Tracking the poles of an AR time-variant model for EEG studies

Time-Variant Autoregressive Models (TVAM) can be used to deal with the non-stationarities of electroencephalographic (EEG) signals induced by the performance of cognitive and motor tasks. The model coefficients matrix update on a sample-by-sample basis allows evaluating changes in time of the quantitative indices that describe the directional relationships between different EEG channels. However, these indices are commonly assessed within a fixed frequency range, which does not guarantee that the total power of the brain rhythms under investigation is considered during the performance of the whole task. Here we analyzed EEG data acquired on healthy subjects performing a motor task (i.e., visually guided cue-paced pointing movement). We exploited the time-variant coefficients matrix, as obtained by a bivariate TVAM, to define a time-variant frequency band associated to the sensorimotor mu rhythm from the knowledge of the model poles position in each time instant. The directional interactions between different brain regions (i.e., premotor, motor and posterior parietal cortices) were assessed within these time-variant frequency bands. The contralateral motor cortex was the source of the information flow towards the other areas. The premotor cortex was enrolled after the motor cortex and was found to lead the posterior parietal cortex within a fronto-parietal network.