Deep Learning Application to Clinical Decision Support System in Sleep Stage Classification.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroscience
• Sleep Medicine

Background:

• Deep learning shows promise for automated sleep stage classification.
• Current polysomnography (PSG) methods require multiple channels and are inconvenient for patients.
• Challenges hinder the routine clinical application of automatic sleep scoring algorithms.

Purpose of the Study:

• To develop a deep learning model for sleep stage classification using single-channel EEG data.
• To create a model suitable for clinical decision support systems (CDSSs).
• To overcome the inconvenience of multi-channel PSG recordings.

Main Methods:

• A deep learning model combining convolutional neural networks and a transformer was developed.
• The model was trained for supervised learning of three sleep stages using single-channel EEG (C4-M1).
• Data from 1590 (training), 341 (validation), and 343 (test) PSG recordings were utilized.

Main Results:

• The developed model achieved an overall accuracy of 91.4%, comparable to human experts.
• Accuracy varied by obstructive sleep apnea severity: 94.3% (normal), 91.9% (mild), 91.9% (moderate), and 90.6% (severe).
• The model demonstrated accurate and rapid delineation of three-class sleep staging.

Conclusions:

• A deep learning model using single-channel EEG can accurately classify sleep stages.
• This approach offers a convenient alternative to multi-channel PSG.
• The model has potential as a clinical decision support system (CDSS) for real-world applications.