Covariance based BSS Algorithm for Functional Magnetic Resonance Imaging (fMRI) data Source Separation

Functional Magnetic Resonance Imaging (fMRI) is a measuring technique used for brain functionality. Observed fMRI data is generally a mixture of hidden sources and corresponding time courses. Different blind source separation (BSS) techniques are used for extracting these hidden sources and time courses. In this work a differential covariance based blind source separation technique is proposed which relies on the difference of covariance of observed data and covariance of sources and mixing matrix. Performance of the proposed method is evaluated on the synthetic fMRI data. Finally comparison of ontained results is done with joint Diagonalization (JD) and Algorithm for multiple unknown source extraction (AMUSE). Comparision results shows that the proposed algoritham is better in terms of time and quality of extracted sources and time courses.