Identification of Nonstationary Brain Networks Using Time-Variant Autoregressive Models

Electroencephalographic (EEG) data provide a direct, non-invasive measurement of neural brain activity. Nevertheless, the common assumption of EEG stationarity (i.e., time-invariant process) neglects information about the underlying neural networks connectivity. We present an approach for finding networks of brain regions, which are connected by effective associations varying over time (effective connectivity). Aiming to improve the performed connectivity analysis, brain source activity is initially reconstructed from EEG recordings, applying an inverse EEG solution with enhanced spatial resolution. Further, a time-variant effective connectivity measure is used to investigate the information flow over some predefined regions of interest. For testing purposes, validation is carried out simulated and real EEG data, promoting non-stationary dynamics. The obtained results of performance prove that inherent interpretability provided by the time-variant processes can be useful to describe the underlying neural networks flow.