Discrimination of Wakefulness From Sleep Sage I Using Nonlinear Features of a Single Frontal EEG Channel

Objective: Although a wide range of researches have shown the utility of electroencephalography (EEG) for the sleep monitoring, the majority of them reported a low sensitivity for classification of wakefulness from sleep stage I. This paper, therefore, validates a reliable nonlinear feature set for discriminating the wakefulness from sleep stage I using a single frontal EEG channel. Methods: Effectiveness of the proposed feature set was evaluated using four public databases namely Sleep Telemetry, DREAMS, MESA, and DCSM. After splitting the EEG signal into its sub-bands using discrete wavelet transform, Katz and Higuchi's fractal dimensions, dispersion entropy, and bubble entropy were computed as the features. Then, 70% of the samples was randomly fed to the support vector machine, linear discriminant analysis, and k-nearest neighbors classifiers with 10-fold cross-validation for the training, whereas the rest was used for the unseen data testing. We also compared performance of the proposed feature set against the relative band power (RBP) analysis. Results: While the best classification results for both feature sets were achieved by the support vector machine, the proposed outperformed the RBP with the higher mean of sensitivity to sleep stage I for the Sleep Telemetry (82.6% vs. 71.8%, p <0.05), DREAMS (87.6% vs. 71.8%, p <0.05), DCSM (91.0% vs. 74.2%, p <0.05), and MESA (82.0% vs. 76.1%, p <0.05) databases. Significance: Considering interchangeability of the proposed feature set for discriminating the wakefulness from sleep stage I, it has the potential to be used for estimating the sleep onset latency in clinical applications.

Automated summary

This study validates a reliable nonlinear feature set for discriminating wakefulness from sleep stage I. Experimental results on four public databases show that the proposed feature set outperforms relative band power analysis in distinguishing sleep stage I.