Towards interpretable sleep stage classification with a multi-stream fusion network.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Sleep Medicine

Background:

• Automatic sleep stage classification is crucial for sleep quality assessment and diagnosing sleep disorders.
• Existing methods often overlook the fusion of heterogeneous spatial-temporal and spectral-temporal features from multi-channel sleep signals.

Purpose of the Study:

• To propose an interpretable multi-stream fusion network (MSF-SleepNet) for enhanced sleep stage classification.
• To address the limitations of current methods by integrating diverse feature representations.

Main Methods:

• Utilized Chebyshev graph convolution and temporal convolution for spatial-temporal feature extraction.
• Employed short-time Fourier transform and gated recurrent units for spectral-temporal feature learning.
• Integrated a contrastive learning scheme and LIME for feature enhancement and model interpretability.

Main Results:

• MSF-SleepNet demonstrated competitive performance on ISRUC-S1 and ISRUC-S3 datasets.
• The proposed method outperformed state-of-the-art approaches in most performance metrics.
• The fusion of heterogeneous features significantly improved classification accuracy.

Conclusions:

• MSF-SleepNet offers a robust and interpretable solution for automatic sleep stage classification.
• The study highlights the importance of multi-stream feature fusion for accurate sleep analysis.
• The findings contribute to advancing automated sleep disorder diagnosis and management.