Single-channel EEG sleep stage classification using synchrosqueezed transform and sequential representation learning

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Accurate sleep stage classification is crucial for sleep research and clinical diagnosis.
• Challenges in sleep stage classification stem from the complexity and feature overlap in electroencephalogram (EEG) signals.
• Existing methods often struggle with the intricate nature of EEG data.

Purpose of the Study:

• To develop a novel, three-stage approach for accurate sleep stage classification using single-channel EEG.
• To leverage SynchroSqueezed Transform (SST) for time-frequency representation (TFR) of EEG signals.
• To enhance feature extraction and classification performance for sleep stages.

Main Methods:

• Utilized the Sleep-EDF dataset, focusing on the Fpz-Cz EEG channel.
• Applied SynchroSqueezed Transform (SST) to generate time-frequency representations (TFRs).
• Employed a three-stage deep learning model: CNN pretraining, contrastive learning for encoder refinement, and stacked Gated Recurrent Units (GRUs) for classification.

Main Results:

• The proposed three-stage model achieved an accuracy of 83% in sleep stage classification.
• Contrastive learning effectively improved feature separability between different sleep stages.
• The combination of CNN, contrastive learning, and GRUs demonstrated robust feature extraction and dependency capture.

Conclusions:

• The novel three-stage deep learning approach significantly enhances sleep stage classification accuracy.
• The method effectively handles complex single-channel EEG signals by employing a multi-stage feature extraction and classification strategy.
• This technique shows promise for improving the efficiency and reliability of sleep analysis in both research and clinical settings.