Applying multivariate empirical mode decomposition to the analysis of broad-band EEG microstates

EEG microstates models spontaneous resting-state EEG as continuous transitions among a few quasi-stable scalp topographies that remain unchanged for 60-120ms. The microstates are extracted from band-passed EEG signals of 2-20Hz or 1-40Hz. Microstates are typically described as broad-band phenomena. A single microstate could model temporal dynamics of a broad range of time scales. The present study investigates a novel method of microstates extraction to examine the broad-band perspective. The data-driven noise-assisted multivariate empirical mode decomposition (NA-MEMD) was applied to decompose time-domain EEG into a set of intrinsic mode functions (IMFs). Each IMF carries information of the original signal at different time scales (~2-150Hz). IMFs can be combined to reconstruct the original signal. EEG microstates were extracted from healthy young (age 20.7 (1.56), n=22) and older adults (age 72.3 (3.34), n=24) utilizing 2-20Hz band-passed signals or reconstructed signals from different IMFs combinations. The proposed approach could recover the four traditional microstate classes from both subject groups while the existing method failed in the elderly group, recovering only two of four classes. Microstates extracted from IMFs of frequency range (~2-15Hz) explained 54% and 59% of total variances of young and old group respectively, which are higher than using existing method (53% and 56%). It is found that microstate classes A and B were more consistent across frequency ranges, while classes C and D were more frequency-specific. The proposed approach provides new insights on the frequency composition of EEG microstates.