FREQUENCY-SPECIFIC NON-LINEAR GRANGER CAUSALITY IN A NETWORK OF BRAIN SIGNALS

We propose a novel algorithm to extract frequency-band specific and non-linear Granger causality (Spectral NLGC) connections between components of a multivariate time series. The advantage of our model over traditionally used VAR based models, as demonstrated in simulations, is the ability to capture complex dependence structures in a network. In addition to the simulations, the proposed method uncovered non-linear dynamics in an epileptic seizure EEG data. Spectral NLGC gives new meaningful insights into frequency specific connectivity changes at the onset of epileptic seizure. Results of both simulated and brain signals confirm the viability of the proposed algorithm as a good tool for exploration of directed connectivity in a network.