Abstract

Background: The diagnosis and efficacy evaluation of methamphetamine use disorder (MUD) have long relied on subjective scales, lacking objective biomarkers. Current metabolite detection methods only reflect short-term drug exposure and fail to assess chronic neural damage. EEG microstate analysis captures spatiotemporal dynamic features of resting-state brain activity, offering a novel approach for addiction identification when combined with machine learning.

Objective: To reveal EEG microstate characteristics in methamphetamine users and their association with drug craving, and to construct a machine learning-based classification model for addiction identification.

Methods: Resting-state EEG data were collected from 59 individuals with MUD and 48 healthy controls. Microstate parameters (duration, frequency, coverage, etc.) were analyzed and correlated with craving scores (Desires for Drug Questionnaire, DDQ). Machine learning models—Random Forest (RF), Support Vector Machine (SVM), Adaptive Boosting (AdaBoost), and Convolutional Neural Network (CNN)—were developed using significantly differentiated microstate features, with performance evaluated via 10-fold cross-validation.

Results:

1. Microstate Features: The MUD group exhibited significantly higher occurrence frequency, mean correlation, and global explained variance (GEV) of microstates, along with shorter durations (all p < 0.01), demonstrating a "high-frequency, short-duration" pattern. The GEV of microstate B independently predicted total craving scores (β = 0.274, p = 0.036). Microstate C frequency positively correlated with the negative reinforcement dimension, while microstate D duration showed a negative correlation.
2. Classification Models: CNN achieved the highest accuracy (92%, sensitivity 100%). AdaBoost demonstrated optimal robustness (F1-score 88%). SVM and RF underperformed due to low sensitivity in identifying the healthy group.

Conclusion: Methamphetamine users’ EEG microstate features reflect compensatory hyperactivation of the visual network and inhibition of the executive control network, validating the impaired response inhibition and salience attribution (iRISA) model. Microstate B serves as a potential biomarker for craving assessment, while the CNN-based model provides a novel tool for objective diagnosis. This study establishes an electrophysiological foundation for understanding addiction mechanisms and advancing precision interventions.