Performance of beamformers on EEG source reconstruction

Recently a number of new beamformers have been introduced for reconstruction and localization of neural sources from EEG and MEG. However, little is known about the relative performance of these beamformers. In this study, 8 scalar beamformers were examined with respect to several parameters to determine how effective they are at reconstruction of a dipole time course from EEG. A simulated EEG signal was produced by means of forward head modelling for projection of an artificial dipole on scalp electrodes then superimposed on background signal. Both real EEG and white noise were applied as background activity. Although the eigenspace beamformer can perform slightly better than other beamformers for small dipoles, and even more so for large dipoles, it is not a contender for real-time beamforming of EEG as it cannot be completely automated. Overall, in terms of performance, robustness to variations in parameters, and ease of application, the minimum variance and Borgiotti-Kaplan beamformers were found to be the best performers.