Causal information to characterize the dynamics of the EEG

ROMAN BARAVALLE; OSVALDO A. ROSSO; FERNANDO MONTANI

Conferencia; XX Conference on Nonequilibrium Statistical Mechanics and Nonlinear Physics; 2018

Electroencephalograms (EEG) reflect electrical activity in the brain, which can be considered governed by a non-linear and chaotic dynamic. In this work a new methodology ofEEG analysis will be shown, based on Information Theory tools. In particular, they are considered the EEG recordings of humans during different motor-type activities and imaginingthat perform these activities. We characterize the different regions of the cortex accordingto the different motor and imaginative activities using the ordinal patterns methodologyintroduced by Bandt and Pompe, and using different quantifiers from the Theory of Information: Shannon entropy, statistical complexity and Fisher information. In this way we candetermine the frequency bands and the most relevant regions for these tasks. We also definea measure of connectivity based on Jensen-Shannon Divergence. In addition we considerthat the problem with the real measures of entropy is that they depend on a limited numberof samples provided by the experiment. Thus, it is important to use a theoretical approachthat eliminates the bias dependent on the sample size of the entropy estimates. We take theBayesian NSB prior that allows us to generate an almost uniform distribution of entropiesto correct the bias dependent on the sample size at its source. Results obtained by applyingwavelet theory to these signals will also be briefly discussed.