Linearly constrained MEG beamformers for MVAR modeling of cortical interactions

Among the many methods for modeling cortical interactions using EEG and MEG data, multivariate autoregressive (MVAR) functional connectivity measures have the advantage of providing parametric directional and frequency specific information. While MVAR models have been successfully applied to depth electrode data, they are more difficult to use with external EEG and MEG data since they are not robust to the cross-talk between cortical regions that may arise because of the limited spatial resolution of EEG/MEG inverse procedures. Here we describe a modified beamforming approach for processing EEG/MEG data, designed to eliminate cross-talk between cortical regions. The output of the beamformer is then used to estimate the coefficients of an MVAR model of cortical interactions. We illustrate this method using simulated dynamic MEG data.