An Improved Group BSS-CCA Method for Blind Source Separation of Functional MRI Scans of the Human Brain

摘要

An improved canonical correlation analysis (CCA) approach for multi-subject blind source separation (BSS) of brain functional magnetic resonance imaging (fMRI) data is proposed. Group-level comparison analysis has attracted increasing interest in the human brain fMRI analysis. Canonical correlation analysis for blind source separation (BSS-CCA) relies on the fact that all meaningful real signals are auto-correlated, compared with white noise which generally is not be cared for. Merely requiring that secondorder statistic is zero, BSS-CCA is looser than independent component analysis (ICA) which demands mutual statistics of all orders to be zero. We proposed an improved group BSSCCA approach for the analysis of multi-subject fMRI data based on spatial BSS-CCA, termed as group dual BSS-CCA. The real data experiment results revealed that the group dual BSS-CCA approach was effective to extract networks which were functionally distinct from the resting-state fMRI data of the human brain.