Slow wave dynamics of scalp EEG can be explained by simple statistical models of long-range connections

Scalp-recorded electroencephalography (EEG) is thought to be driven by both local and global oscillations dependent on the cognitive state and task of the individual. However, many EEG studies assume that the activity is local, especially when inverse modeling EEG activity. In this work, we show that a simple model of purely macroscopic connections derived from biologically plausible distributions of long-range axon delays can drive many of the traditional features of scalp-recorded EEG dynamics. All that is required is a simple linear model of time delays in a linear vector autoregressive framework with a few parameters. We make several simplifying modeling assumptions in the model: only long-range excitatory connections are included, local activity is treated as stochastic noise, and nonlinear synaptic dynamics are omitted. As a proof of concept, we restrict the model to five broad brain regions (frontal, parietal, occipital, temporal, thalamic) and model resting state EEG with no external input. We show how this simple connection model is derived from theoretical principles of synaptic activity. The model is able to replicate many features of real EEG data, including resting-state alpha power and coherence (8-13 Hz). We show that model parameters can also be informed by empirical work on structural connectivity, axon diameter estimation, and functional connectivity of fMRI BOLD measures. However, some features of the macroscopic simulations are not ideal as a model for all features of resting EEG, such as high coherence in low-frequencies in the simulation as opposed to real data. Overall, the results support the explanation of many classical EEG findings in terms of macroscopic network behavior as opposed to local activity.