Physiologically-based Brain State Modeling

Abstract

The brain is the most complex organ in the body, and exhibits rich multiscale dynamics. Changes in arousal state (such as falling asleep) are accompanied by large scale changes in whole-brain dynamics that are readily observable by a range of methods including electroencephalography (EEG). However, most analysis of arousal state focuses either on coarse quantitative measures like the relative amount of power in certain frequency bands, or by assigning the state into one of several classes based largely on qualitative observations. Both of these approaches are ultimately phenomenological, and provide limited insight into the physiological mechanisms that give rise to the EEG. Predicting EEG using modeling approaches has been highly successful at relating large scale brain physiology to experimental observations. In particular, physiologically based modeling addresses significant issues that commonly arise in high-dimensional models, by constraining each parameter on the basis of experimental data, and by providing a physiologically meaningful interpretation of all model parameters. One class of brain models is based on neural field theory, which averages the properties of neurons over short temporal and spatial scales to form continuous fields that represent neural activity. These models are ideally suited to EEG comparison and analysis because the EEG reflects the combined activity of millions of individual neurons. This thesis uses an established neural field model of the brain to investigate the physiological basis for changes in the EEG over a range of brain states. In particular, the model is used to investigate sleep-wake dynamics, and newly discovered nonlinear dynamics occurring in normal brain states. In this thesis, classical sleep stages are associated with model parameters by characterizing each of them in terms of EEG power spectral features. The match between theory and experiment is quantified in terms of these spectral features. Linear interpolation of parameter values between pairs of states is used to predict changes in the EEG accompanying brain state changes. In particular, continuous parameter trajectories in the model are associated with classical, discrete sleep stage transitions. The model is fitted to EEG data from a healthy subject, and this experimentally observed trajectory is compared with predicted trajectories and with the classified regions of the model parameter space. The continuous experimental trajectory is consistent with predictions, and is able to represent the individual physiology corresponding to the experimental observations, unlike classical staging. This thesis also presents a detailed theoretical examination of nonlinear effects occurring in conjunction with sleep spindles, which are transient bursts of electrical activity during sleep that originate in the thalamus. Nonlinear effects are commonly observed in the brain during seizures, but observing nonlinearity in normal brain states is more challenging. Our neural field model predicts strong nonlinear effects during sleep spindles, that give rise to an experimentally observable harmonic in the EEG power spectrum. The properties of this harmonic are analyzed in the model, and the power in the spindle harmonic is predicted to scale approximately quadratically with the power in the fundamental spindle oscillation, once background sleep EEG activity is accounted for. The sleep spindle harmonic is readily observable in experimental data by using novel signal processing developed in this thesis to minimize the background sleep signals that otherwise obscure the spindle harmonic. A spindle harmonic at twice the fundamental spindle frequency was visible in all nine subjects studied. The primary spindle oscillation has a different amplitude in each subject, which enables the power scaling of the spindle harmonic to be estimated. The power scaling is found to be consistent with the nonlinear scaling prediction made by the model.