How to derive Granger causality measures from EEG signals?

Mattia Pagnotta,

Perceptual Network lab, Department of Psychology, University of Fribourg

Brain function arises from directed interactions among multiple distributed brain areas. Functional directed connectivity measures based on the concept of Granger causality provide promising tools to investigate such interactions and characterize how these vary over time. Granger causality measures can be in fact extended to the nonstationary case either by using adaptive multivariate autoregressive modeling (parametric methods) or by exploiting time-varying spectral factorization (nonparametric methods). Recently we systematically assessed the performance of several of these methods by using a benchmark approach. The results of our analyses showed strengths and limitations of the two families of methods, and highlighted some common pitfalls that can arise in practice when applying these methods to EEG data. Furthermore, we used EEG recordings and applied the Granger causal framework to investigate visual selective attention in healthy participants. This allowed to identify differences in the network of time-varying spectral causal influences among source-reconstructed cortical regions, as a function of whether the visual stimuli were task-relevant or task-irrelevant.