Automatic sleep stages classification using respiratory, heart rate and movement signals

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Machine Learning

Background:

• Sleep stage identification is crucial for diagnosing sleep disorders.
• Traditional polysomnography is invasive and not suitable for long-term home monitoring.
• Bio-vital signals correlate strongly with sleep stages.

Purpose of the Study:

• To develop a non-invasive algorithm for sleep stage identification.
• To enable long-term sleep monitoring in a home environment.
• To support sleep experts in analyzing sleep data.

Main Methods:

• Utilized multinomial logistic regression for sleep stage classification.
• Employed derived parameters from respiratory, heart rate, and movement signals as input.
• Trained a machine learning model on sleep recordings from 35 individuals (5 for training, 30 for evaluation).

Main Results:

• Achieved 72% accuracy for Wake, NREM, REM classification (Cohen's kappa = 0.67).
• Achieved 58% accuracy for Wake, Light, Deep, REM classification (Cohen's kappa = 0.50).
• Demonstrated the potential of non-invasive signals for sleep studies.

Conclusions:

• Respiratory, heart rate, and movement signals can be used for non-invasive sleep studies with reasonable accuracy.
• The developed algorithm is suitable for long-term home monitoring systems.
• The algorithm's adaptability allows for use with various hardware systems.