Evaluating a Multivariate Directional Connectivity Measure for Use in Electroencephalogram (EEG) Network Analysis Using a Conductance-Based Neuron Network Model.

摘要

Neuroscientist have shown increasing interest in understanding brain networks, and directional connectivity measures based on multivariate autoregressive (MVAR) models of signals have been proposed as a method to recover patterns of functional brain connectivity. In the past, the mathematical properties of these measures have been studied by applying them to well defined linear models or differential equations describing nonlinear oscillations. Our work tests these measures on more realistic electroencephalogram (EEG) signals to evaluate their applicability for experimental use. This report evaluates how well an effective functional connectivity measure, the directed transfer function (DTF), captures the connectivity patterns across 6 simulated 4-node networks. Each node in the network is composed of conductance-based model neurons that have been tuned to oscillate in frequency bands common in experimental EEG studies, alpha and gamma. The resulting time series from each simulation was separately modeled using multivariate autoregressive modeling, and DTF was computed from the model coefficients. Our results demonstrate that this measure can lead to spurious results in both simple feed forward networks and more complex networks with feedback loops and a ring structure; however , we propose 2 guiding principles that can increase the reliability of the results and discuss their applicability and interpretability for the future of EEG experimental analysis.