Abstract
BACKGROUND: The segmentation of electroencephalography (EEG) signals into a limited number of microstates is of significant importance in the field of cognitive neuroscience. Currently, the microstate analysis algorithm based on global field power has demonstrated its efficacy in clustering resting-state EEG. The task-related EEG was extensively analyzed in the field of brain-computer interfaces (BCIs); however, its primary objective is classification rather than segmentation. METHODS: We propose an innovative algorithm for analyzing task-related EEG microstates based on spatial patterns, Riemannian distance, and a modified deep autoencoder. The objective of this algorithm is to achieve unsupervised segmentation and clustering of task-related EEG signals. RESULTS: The proposed algorithm was validated through experiments conducted on simulated EEG data and two publicly available cognitive task datasets. The evaluation results and statistical tests demonstrate its robustness and efficiency in clustering task-related EEG microstates. CONCLUSIONS: The proposed unsupervised algorithm can autonomously discretize EEG signals into a finite number of microstates, thereby facilitating investigations into the temporal structures underlying cognitive processes.