Multiscale information for network characterization in epilepsy

We have developed a multiscale approach for the estimation of neuronal network coordination in the epileptic brain, from continuous (long-term) non-invasive electroencephalograms (EEG). The proposed approach specifically assesses the effect of large-scale network behavior on local network coordination, at individual dominant frequencies (modes) of the EEG spectrum. For this purpose a set of conditional information parameters is proposed to explicitly quantify the effect of global network correlation in the brain on pairwise (local) mutual information, via conditioning. These parameters are shown to be modulated in a frequency-specific manner at baseline, as well as during seizure evolution.