Jumping Knowledge Based Spatial-Temporal Graph Convolutional Networks for Automatic Sleep Stage Classification.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Sleep Medicine

Background:

• Accurate sleep stage classification is crucial for diagnosing sleep disorders.
• Existing methods often struggle with complex spatio-temporal patterns in multi-channel bio-signals.
• There is a need for efficient and accurate automated sleep analysis tools.

Purpose of the Study:

• To propose a novel deep learning model, the jumping knowledge spatial-temporal graph convolutional network (JK-STGCN), for enhanced sleep stage classification.
• To leverage multi-channel bio-signals (EEG, EMG, EOG, ECG) for improved classification accuracy.
• To evaluate the efficiency and generalizability of the proposed JK-STGCN model.

Main Methods:

• Feature extraction using a standard convolutional neural network (CNN) named FeatureNet.
• Adaptive learning of two adjacency matrices to capture intrinsic connections among bio-signal channels across epochs.
• Implementation of a jumping knowledge spatial-temporal graph convolution module for efficient spatial feature extraction and standard convolutions for temporal feature extraction.

Main Results:

• On the ISRUC-S3 dataset, the JK-STGCN model achieved an overall accuracy of 0.831, F1-score of 0.814, and Cohen kappa of 0.782.
• The model demonstrated high computational efficiency with a training time of 2621s for 10 subjects and a testing time of 6.8s for 50 subjects.
• On the ISRUC-S1 dataset, the model achieved accuracy, F1-score, and Cohen kappa of 0.820, 0.798, and 0.767, respectively, indicating good generalizability.

Conclusions:

• The proposed JK-STGCN model offers a novel and effective approach for automated sleep stage classification using multi-channel bio-signals.
• The model achieves competitive classification performance compared to state-of-the-art methods.
• JK-STGCN exhibits superior computational efficiency and generalizability across different datasets.