MultiChannelSleepNet: A Transformer-Based Model for Automatic Sleep Stage Classification With PSG

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Automatic sleep stage classification is crucial for sleep quality assessment and diagnosing sleep disorders.
• Current methods often rely on single-channel electroencephalogram (EEG) signals, limiting performance.
• Polysomnography (PSG) offers multi-channel data for potentially more accurate sleep staging.

Purpose of the Study:

• To introduce MultiChannelSleepNet, a novel transformer encoder-based model for automatic sleep stage classification using multichannel PSG data.
• To enhance sleep staging performance by effectively extracting and integrating information from multiple physiological signals.
• To improve the precision of sleep staging for clinical applications.

Main Methods:

• Developed a transformer encoder architecture for both single-channel feature extraction and multichannel feature fusion.
• Employed transformer encoders to process time-frequency images from individual PSG channels.
• Implemented a multichannel fusion block with additional transformer encoders to capture joint features and a residual connection to preserve channel-specific information.

Main Results:

• MultiChannelSleepNet demonstrated superior classification performance compared to existing state-of-the-art techniques on three public datasets.
• The model efficiently extracts and integrates information from multichannel PSG data.
• Achieved higher accuracy in automatic sleep stage classification.

Conclusions:

• MultiChannelSleepNet offers an efficient and effective approach for multichannel PSG-based sleep staging.
• The model's ability to integrate diverse signal information facilitates precision sleep staging in clinical settings.
• This method holds significant potential for advancing sleep medicine and diagnostics.