Source to sensor coupling (SoSeC) as an effective tool to localize interacting sources from EEG and MEG data

Background:

A standard approach to estimate interacting sources from EEG or MEG data is to first calculate a coupling between all pairs of voxels on a predefined grid within the brain and then average or maximize this coupling matrix along each column or row. Depending on the chosen coupling measure and grid size this approach can be computationally very costly, in particular when a bias is supposed to be removed.

New Method:

We here suggest to replace this approach by a maximization of coupling between each source and the signal in sensor space. The idea is that any neuronal activity which can be estimated from recorded data must be present in sensor space in the first place. Using the imaginary part of coherency as coupling measure makes sure that we do not confuse this source to sensor coupling with a coupling of a source to itself. The presentation of this specific method is augmented with a discussion of conceptual issues for various forms of vector beamformers and eLoreta.

Results:

We found in simulations and empirical EEG data that the method is capable to robustly detect coupled sources.

Comparison with existing methods:

We found that the approach is hundreds of times faster than comparable conventional approaches. Results for EEG resting state data indicate that the new approach has also more statistical power than conventional approaches.

Conclusion:

The new approach is an effective tool to identify interacting sources from cross-spectra of EEG and MEG data.