Single-channel EEG sleep stage classification based on a streamlined set of statistical features in wavelet domain

Area of Science:

• Biomedical Engineering
• Signal Processing
• Sleep Medicine

Background:

• Single-channel electroencephalograms (EEGs) are crucial for developing accessible sleep monitoring technologies.
• Accurate sleep stage classification is essential for diagnosing sleep disorders and advancing telemedicine.
• Existing methods often require multi-channel EEGs, limiting their application in home-based or portable devices.

Purpose of the Study:

• To enhance sleep stage classification performance using single-channel electroencephalograms (EEGs).
• To develop a robust feature extraction method applicable to various sleep scoring systems.
• To validate the proposed method's efficacy and stability using a large public dataset.

Main Methods:

• Decomposition of single-channel EEGs using discrete wavelet transform.
• Computation of statistical moments (kurtosis, skewness, variance) from wavelet coefficients.
• Classification of sleep epochs using a random forest predictor trained on extracted features.

Main Results:

• The proposed method achieved high accuracy (over 90%) and kappa coefficients (over 0.8) for 2- to 6-state sleep classifications.
• Feature extraction using wavelet coefficients demonstrated superior performance compared to time-domain analysis.
• ReliefF tests confirmed the stability and non-redundancy of the selected statistical features.

Conclusions:

• The developed method offers a robust and accurate approach for sleep stage classification from single-channel EEGs.
• The feature set is stable across different sleep scoring systems, making it versatile for various applications.
• This technique holds significant potential for improving sleep monitoring in telemedicine and home care settings.