Sleep staging classification based on a new parallel fusion method of multiple sources signals

Area of Science:

• Medical Informatics
• Biomedical Engineering
• Sleep Medicine

Background:

• Automatic sleep staging is crucial for sleep studies but faces challenges like data variability and inefficiency.
• Traditional polysomnography (PSG) for sleep staging degrades sleep quality.
• Electrooculogram (EOG) and electrocardiogram (ECG) offer a less intrusive alternative for home sleep monitoring.

Purpose of the Study:

• To develop a new automatic sleep staging algorithm using EOG and ECG signals.
• To address data-variability and data-inefficiency issues in current sleep staging models.
• To provide a more comfortable and effective method for sleep monitoring.

Main Methods:

• Extracted heart rate variability (HRV) from EOG using specific algorithms.
• Derived time-domain, frequency-domain, and nonlinear-domain features from HRV and EOG segments.
• Employed a novel Parallel Fusion Method (PFM) to fuse feature sets.
• Utilized Extreme Gradient Boosting (XGBoost) for sleep stage classification.

Main Results:

• The proposed method achieved an average accuracy of 83% and a Kappa coefficient of 0.7.
• Demonstrated significant performance improvement in automatic sleep staging.
• Showed competitive performance compared to current state-of-the-art methods.
• Significantly improved the recognition rate for the S1 sleep stage.

Conclusions:

• Fusing EOG and HRV signals enhances the quality of automatic sleep staging models.
• The developed algorithm offers a beneficial approach for monitoring sleep quality and health conditions.
• Provides valuable research methods for scholars, clinicians, and individuals in sleep studies.