Linear and nonlinear features of EEG microstate associated with insomnia.

BACKGROUND

Numerous studies have revealed abnormalities in EEG microstate in insomnia, primarily quantified using linear features, whereas nonlinear metrics remain underexplored. This study aimed to compare linear and nonlinear features and further assess their contributions to insomnia classification using machine learning.

METHODS

Five minutes of resting-state EEG was recorded from 55 patients with insomnia (IN, 32 females, 20.057 ± 1.499 years) and 67 healthy controls (HC, 40 females, 20.138 ± 1.944 years). Microstate analysis was performed, and both linear (duration, occurrence, coverage, transition probability, and adjusted transition probability) and nonlinear (excess entropy, entropy rate, Lempel-Ziv complexity, and Hurst exponents) features were extracted for between-group comparisons. Additionally, these features were used to train a support vector machine to classify patients with insomnia and HC.

RESULTS

For linear features, compared with HC, patients with insomnia showed a shorter duration of microstate B, lower occurrence and coverage of microstate B and C, and higher occurrence and coverage of microstate D and E. Moreover, they exhibited abnormal microstate transition dynamics. For nonlinear features, the IN group showed a higher entropy rate and lower excess entropy in the full microstate sequences, whereas a lower entropy rate was observed in the jump sequences. When both linear and nonlinear microstate features were integrated for machine learning-based classification, the model achieved satisfactory performance, yielding an accuracy of 82.8 %.

CONCLUSIONS

In summary, both linear and nonlinear EEG microstate features reflect abnormal brain network dynamics in insomnia and may serve as reliable biomarkers for distinguishing patients from HC.