Sleep Stage Classification with CNN-Transformer-combined Structure Using Single-Channel Raw ECG.

Area of Science:

• Biomedical Engineering
• Cardiology
• Sleep Medicine

Background:

• Sleep disorders are increasingly common, necessitating accurate sleep stage classification for diagnosis.
• Current methods rely on polysomnography, which is often impractical.
• There is a growing need for single-channel sleep stage classification models.

Purpose of the Study:

• To develop and validate a user-friendly, single-channel model for automatic sleep stage classification.
• To leverage electrocardiogram (ECG) signals for sleep stage analysis.
• To improve the accessibility and practicality of sleep disorder diagnosis.

Main Methods:

• A novel model integrating convolutional neural networks and transformer structures was proposed.
• The model was designed to learn both local and global information from a single ECG channel.
• Four sleep stages were classified: wake, light sleep, deep sleep, and rapid eye movement (REM).

Main Results:

• The model achieved 76.12% accuracy on the ISRUC-1 dataset.
• The model achieved 63.42% accuracy on the SHHS-1 dataset.
• Performance surpassed existing baseline models for single-channel classification.

Conclusions:

• The proposed single-ECG channel model demonstrates significant potential for automatic sleep stage classification.
• This approach offers a practical and user-friendly method for diagnosing sleep disorders.
• Enhanced sleep stage classification precision can aid in the diagnosis of conditions like sleep apnea.