Constrained Spatiotemporal ICA and Its Application for fMRI Data Analysis

In general, Independent Component Analysis (ICA), a blind source separation technique, is used in either spatial or temporal domain. For spatiotemporal data, however, in which the statistical independence criteria of the underlying sources cannot be guaranteed, ICA does not perform well simultaneously in both domains. Thus, spatiotemporal I A (stICA) has been proposed. However, these conventional ICAs in spatial, temporal, or spatiotemporal mode suffer from a problem of source ambiguity: each source must be identified by users. To extract only the desired sources, recently constrained ICA (cICA) has been proposed in which a priori information of the desired sources are used as constraints in either spatial or temporal domain. In this study, we propose constrained spatiotemporal ICA (constrained-stICA), as an extension of cICA, for better analysis of spatiotemporal data. The proposed algorithm tries to find the maximally independent, yet desired sources in both the spatial and temporal domains. The performance of the proposed algorithm is tested against the conventional ICAs using simulated data. Then, its application to the analysis of spatiotemporal fMRI data indicates improved data analysis against the conventional functional MRI data analysis via SPM.