Multi-View Spatial-Temporal Graph Convolutional Networks With Domain Generalization for Sleep Stage Classification.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Sleep stage classification is crucial for diagnosing sleep disorders and assessing sleep quality.
• Existing methods struggle with utilizing complex spatial-temporal brain signal features and generalizing across individuals.
• Deep learning models often lack interpretability in the context of brain activity.

Purpose of the Study:

• To develop an advanced deep learning framework for robust sleep stage classification.
• To enhance the utilization of multi-channel brain signals by integrating spatial and temporal information.
• To improve model generalization across different subjects and provide interpretable insights.

Main Methods:

• Proposed a multi-view spatial-temporal graph convolutional network (MSTGCN) integrating functional connectivity and physical proximity brain graphs.
• Employed graph convolutions for spatial feature extraction and temporal convolutions for sleep stage transitions.
• Integrated attention mechanisms and domain generalization for subject-invariant feature extraction.

Main Results:

• The MSTGCN model demonstrated superior performance compared to state-of-the-art methods on public sleep datasets.
• Effectively captured time-varying spatial-temporal dynamics and topological information from brain signals.
• Achieved improved generalization across subjects, addressing individual biological signal variability.

Conclusions:

• The proposed MSTGCN framework offers a significant advancement in sleep stage classification accuracy and generalization.
• The multi-view graph approach effectively leverages complex brain signal features.
• This method holds promise for more reliable sleep assessment and personalized diagnostics.