Fusion of Multi-Paradigm EEG Microstate Features to Enhance the Recognition of Mild Cognitive Impairment

Mild cognitive impairment (MCI) represents a transitional stage between normal aging and Alzheimer's disease. Although neuropsychological screening tools such as the Montreal Cognitive Assessment (MoCA) are widely used, they do not directly capture underlying neural dynamics. EEG microstate analysis provides a fast and noninvasive approach to characterize large-scale brain network activity. However, most previous studies have relied on single-paradigm recordings. This study proposes a unified framework integrating resting-state and task-based EEG microstate features to enhance MCI recognition.

EEG data were acquired from 63 age- and sex-matched participants, comprising 32 patients with MCI and 31 healthy controls, during both resting-state and Stroop task paradigms. Data augmentation was applied to generate 10 samples per subject, improving model robustness. We extracted 26 microstate features, including global explained variance (GEV), duration, coverage, transition probability, and Lempel-Ziv complexity. Subsequently, statistically informed minimum redundancy maximum relevance (mRMR) selection was used to determine the optimal feature subset (OFS; ≤ 5 features), followed by classification via a support vector machine (SVM) with stratified fivefold cross-validation.

Based on the OFS, resting-state and task-based models achieved accuracies of 81.8% and 88.0%, respectively. Multi-paradigm EEG fusion improved accuracy to 90.3%. MoCA alone achieved 87.6% accuracy. When MoCA was integrated with multi-paradigm EEG features under the same selection framework, accuracy further increased to 93.8%, with improved balance between sensitivity and specificity.

Integrating resting-state and task-evoked microstate dynamics enhances MCI classification beyond single-paradigm EEG. Importantly, EEG features provide complementary diagnostic information beyond conventional cognitive screening, supporting a hybrid electrophysiological-neuropsychological framework for early detection of MCI.