Dynamics of parameters of neurophysiological models from phenomenological EEG modeling E. Olbrich a,∗ a Max Planck Institute for Mathematics in the Sciences, Inselstr. 22, D 04103 Leipzig, Germany T. Wennekers b b Centre for Theoretical and Computational Neuroscience, University of Plymouth, PL4 8AA Plymouth, United Kingdom Abstract 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 (< 1 Hz) 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. Key words: EEG, oscillations, data analysis, sleep, slow oscillation, thalamocortical network 1 Introduction The EEG was traditionally analysed visually by looking for characteristic patterns such as alpha waves, sleep spindles, slow waves in sleep, spikes and spike waves in epilepsy, and other specific patterns in pathological conditions. Today computer-based spectral analysis is still the main method of analysis using the spectral power of activity in ∗ Corresponding author. Tel.: ++49 341 9959 568 Email address: olbrich@mis.mpg.de (E. Olbrich). Preprint submitted to Neurocomputing 28 September 2006