RimeSleepNet: A hybrid deep learning network for s-EEG sleep stage classification

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Neuroscience

Background:

• Sleep stage classification is crucial for sleep research and clinical diagnostics.
• Frequency aliasing in sleep electroencephalogram (s-EEG) signals presents a persistent challenge for existing methods.
• Accurate sleep staging aids in the diagnosis and management of sleep disorders.

Purpose of the Study:

• To develop a novel hybrid deep-learning model, RimeSleepNet, to address frequency aliasing in s-EEG signals.
• To enhance the accuracy and robustness of automated sleep stage classification (NREM, REM, WAKE).
• To provide an advanced tool for sleep disorder diagnosis and personalized monitoring.

Main Methods:

• Proposed RimeSleepNet model combining rime optimization algorithm with variational mode decomposition (VMD) to generate intrinsic mode functions (IMFs).
• Utilized a convolutional neural network (CNN) for feature extraction from IMFs, followed by multi-head self-attention (MHSA) for feature weighting.
• Employed long short-term memory (LSTM) networks to model temporal dynamics for final sleep stage classification.
• Evaluated model performance on the Chengdu People's Hospital and Sleep-EDF datasets.

Main Results:

• RimeSleepNet achieved high F1 scores: 0.94 (NREM), 0.89 (REM), and 0.92 (WAKE), with an AUC of 0.92.
• Outperformed baseline models including CNN and LSTM in sleep stage classification accuracy.
• Demonstrated robust generalization across datasets with a Cohen's kappa of 0.90.
• Reduced validation loss by 53% compared to LSTM, indicating improved learning efficiency.

Conclusions:

• RimeSleepNet effectively mitigates frequency aliasing in s-EEG signals, leading to superior sleep stage classification.
• The hybrid deep-learning approach offers a promising solution for automated sleep analysis in clinical settings.
• This model represents an advancement in diagnosing sleep disorders and enabling personalized sleep monitoring.