EEG microstate dynamics and metal metabolism in Parkinson's disease dementia: An exploratory study with machine learning-assisted sensitivity analysis.

BACKGROUND

Electroencephalogram (EEG) microstates effectively characterise cognitive-related brain networks, and metal homeostasis is crucial for maintaining cognitive function. However, the relationship between these factors in Parkinson's disease (PD) with dementia (PDD) remains unclear. We investigated this association and evaluated the performance of these combined features in classifying PDD.

METHODS

We analysed linear and nonlinear features of EEG microstates in 60 patients with PDD, 56 patients with PD without dementia (PDND), and 60 healthy controls (HC). We measured indicators of metal metabolism in patients with PD and examined correlations between their microstate characteristics, clinical scale scores, and metal metabolism levels. We trained a support vector machine (SVM) with these integrated features to differentiate PDD from PDND.

RESULTS

(1) Compared with the PDND group, the PDD group showed increased occurrence and coverage of microstate A, prolonged duration of microstates B and C, increased occurrence of microstate C, but reduced occurrence and coverage of microstate B. Compared with HC, the PDND group showed decreased duration and occurrence of microstate C, along with prolonged duration of microstate B. Both PD subgroups showed lower nonlinear dynamic metrics than HC. (2) The PDD group had lower serum transferrin (STf) than PDND. Lower STf was associated with longer duration of microstate C and higher TP from microstate D to A. (3) Although microstate parameters achieved the highest area under the curve (AUC), the multimodal ensemble provided superior clinical utility for identifying PDD by maximizing recall and minimizing false negatives.

CONCLUSIONS

Altered STf levels in PDD suggest an association between metal metabolism and brain network dysfunction relevant to pathophysiology. Integrating EEG microstates with metal metabolic indicators showed potential to enhance diagnostic sensitivity, warranting further investigation as a candidate approach for PDD identification.