Fast Sleep Stage Classification Using Cascaded Support Vector Machines with Single-Channel EEG Signals

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Long-term sleep stage monitoring is crucial for diagnosing and treating insomnia.
• Wearable electroencephalogram (EEG) devices enable continuous, practical sleep analysis.

Purpose of the Study:

• To develop a fast and accurate sleep stage classification method using single-channel EEG signals.
• To improve the accuracy and generalization of sleep stage classification for practical applications.

Main Methods:

• Applied wavelet threshold denoising (WTD) and wavelet packet transformation (WPT) for EEG signal preprocessing.
• Extracted time, frequency, and nonlinear dynamics features from denoised EEG signals.
• Utilized cascaded Support Vector Machine (SVM) models for sleep stage classification.

Main Results:

• Achieved an average classification accuracy of 88.11% with the inclusion of nonlinear dynamics features.
• Enhanced the classification accuracy for non-rapid eye movement sleep stage 1 (N1) from 41.5% to 55.65% using cascaded SVM models.
• Maintained an overall classification time of less than 1.7 seconds per epoch.

Conclusions:

• The proposed cascaded SVM method with comprehensive features is effective for accurate, long-term sleep stage monitoring.
• This approach demonstrates potential for practical applications using wearable EEG devices.
• The inclusion of nonlinear dynamics features significantly improves classification performance, especially for subtle sleep stages.