Blind Source Separation of fMRI Signals Using Joint Diagonalization Algorithm

Blind Source Separation (BSS) is a model free source separation technique which decomposes observed mixture data into mixing matrix and source matrix both of which are unknown beforehand. One well known BSS algorithm is joint diagonalization which is from the algebraic class and in which mixing structures are recovered by jointly diagonalizing the source condition matrix. In this study we first review the existing joint diagonalizing algorithm and then propose a modified high order and exponential gradient of the algorithm. The proposed algorithm is tested on simulated images and synthetic fMRI signals. Quality and execution time of the extracted sources and time courses is compared with conventional JD algorithm.