A Noise-Assisted Data Analysis Method for Automatic EOG-Based Sleep Stage Classification Using Ensemble Learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Sleep staging research traditionally relies on multiple recording modalities.
• Reducing the number of sensors can improve patient comfort and simplify data acquisition.
• Conventional sleep staging methods require complex polysomnography setups.

Purpose of the Study:

• To investigate the feasibility of automatic sleep staging using only electrooculography (EOG) signals.
• To develop and validate a novel signal processing and machine learning approach for sleep staging.
• To assess if reduced modalities can yield results comparable to conventional systems.

Main Methods:

• Utilized electrooculography (EOG) signals as the sole data source.
• Applied Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) for signal processing.
• Employed a Random Forest classifier for automatic sleep stage classification.

Main Results:

• Achieved an overall accuracy of 82% for automatic sleep staging.
• Obtained a Cohen's kappa coefficient of 0.74, indicating substantial agreement with manual scoring.
• Demonstrated the effectiveness of EOG-based analysis for sleep staging.

Conclusions:

• Automatic sleep staging using EOG signals is a viable alternative to conventional multi-modal systems.
• The proposed CEEMDAN and Random Forest method offers a promising approach for simplified sleep research.
• Further research can explore optimizing this method for clinical applications.