Multimodal sleep staging network based on obstructive sleep apnea

Area of Science:

• Artificial Intelligence
• Biomedical Engineering
• Computational Neuroscience

Background:

• Automatic sleep staging is crucial for sleep quality assessment and diagnosing sleep disorders.
• Existing sleep staging networks often lack validation in patient populations like those with Obstructive Sleep Apnea (OSA).
• Challenges remain in fine-grained polysomnography (PSG) detection and capturing multi-scale sleep stage transitions.

Purpose of the Study:

• To develop a widely applicable deep learning network for automatic sleep stage classification.
• To address the limitations of current methods in handling OSA and multi-scale sleep transitions.
• To enhance the robustness and interpretability of sleep staging for clinical applications.

Main Methods:

• Introduced MSDC-SSNet, a deep learning network utilizing electroencephalogram (EEG) and electrooculogram (EOG) signals.
• Transformed signals into time-frequency representations for multi-scale feature extraction using an improved Transformer encoder and Multi-Scale Feature Extraction Module (MFEM).
• Integrated multi-channel data and employed multi-scale attention to capture inter-channel features and sleep stage transitions, alleviating OSA impact.

Main Results:

• MSDC-SSNet achieved 80.4% accuracy on an Obstructive Sleep Apnea (OSA) dataset.
• The network outperformed state-of-the-art methods in accuracy, F1 score, and Cohen's Kappa coefficient on three public datasets.
• Demonstrated enhanced robustness and effective integration of multimodal information.

Conclusions:

• The proposed MSDC-SSNet architecture enhances system applicability through inter-channel feature supplementation and multi-scale attention.
• The method effectively integrates multimodal information, addressing limitations of single-channel approaches.
• The approach improves interpretability for clinical applications in sleep staging.