Sparse variable principal component analysis with application to fMRI

Multivoxel methods such as Principal Component analysis (PCA) and Independent component analysis (ICA) have been found to be useful in fMRI data analysis. They can extract biologically interpretable components without any knowledge of the experimental settings. Interesting brain networks such as the motor or the visual cortex typically have sparse spatial structure that PCA or ICA do not make use of. Sparse PCA is a new class of methods that is able to null out voxels containing only noise therefore getting more accurate results. In this paper we apply our own previously introduced sparse PCA method for the first time on real fMRI data. Additionally, we use different estimation method, which is much faster than the one previously introduced, therefore making the method more attractive for large fMRI data sets.