White matter hyperintensity drives EEG microstate abnormalities in arteriosclerotic cerebral small vessel disease

Abstract

Objectives Although MRI is the optimal imaging method for assessing white matter hyperintensity (WMH), it lacks sensitivity to the underlying pathophysiological changes of WMH. This study aims to investigate EEG changes associated with WMH and identify potential EEG biomarkers for WMH-related dementia. Methods This cross-sectional study enrolled 90 subjects: 30 patients with WMH with dementia (WMHD), 30 patients with WMH without dementia (WMH-ND), and 30 age-and sex-matched healthy controls. Brain MRI was used to segment and quantify deep, periventricular, and total WMH volumes (DWMH, PVWMH, TWMH). Resting-state EEG data were recorded for microstate analysis. Four microstate classes (A, B, C, D) were extracted, and their occurrence, mean duration, time coverage, and transition probabilities were calculated. Partial correlation analysis was used to evaluate the associations between microstate features and WMH volume as well as MMSE scores. Based on microstate features, classification prediction models were constructed using machine learning algorithm. Results In the WMHD group, the temporal parameters (occurrence, mean duration, time coverage) of microstate C were decreased (P < 0.05–0.001), while the mean duration of microstates A and B was prolonged (P < 0.05–0.001). The WMH-ND group also showed reduced microstate C temporal parameters (P < 0.001), but exhibited compensatory increases in mean duration and time coverage of microstate D (P < 0.001). Transition probabilities toward microstate C were reduced in both patient groups (P < 0.001), and all transition probabilities toward microstate D were lower in WMHD than in WMH-ND (P < 0.05–0.001). Temporal parameters of microstates A and B and transition probabilities toward them were positively correlated with WMH volume and negatively correlated with MMSE score. Conversely, those of microstate D showed the opposite correlations. For distinguishing WMHD from WMH-ND, support vector machine (SVM) and logistic regression (LR) performed best (accuracy: 76.67%, AUC: 0.84). Classification performance was highest for WMH-ND vs. HC (LR: 79.17%, AUC 0.86), and lowest for WMHD vs. HC (SVM: 68.33%, AUC 0.77). Conclusion Patients with WMH exhibit alterations in EEG microstate features that are significantly correlated with WMH burden and cognitive function. Classification models based on these microstate features show promise for practical application in WMH-related dementia.