model performs comparably to TL models for binary classification, and interestingly perform better for diagnosis of multiple disorders. The results confirm the superiority of the custom 3D-CNN in providing better explainable model compared to TL adopted ones.
Sections du résumé
BACKGROUND
In recent years, neuroimaging with deep learning (DL) algorithms have made remarkable advances in the diagnosis of neurodegenerative disorders. However, applying DL in different medical domains is usually challenged by lack of labeled data. To address this challenge, transfer learning (TL) has been applied to use state-of-the-art convolution neural networks pre-trained on natural images. Yet, there are differences in characteristics between medical and natural images, also image classification and targeted medical diagnosis tasks. The purpose of this study is to investigate the performance of specialized and TL in the classification of neurodegenerative disorders using 3D volumes of 18F-FDG-PET brain scans.
RESULTS
Results show that TL models are suboptimal for classification of neurodegenerative disorders, especially when the objective is to separate more than two disorders. Additionally, specialized CNN model provides better interpretations of predicted diagnosis.
CONCLUSIONS
TL can indeed lead to superior performance on binary classification in timely and data efficient manner, yet for detecting more than a single disorder, TL models do not perform well. Additionally, custom 3D model performs comparably to TL models for binary classification, and interestingly perform better for diagnosis of multiple disorders. The results confirm the superiority of the custom 3D-CNN in providing better explainable model compared to TL adopted ones.
Identifiants
pubmed: 36476613
doi: 10.1186/s12911-022-02054-7
pii: 10.1186/s12911-022-02054-7
pmc: PMC9727842
doi:
Types de publication
Journal Article
Research Support, N.I.H., Extramural
Research Support, U.S. Gov't, Non-P.H.S.
Research Support, Non-U.S. Gov't
Langues
eng
Sous-ensembles de citation
IM
Pagination
318Subventions
Organisme : VINNOVA
ID : 2017-02447
Organisme : NIA NIH HHS
ID : U01 AG024904
Pays : United States
Informations de copyright
© 2022. The Author(s).
Références
Arch Neurol. 2009 Sep;66(9):1151-7
pubmed: 19752306
Can J Neurol Sci. 2016 Apr;43 Suppl 1:S83-95
pubmed: 27307129
Radiology. 2019 Feb;290(2):456-464
pubmed: 30398430
Neuroimaging Clin N Am. 2005 Nov;15(4):869-77, xi-xii
pubmed: 16443497
Neurology. 2005 Dec 27;65(12):1863-72
pubmed: 16237129
Eur J Nucl Med Mol Imaging. 2020 Feb;47(2):403-412
pubmed: 31768599
Lancet. 2005 Dec 17;366(9503):2112-7
pubmed: 16360788
Radiographics. 2014 May-Jun;34(3):684-701
pubmed: 24819789
Eur J Nucl Med Mol Imaging. 2022 Jan;49(2):563-584
pubmed: 34328531
Eur J Neurol. 2018 Oct;25(10):1201-1217
pubmed: 29932266
IEEE Rev Biomed Eng. 2019;12:19-33
pubmed: 30561351
J Alzheimers Dis. 2017;57(3):787-795
pubmed: 28304294
Proc SPIE Int Soc Opt Eng. 2017 Oct;10572:
pubmed: 29263566
Sensors (Basel). 2019 Jun 11;19(11):
pubmed: 31212698
Neuroimage. 1995 Jun;2(2):89-101
pubmed: 9343592