EE Seminar: Underdetermined Blind Source Separation in the Wavelet Space Using Periodicity Priori for Removal of fMRI Artifacts from Simultaneous EEG-fMRI Acquisitions
Speaker: Shauli Gur Arieh
M.Sc. student under the supervision of Prof. Nathan Intrator and Dr. Raja Giryes
Wednesday, March 6th, 2019 at 15:00
Room 011, Kitot Bldg., Faculty of Engineering
Underdetermined Blind Source Separation in the Wavelet Space Using Periodicity Priori for Removal of fMRI Artifacts from Simultaneous EEG-fMRI Acquisitions
Abstract
Both EEG and fMRI are common method to measure the brain's neuron activity. EEG acquisition measures neuron cells' electrical activity using small electrodes on the scalp, thus it has high time resolution and low localization. On the contrary, fMRI acquisition uses periodically alternating magnetic fields to measure in 3d the amount of oxygen consumed by every neuron. Thus, it has low time resolution and high localization. To combine both methods' advantages, simultaneous EEG-fMRI acquisition is used both on patients and in research. To allow the simultaneous acquisition, one should remove the artifact current conducted on the EEG electrodes by the fMRI gradient magnetic field. This gradient artifact (GA) is periodic with fluctuations that have a dynamic range greater by an order of magnitude than the EEG signal.
Our methodology aims to filter out the periodical GA while minimizing damage to the EEG signal. It consists of development and analysis of a method to extract low power non-periodic signals which are contaminated by fluctuated high power periodic artifacts. The method suggests a new combination of the advantages of sparse representation in wavelet bases and two criteria based on the coefficient histogram through the periods. First is the RSD criterion, which distinguishes between the non-periodic signal and the periodic GA by the normalized standard deviation of each component. Second is the Clustering Index, which does the same distinction by the similarity of each component's histogram to normal distribution.
This method is later adapted for simultaneous EEG-fMRI signal filtering and it shows superior results over the conventional FASTR method.
