Single-channel EEG classification of sleep stages based on REM microstructure

Area of Science:

• Neuroscience
• Sleep Medicine
• Biomedical Engineering

Background:

• Rapid-Eye Movement (REM) sleep is crucial, comprising 20-25% of adult sleep.
• Sleep disorders, including REM sleep behavior disorder, are prevalent in Parkinson's disease.
• Accurate sleep stage classification is vital for identifying and monitoring parasomnias.

Purpose of the Study:

• To develop and validate a method for identifying REM sleep from single-channel electroencephalogram (EEG) data.
• To explore novel features derived from REM sleep microstructures for improved classification accuracy.
• To assess the feasibility of using this method for home sleep monitoring devices.

Main Methods:

• Utilized novel features based on REM sleep microstructures.
• Employed machine learning classifiers: Random Forest (RF), K-nearest neighbour (K-NN), and RUSBoost.
• Trained classifiers on a combination of published and novel features using single-channel EEG data.

Main Results:

• Achieved REM detection accuracy ranging from 89% to 92.7%.
• Obtained F1-scores for the REM class: 0.83 (RF), 0.80 (K-NN), and 0.70 (RUSBoost).
• Demonstrated encouraging outcomes for automatic sleep scoring and REM detection.

Conclusions:

• The proposed method shows high accuracy in REM sleep detection using single-channel EEG.
• This approach supports the development of less invasive, multi-channel home sleep monitoring systems.
• Effective REM sleep identification can improve the management of sleep disorders and related conditions.