Modeling and tracking brain nonstationarity in a sustained attention task

Abstract

In real-life situations, where humans optimize their behaviors to effectively interact with unknown and dynamic environments, their brain activities are inevitably nonstationary. Electroencephalogram (EEG), a widely used neuroimaging modality, has a high temporal resolution for characterizing the brain nonstationarity. However, quantitative measurements of EEG nonstationarity and its relations with human cognitive states and behaviors are still elusive. This study hypothesized that EEG nonstationarity could be modeled as changes of active sources decomposed by an Independent Component Analysis and proposed a model-based nonstationarity index (NSI) to quantitatively assess these changes. We tested the hypothesis and evaluated the NSI on EEG data collected from eight subjects performing a sustained attention task. Empirical results showed that values of the proposed NSI were significantly different when the subjects exhibited different levels of behavioral performance that inferred their brain states. The proposed approach is online-capable and can be used to track EEG nonstationarity in near real-time, which enables applications such as monitoring brain states during a cognitive task or predicting human behaviors in a brain-computer interface.