Noisy network attractor models for transitions between EEG microstates
arXiv Preprint Archive March 13, 2019 Jennifer Creaser, Peter Ashwin, Claire Postlethwaite et al.
The brain's large-scale networks reorganize constantly, even at rest. EEG microstates—brief periods of stable scalp electrical activity—correspond to fMRI-defined resting-state networks and exhibit scale-free, long-range temporal correlations. This paper proposes modeling microstate sequences with nonlinear stochastic differential equations that form a noisy network attractor. A single-layer network of four nodes reproduces transition probabilities between microstates but not the heavy-tailed residence time distributions. A two-layer network with a hidden layer captures these heavy tails and long-range correlations. Fitting these models to EEG data recorded inside and outside an MRI scanner shows that separating EEG from fMRI machine noise causes a loss of information, reflected in differences in the long tail of dwell-time distributions.