An Attention-Guided Spatiotemporal Graph Convolutional Network for Sleep Stage Classification.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Sleep staging is crucial for diagnosing sleep disorders in clinical settings.
• Graph neural networks (GNNs) show promise for automatic sleep stage classification.
• Existing GNN methods use static adjacency matrices, neglecting electrode-specific information and spatiotemporal dynamics.

Purpose of the Study:

• To address limitations in current GNN-based sleep staging methods.
• To propose a dynamic and static spatiotemporal graph convolutional network (ST-GCN) with inter-temporal attention.
• To enhance the accuracy of automatic sleep stage classification by capturing electrode-level and spatiotemporal features.

Main Methods:

• Developed a hybrid GCN-CNN model to extract spatial and temporal features from electroencephalogram (EEG) data.
• Incorporated a CNN to analyze intra-frame dependencies within brain regions.
• Utilized an attention block to capture long-range dependencies between electrodes, improving sleep stage dynamics classification.

Main Results:

• The proposed ST-GCN method demonstrated superior performance compared to existing approaches on the sleep-EDF and ISRUC-SLEEP datasets.
• Achieved accuracy improvements ranging from 4.6% to 5.3%.
• Showcased enhancements in Kappa (0.06–0.07) and macro-F score (4.9%–5.7%).

Conclusions:

• The novel ST-GCN with attention mechanisms effectively captures crucial spatiotemporal information for sleep staging.
• This method offers a significant advancement over static GNN approaches.
• The proposed technique holds potential as an effective tool for improving the diagnosis and management of sleep disorders.