Related Experiment Video
Updated: Jul 3, 2026
13:51
Cross-Modal Multivariate Pattern Analysis
Published on: November 9, 2011
Combining multivariate voxel selection and support vector machines for mapping and classification of fMRI spatial
Federico De Martino1, Giancarlo Valente, Noël Staeren
1Department of Cognitive Neurosciences, Faculty of Psychology, University of Maastricht, Maastricht, Postbus 616, 6200 MD, Maastricht, The Netherlands. f.demartino@psychology.unimaas.nl
Neuroimage
|August 2, 2008
Summary
This study introduces Recursive Feature Elimination (RFE) for improved functional brain mapping. RFE effectively identifies informative brain activation patterns, enhancing classification accuracy in fMRI data analysis.
Area of Science:
- Neuroimaging
- Machine Learning
- Pattern Recognition
Background:
- Functional brain mapping uses pattern recognition to detect informative multivoxel activation patterns.
- Sensitivity is reduced when discriminative voxels are a small fraction of total voxels, posing a dimensionality problem.
- Previous methods used univariate voxel selection or region-of-interest strategies before machine learning.
Purpose of the Study:
- To introduce and evaluate a multivariate feature selection algorithm, Recursive Feature Elimination (RFE), for functional imaging data classification.
- To compare RFE's performance against univariate voxel selection strategies.
- To assess RFE's effectiveness in identifying informative spatial patterns and improving generalization performance.
Main Methods:
- Employed Recursive Feature Elimination (RFE), a multivariate feature selection algorithm using support vector machines recursively.
- Evaluated RFE using simulated fMRI data and real high-resolution auditory fMRI data.
- Compared RFE with univariate activation-based (F-test) voxel reduction and statistical learning without feature selection.
Main Results:
- RFE demonstrated suitability for mapping discriminative patterns in fMRI data.
- Combining univariate activation-based reduction with multivariate RFE yielded optimal results, especially with low contrast-to-noise ratios.
- The recursive algorithm successfully detected and classified multivoxel spatial patterns in auditory fMRI data, identifying the superior temporal gyrus's role in sound category encoding.
Conclusions:
- Recursive Feature Elimination (RFE) enhances the sensitivity and generalization performance of functional brain mapping.
- The combined approach of univariate and multivariate feature selection is superior to univariate methods alone.
- This method accurately classifies multivoxel patterns and reveals specific brain regions involved in cognitive processes, outperforming traditional statistical analyses.
