Dynamics of parameters of neurophysiological models from phenomenological EEG modeling

We investigate a recently proposed method for the analysis of oscillatory patterns in EEG data, with respect to its capacity of further quantifying processes on slower (<1Hz) time scales. The method is based on modeling the EEG time series by linear autoregressive (AR) models with time dependent parameters. Systems described by such linear models can be interpreted as a set of coupled stochastically driven oscillators with time dependent frequencies and damping coefficients. It is an open question to which extent the estimated frequencies and dampings correspond to true properties of oscillatory eigenmodes in the underlying networks. The present study investigates this relationship using simple neural network models to generate artificial data with controllable properties. We demonstrate that the method detects changes of the eigenmodes induced by slow parameter changes of the network very well.