Brain state quantification using dynamical nonstationarity analysis of EEG

Abstract

The Brain is composed in average of $10^{11}$ neurons each connected via synapses to about 7,000 to 10,000 other neurons. The characteristics of neurons differs depending on their morphology and function, however, they all integrate in nonlinear fashion temporal and spatial information from their afferent connections. In consequence, the Brain is an incredibly complex machinery in which spatio-temporal patterns of information processing seem to play a key-role for the understanding of cognitive and functional integration of information. The electroencephalograms (EEGs) are a high-temporal resolution, non-expensive, non-invasive and popularly used method to record and investigate the electrical activity of the brain during different cognitive states. This recording method has allowed to partially understand and decipher (electrical) patterns associated with information processing in the Brain using linear (e.g., Fourier transform) and nonlinear approaches (e.g., entropy, phase space-based approached). More recently, the intrinsic nonstationarity of EEG (i.e. change over time of parameters defining the underlying generating system), long considered a limitation to the application of conventional methods, has been exploited using nonlinear approaches to improve our understanding of Brain waves’ patterns. The temporally changing (nonlinear) dynamics of the brain’s activity (i.e. nonstationary dynamics) has demonstrated its usefulness to unravel the organization of spatially and temporally multi-scale information processing. In this thesis, we propose a nonlinear approach named dynamical nonstationarity analysis (DNA) that analyses changes in dynamical properties of EEGs. This method provides a novel quantification method of macrostates from microstates patterns, which is a possible linkage between different temporal-scales of dynamics. We demonstrate, with results among the best in the field, the applicability and usefulness of this approach on simulated E...