The addition of entropy-based regularity parameters improves sleep stage classification based on heart rate...

Area of Science:

• Cardiology
• Sleep Medicine
• Biomedical Engineering

Background:

• Automatic sleep staging is crucial for diagnosing sleep disorders.
• Heart rate variability (HRV) analysis offers a non-invasive method for physiological monitoring.
• Existing HRV-based sleep staging methods require further refinement for accuracy.

Purpose of the Study:

• To develop and evaluate an automatic sleep stage classification system using HRV analysis.
• To investigate the efficacy of novel regularity features derived from RR series.
• To compare classification performance for wakefulness versus sleep and REM versus NREM sleep.

Main Methods:

• Utilized 20 automatically annotated polysomnographic recordings.
• Employed artificial neural networks for classification tasks.
• Extracted features from RR series, including novel regularity parameters, using varying epoch lengths (2, 6, 10 epochs).

Main Results:

• Two feature sets, each with four features including a novel regularity parameter, captured 99% of data variance.
• Higher classification accuracy was achieved for REM versus NREM sleep (up to 83.8%) compared to WAKE versus SLEEP (up to 71.3%).
• Reliability (Cohen's Kappa) was higher for REM vs. NREM (0.68) than WAKE vs. SLEEP (0.45).

Conclusions:

• HRV-only sleep classification, incorporating novel regularity features, demonstrates sufficient precision for many applications.
• While less precise than multi-signal polysomnography, this method offers a cost-effective and unobtrusive alternative.
• The proposed approach advances automated sleep analysis using physiological signals.