Stochastic Generative Models of EEG and Brain State Analysis

Electroencephalography (EEG) is the measurement of electrical activity generated by the brain from the scalp. It is known that characteristic waveform patterns appear in EEG depending on the brain state. For example, during sleep, characteristic waves such as spindles and K-complexes appear in EEG according to sleep depth. Additionally, when people with epilepsy have seizures, sharp waves such as spikes and sharp waves are observed. In this research, we are developing an approach based on stochastic generative models to accurately capture such EEG characteristics. Based on the modeling approach for EMG signals, and by considering the multidimensionality and time-series characteristics of EEG, we aim to capture the characteristic changes in EEG that change moment by moment according to brain states. By extending the model to represent not only the tail weight of the amplitude distribution but also its lateral asymmetry (skewness), features that appear in EEG according to sleep depth, such as slow waves and spindles, can be quantified as parameters of a probability distribution rather than detected directly from the waveform. The resulting features of stochastic fluctuation are applied not only to the evaluation of sleep states but also to the estimation of brain states more broadly, including the decoding of emotional valence. Main research features: - Quantification of EEG non-Gaussianity via probabilistic models that account for skewness - Analysis of stage-specific changes in sleep EEG characteristics - Application to decoding of emotional states