Electroencephalography microstates as biomarkers for screening Alzheimer's disease: Feasibility analysis and a machine learning classification scheme

Abstract

Background

Electroencephalography (EEG) microstate analysis has emerged as a key methodology for elucidating the brain's dynamic repertoire, providing a pivotal neurophysiological framework for the identification of cognitive impairment.

Objective

This study was aimed to analyze the EEG microstates in Alzheimer's disease (AD) based on a publicly accessible EEG dataset and additionally using support vector machine models to separate the healthy controls and AD patients.

Methods

This scalp EEG dataset from an open-source included 36 AD patients and 29 healthy controls. All EEG data underwent standardized preprocessing incorporating a 0.5–35 Hz band-pass filter and automated artifact rejection. The EEG data were subsequently partitioned into 20-s segments for microstate analysis, generating temporally aligned sequences characterized by canonical four-class spatial configurations.

Results

A total of 24 features were extracted from microstate sequences, including coverage, mean duration, occurrence, and transition probabilities between each two microstates. The statistical testing results indicated that there were significant differences in 21 features between AD patients and healthy controls. Based on the features of statistical significance, we implemented support vector machine models to distinguish the AD patients from the healthy controls, achieving an averaged classification accuracy of 75.8% in a 5-fold cross-subject validation via 10 times repeated random trials.

Conclusions

The EEG microstate analysis methods was a non-invasive, convenient, and efficient technical pathway and could be adopted for identifying AD.