Deep residual networks for automatic sleep stage classification of raw polysomnographic waveforms

Area of Science:

• Artificial Intelligence
• Neuroscience
• Biomedical Engineering

Background:

• Accurate sleep stage classification is crucial for diagnosing sleep disorders.
• Current methods often rely on manual scoring or complex feature engineering.
• Deep learning offers a promising avenue for automated analysis of physiological signals.

Purpose of the Study:

• To develop and evaluate an automated sleep stage classification algorithm.
• To utilize deep residual neural networks with raw polysomnogram signals.
• To compare the performance of different model configurations.

Main Methods:

• A deep residual neural network architecture with 50 convolutional layers was employed.
• The algorithm processed raw polysomnogram signals for sleep stage classification.
• Three model configurations were trained on 1850 recordings and tested on 230 independent recordings.

Main Results:

• The best performing model achieved an accuracy of 84.1% and a Cohen's kappa of 0.746.
• This performance surpasses previous results using only raw polysomnogram data.
• Classification errors were most frequent between non-REM stage 1 and stage 3.

Conclusions:

• The developed deep learning algorithm shows high performance for automatic sleep stage classification.
• Further validation on independent clinical cohorts is recommended.
• The algorithm has potential for improving sleep disorder diagnosis and research.