Optimal Electroencephalogram and Electrooculogram Signal Combination for Deep Learning-Based Sleep Staging.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Sleep Medicine

Background:

• Manual sleep staging via polysomnography (PSG) is time-consuming and prone to errors.
• Deep learning models show promise for automated sleep staging using fewer signals.
• Optimal signal combinations for accurate automated sleep staging require further investigation.

Purpose of the Study:

• To identify optimal electroencephalogram (EEG) and electrooculogram (EOG) signal combinations for deep learning-based automatic sleep staging.
• To test the hypothesis that various EEG signal combinations yield comparable performances for automated sleep staging.
• To determine if simplified measurement setups can achieve accurate automatic sleep staging.

Main Methods:

• Utilized four EEG signals and one EOG signal from 876 subjects with suspected obstructive sleep apnea.
• Trained 31 deep learning models using diverse combinations of EEG and EOG signals.
• Evaluated classification performance for automatic sleep staging against manual scoring across five sleep stages.

Main Results:

• Classification performance differences among EEG signal combinations were minimal, with accuracies between 81% and 85%.
• Incorporating EOG signals improved accuracy by 1-2% in single EEG configurations.
• Minimal accuracy gains were observed when combining EOG with multiple EEG signals.

Conclusions:

• Automated sleep staging can be achieved using simplified EEG and EOG measurement setups.
• Signal combination choices do not significantly compromise classification performance.
• This finding supports the development of more efficient sleep staging systems.