EEG-Based Cognitive State Classification and Analysis of Brain Dynamics Using Deep Ensemble Model and Graphical Brain Network

Estimating the cognitive state of an operator has drawn increasing attention in recent years. Cognitive behavior has been effectively studied by analyzing electroencephalogram (EEG) signals. EEG has an excellent temporal resolution but poor spatial resolution; therefore, it cannot be efficiently used for the cognitive state assessment. To obtain the spatial as well as the temporal resolution of the EEG signal, we propose to combine the filter bank common spatial pattern (FBCSP) method and long short-term memory (LSTM)-based deep ensemble model for classifying the cognitive state of a user during the mental arithmetic experiment. The proposed deep ensemble model produces 87.04% classification accuracy, and it achieves an improvement of 2.04% over the state-of-the-art method. In addition, we propose an algorithm that enables us to identify the brain dynamics for each cognitive state, which is equally important to the estimation of the cognitive state in cognitive neuroscience. For each cognitive state, the proposed information flow algorithm constructs a graphical brain network using functional and effective brain connectivity patterns, which indicates the connection strength as well as the causal effects between two anatomically separated brain regions. The proposed information flow algorithm provides an effective way of communication between different brain regions.