Simplifying Multimodal With Single EOG Modality for Automatic Sleep Staging.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Sleep Medicine

Background:

• Polysomnography (PSG) is standard for sleep staging, utilizing signals like electroencephalography (EEG) and electrooculography (EOG).
• Combining EEG and EOG enhances sleep staging accuracy, but EEG acquisition is complex and sensitive to noise.
• Electrooculography (EOG) offers a simpler, more robust alternative for clinical sleep analysis.

Purpose of the Study:

• To develop a novel framework for simplified multimodal sleep staging using only EOG signals.
• To leverage the correlation between EEG and EOG to generate effective EOG-only features.
• To achieve high sleep staging performance with EOG, even without EEG data.

Main Methods:

• A novel framework was developed to model the correlation between EEG and EOG signals.
• Generative adversarial learning was employed with time and frequency-guided generators to create multimodal features from EOG.
• The framework was evaluated on a real-world dataset (67 recordings) and four public datasets.

Main Results:

• The proposed framework achieved the best performance using only the EOG modality compared to existing methods.
• Sleep staging performance using EOG within this framework was comparable to using EEG.
• The method effectively simplifies multimodal sleep staging by relying solely on EOG.

Conclusions:

• The novel framework successfully simplifies sleep staging by utilizing EOG signals effectively.
• This approach offers a practical alternative for clinical sleep analysis, overcoming EEG acquisition challenges.
• EOG-based sleep staging can achieve performance comparable to EEG, enhancing accessibility and reducing complexity.