G5 Doctoral dissertation (article)
Harmonization of multi-site MRI data (2023)
Monen sivuston MRI-tietojen harmonisointi
Xu, H. (2023). Harmonization of multi-site MRI data [Doctoral dissertation]. University of Jyväskylä. JYU dissertations, 736. https://urn.fi/URN:ISBN:978-951-39-9884-4
JYU authors or editors
Publication details
All authors or editors: Xu, Huashuai
eISBN: 978-951-39-9884-4
Journal or series: JYU dissertations
eISSN: 2489-9003
Publication year: 2023
Number in series: 736
Number of pages in the book: 1 verkkoaineisto (73, 12 sivua, 41 numeroimatonta sivua)
Publisher: University of Jyväskylä
Place of Publication: Jyväskylä
Publication country: Finland
Publication language: Finnish
Persistent website address: https://urn.fi/URN:ISBN:978-951-39-9884-4
Publication open access: Openly available
Publication channel open access: Open Access channel
Abstract
Combining magnetic resonance imaging (MRI) data from different sites is now common to improve research with larger, more varied groups, which makes studies more powerful and representative. However, this approach faces challenges due to differences in MRI scanners that can distort results. Two methods, independent component analysis (ICA) and general linear model (GLM), are used to correct these site effects, but they struggle to fully remove them without affecting the data's real signals, especially when these signals are related to the very scanner differences they aim to correct. In this thesis, we introduced an effective noise-reduction method utilizing the dual-projection (DP) concept grounded on independent component analysis (ICA) to mitigate site-specific influences in combined data. This method can separate the signal effects from the identified site-related components and then remove site effects without losing signals of interest. To validate the method's effectiveness, we simulated two scenarios, one where the site and signal variables are correlated and another where they are not. Structural and functional MRI data from the Autism Brain Imaging Data Exchange II and a traveling subject dataset from the Strategic Research Program for Brain Sciences were employed to test the ICA-DP methods for removing site effects and preserving signal effects. We also proposed an innovative multimodal denoising approach that employs a dual projection (DP) methodology grounded on linked independent component analysis (LICA) to remove the site effects. Compared with unimodal studies, using LICA on multimodal MRI data offers a more precise estimation of site effects. Structural and functional MRI data from Autism Brain Imaging Data Exchange II validated the LICA-DP methods. In conclusion, our approaches using ICA-DP and LICA-DP have demonstrated their efficacy in mitigating site-related influences while maintaining biological variation. Such a strategy can greatly boost the validity of neuroimaging studies, and we are confident it will be an indispensable resource for forthcoming research.
Keywords: magnetic resonance imaging; reliability (general); multimodality; independent component analysis; projection (modelling); doctoral dissertations
Free keywords: multi-site; magnetic resonance imaging; site effects; biological variability; multimodal; dual-projection; independent component analysis
Contributing organizations
Ministry reporting: Yes
VIRTA submission year: 2023