Related Experiment Videos
Mixed-effects and fMRI studies
K J Friston1, K E Stephan, T E Lund
1The Wellcome Department of Imaging Neuroscience, Institute of Neurology, 12 Queen Square, London WC1N 3BG, UK.
Neuroimage
|December 14, 2004
Summary
This study clarifies mixed-effect (MFX) analyses in multisession functional magnetic resonance imaging (fMRI) by comparing a full MFX approach with the common two-stage summary statistics method. The findings highlight when these statistical approaches yield different results in fMRI group analysis.
Area of Science:
- Neuroimaging
- Statistical analysis
- Cognitive neuroscience
Background:
- Multisession functional magnetic resonance imaging (fMRI) studies commonly employ a two-stage summary statistics procedure for group-level analysis.
- The relationship between this widely adopted procedure and full mixed-effect (MFX) analyses in fMRI is not always clear.
- Understanding these statistical approaches is crucial for accurate population inference in neuroimaging research.
Purpose of the Study:
- To clarify the relationship between full mixed-effect (MFX) analyses and the two-stage summary statistics procedure in multisession fMRI.
- To introduce a procedure for full MFX analyses using restricted maximum likelihood (ReML) estimates within statistical parametric mapping.
- To compare the outcomes of full MFX analysis with the two-stage approach and identify conditions for divergence.
Main Methods:
- Description of a novel procedure for full mixed-effect (MFX) analyses in statistical parametric mapping.
- Utilized restricted maximum likelihood (ReML) estimates of covariance components for the MFX approach.
- Comparative analysis between the full MFX procedure and the conventional two-stage summary statistics method for fMRI group data.
Main Results:
- Demonstrated a method to implement full mixed-effect (MFX) analyses in statistical parametric mapping for multisession fMRI.
- Provided a direct comparison between full MFX and two-stage summary statistics analyses.
- Identified specific scenarios where the two analytical approaches produce divergent results in fMRI group inference.
Conclusions:
- The proposed ReML-based procedure enables full mixed-effect (MFX) analyses in statistical parametric mapping for fMRI.
- Full MFX analyses offer a more comprehensive approach to group-level inference in multisession fMRI compared to the two-stage method.
- Researchers should be aware of the conditions under which the two-stage procedure may lead to different conclusions than a full MFX analysis.