PD Dr Thomas Knösche | Estimation of brain connectivity from EEG and MEG

Causal relationships between different neural populations or brain areas are reflected by correlative relationships between their average neural activity characterized by membrane potentials or firing rates (functional connectivity; FC). EEG and MEG have the advantage to relatively directly reflect such quantities and are therefore good candidates for the non-invasive estimation of FC. However, this comes with a number of complications, in particular the signal mixture due to volume conduction, which can be only limitedly disentangled due to the non-uniqueness of the inverse problem and the approximate description of the volume conduction by head models.

After a general discussion of the problem, I will first address the influence of simplified volume conduction models onto the accuracy of FC estimates and propose ways to assess the confidence we may have in EEG/MEG based FC estimates [1]. Next I will turn to inverse procedures that, rather than first estimating the source time courses and then quantifying the connectivity based thereupon, jointly estimate the sources and their connectivity from the data, thereby stabilizing the inverse procedure. This can be achieved by projecting a multivariate autoregressive (MAR) model into the source space using a beamformer approach [2]. Moreover, I will present an approach that simultaneously estimates source amplitudes and interactions with a variational Bayesian learning algorithm that uses prior knowledge from diffusion tractography and functional MRI [3].

References

[1] Cho, Vorwerk, Wolters, Knösche: Influence of the head model on EEG and MEG source connectivity analysis. NeuroImage 110, 60-77 (2015)

[2] Cho, Wolters, Knösche: Source connectivity analysis using the multivariate autoregressive model of MEG sensor data with the beamformer for crosstalk suppression. In preparation.

[3] Fukushima, Yamashita, Knösche, Sato: MEG source reconstruction based on identification of directed source interactions on whole-brain anatomical networks. NeuroImage 105, 408–427 (2015).