Single-Channel Ecg-Based Sleep Stage Classification With End-To-End Trainable Deep Neural Networks

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Sleep Medicine

Background:

• Increasing prevalence of sleep disorders necessitates accessible diagnostic tools.
• Electrocardiograms (ECGs) are easily obtainable physiological signals.
• Existing sleep classification methods often require complex multi-channel recordings or manual feature engineering.

Purpose of the Study:

• To investigate the efficacy of single-channel ECG signals for automatic sleep stage classification.
• To develop a fully automated system leveraging deep learning for sleep analysis.
• To move beyond traditional manual feature extraction in ECG-based sleep studies.

Main Methods:

• Utilized a ContextNet-based neural network for feature extraction from ECG spectrograms.
• Employed a Transformer model to capture temporal dynamics of sleep cycles.
• Developed a fully neural network-based approach, eliminating manual feature engineering.

Main Results:

• Demonstrated the feasibility of using single-channel ECG for sleep stage classification.
• The proposed model effectively captures temporal patterns crucial for accurate sleep staging.
• Achieved promising results in automatic sleep classification using deep learning on ECG data.

Conclusions:

• Single-channel ECG signals hold significant potential for user-friendly automatic sleep stage classification.
• Fully neural network-based feature extraction offers a robust alternative to manual methods.
• This approach could enhance the accessibility and efficiency of sleep disorder diagnosis.