A Siamese Network-Based Method for Improving the Performance of Sleep Staging with Single-Channel EEG

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Accurate sleep staging is crucial for diagnosing sleep disorders.
• Electroencephalogram (EEG) is a primary tool for monitoring brain activity during sleep.
• Existing sleep staging models face challenges with single-channel EEG data.

Purpose of the Study:

• To propose a novel method for improving sleep staging performance using single-channel EEG.
• To leverage Siamese networks and a redesigned loss function with distance metrics.
• To enhance the accuracy and robustness of automated sleep stage classification.

Main Methods:

• Developed a Siamese network architecture with two encoders to extract latent features from EEG epochs.
• Implemented a contrastive loss function, a type of distance metric, to compare EEG epoch similarities.
• Evaluated the method on two public datasets (SleepEDF and MASS-SS3) using single-channel EEG data.

Main Results:

• Achieved high performance metrics: 85.2% accuracy, 78.3% MF1, and 0.79 Cohen's kappa on SleepEDF.
• Attained 87.2% accuracy, 82.1% MF1, and 0.81 Cohen's kappa on MASS-SS3.
• Demonstrated significant improvement over state-of-the-art sleep staging methods.

Conclusions:

• The proposed Siamese network method effectively improves sleep staging performance on single-channel EEG.
• The use of distance metrics within the Siamese network framework captures relevant features for accurate sleep staging.
• This approach offers a promising advancement for automated sleep disorder diagnosis.