Spectrally weighted Granger-causal modeling: Motivation and applications to data from animal models and epileptic patients

In this paper we motivate and describe spectral weighting in methods based on the Granger-causal modeling framework. We show how these methods were validated in recordings from an animal model (rats) with relatively well-understood dynamic connectivity, and provide a comparison of their performances in terms of physiological interpretability and time resolution. Having shown that spectrally weighted Partial Directed Coherence (wPDC) shows good performances in real animal data, we provide an example of the application of this method to EEG data recorded from patients with left or right temporal lobe epilepsy. The result showed that wPDC correctly identified the major drivers of interictal epileptic spiking activity, in line with invasive validation and surgical outcome, and furthermore that right temporal lobe epilepsy is characterized by more inter-hemispheric influence than left temporal lobe epilepsy.