Mctsleepnet: a multiscale waveform and composite attention network with temporal dependency learning for robust...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Sleep staging is vital for evaluating sleep quality and diagnosing sleep disorders.
• Current methods face challenges in representing complex waveforms and dynamic sleep stage transitions.
• Single-channel electroencephalography (EEG) offers a non-invasive approach but requires sophisticated analysis.

Purpose of the Study:

• To introduce MCTSleepNet, a novel deep learning network for automated sleep staging using single-channel EEG.
• To address limitations in waveform representation and temporal dependency modeling in existing sleep staging techniques.
• To improve the accuracy and robustness of sleep staging, particularly for imbalanced datasets.

Main Methods:

• Developed MCTSleepNet incorporating Multiscale waveform representation (dual-scale CNN), Composite Attention, and Time dependency learning (Bi-GRU) modules.
• Utilized multiscale waveform representation to capture diverse signal patterns from EEG.
• Employed Composite Attention for enhanced feature extraction and Bi-GRU for modeling temporal dynamics.
• Introduced an adaptive cross-entropy polynomial loss function to mitigate class imbalance issues.

Main Results:

• MCTSleepNet demonstrated exceptional performance on the Sleep-EDF-20 and Sleep-EDF-78 datasets.
• The multiscale representation and attention mechanisms effectively captured prominent waveform features.
• The Bi-GRU module successfully modeled dynamic transitions between sleep stages.
• The adaptive loss function improved the model's sensitivity to minority sleep stages.

Conclusions:

• MCTSleepNet offers a powerful and effective solution for single-channel EEG-based sleep staging.
• The proposed architecture successfully addresses key challenges in waveform representation and temporal dependency.
• This work contributes to advancing automated sleep analysis and the diagnosis of sleep disorders.