CRLS-PCA Based Independent Component Analysis for fMRI Study

Data reduction through conventional principal component analysis is impractical for temporal independent component analysis (tICA) on fMRI data, since the data covariance matrix is too huge to be manipulated. It is also computationally intensive for spatial ICA (sICA) on long time fMRI scans. To solve this problem, a cascade recursive least squared networks based PCA (CRLS-PCA) was used to reduce the fMRI data in this paper. Without the need to compute data covariance matrix CRLS-PCA can extract arbitrary number of PCs directly from the original data, which simultaneously saves time for data reduction. Experiment results were given to evaluate the performance of CRLS-PCA based tICA and sICA in fMRI study