Differentiating propofol-induced altered states of consciousness using features of EEG microstates

Abstract Objective In this study, we comprehensively studied alternations of EEG microstates in resting state, light, and deep level of propofol-induced unconsciousness. We further investigated the feasibility of using microstate features for differentiating different consciousness levels.

Methods: 60-channel EEG was recorded from 31 male subjects in resting, light, and deep anesthesia states. Microstate analysis was performed on each consciousness levels separately. We first identified the optimal number of microstate templates for each consciousness levels and aligned them based on spatial similarity. Then, we extracted features including duration, coverage, and occurrence. Machine learning models including SVM, LDA and Random Forest were employed to classify different consciousness levels using microstate features. Finally, the cortical source of each microstates was analyzed using standardized low-resolution electromagnetic tomography.

Results: We found that the optimal number of microstates differed in different consciousness levels. 7 common and 2 anesthesia-specific microstates were identified. M4, M5, M1 and M2 were similar to the canonical microstates A to D. Occurrence of M6 and duration of M1, M5 and M7 monotonically decreased with increasing level of consciousness suppression. All three features of the anesthesia-specific microstate M8 significantly decreased with increasing level of anesthesia. The classifiers using microstate features showed a mean accuracy of 85.6% to classify three consciousness levels.

Conclusion: EEG microstate features significantly altered during propofol-induced unconsciousness, and can be used to distinguish between consciousness levels.

Significance: Microstate analysis may be a useful tool in deciphering the neural mechanism of propofol-induced anesthesia and provide useful features for quantifying levels of consciousness.