Automatic detection of overnight deep sleep based on heart rate variability: a preliminary study

Area of Science:

• Cardiology
• Sleep Medicine
• Biomedical Engineering

Background:

• Accurate sleep stage detection is crucial for diagnosing sleep disorders.
• Polysomnography (PSG) is the gold standard but is resource-intensive.
• Exploring non-invasive methods like heart rate variability (HRV) for sleep analysis is valuable.

Purpose of the Study:

• To investigate the efficacy of heart rate variability (HRV) for automatic deep sleep detection.
• To evaluate the impact of Z-score normalization and feature selection on classification performance.
• To develop an automated system for classifying deep sleep versus non-deep sleep stages.

Main Methods:

• Extracted 42 features from HRV data derived from 15 subjects' PSG recordings.
• Applied Z-score normalization to reduce inter-subject physiological variation.
• Utilized Correlation-based Feature Selection (CFS) and a Linear Discriminant (LD) classifier for epoch-by-epoch classification.

Main Results:

• Z-score normalization significantly enhanced classification performance.
• Achieved an overall accuracy of 81.3% for deep sleep detection.
• Obtained a Cohen's Kappa coefficient of 0.42 using leave-one-subject-out cross-validation.

Conclusions:

• HRV analysis is a promising, non-invasive method for automatic deep sleep detection.
• Normalization and feature selection are critical for improving the accuracy of HRV-based sleep classification.
• This approach offers potential for more accessible and automated sleep monitoring.