Microstate-Specific Cross-Frequency Coupling Networks: A Spatiotemporal Framework for MCI Identification

This study integrated EEG microstates with cross-frequency coupling networks to develop a novel spatiotemporal framework for the early identification of mild cognitive impairment (MCI).

Resting-state EEG was recorded from 22 patients with MCI and 22 age-matched healthy controls. Topographic clustering was used to derive four canonical microstates (A-D). Source reconstruction was then performed to extract time series from seven resting-state networks. For each microstate, a microstate-specific cross-frequency coupling network was constructed by computing n:m phase synchronization across the delta, theta, alpha, and beta bands among these networks. Classification was performed using Maximum Relevance Minimum Redundancy (mRMR) feature selection and a nested cross-validated CatBoost classifier.

Significant disruptions in cross-frequency phase synchronization were observed within microstates A, C, and D among patients with MCI. The classification model demonstrated its highest performance within microstate D, achieving a balanced accuracy of 84.4%. The features most effective in discriminating between groups primarily involved the limbic network in the theta band, the frontoparietal and dorsal attention networks in the alpha band, and the sensorimotor network in the delta band.

MCI is associated with microstate-dependent impairments in cross-frequency synchronization, which suggests compromised spatiotemporal integration across large-scale brain networks. The microstate-specific cross-frequency coupling network framework proposed offers a novel and physiologically interpretable EEG biomarker for the early diagnosis of MCI.