A multi-scale ST-GCN sleep staging method based on functional connectivity matrices

Although graph convolutional neural networks can well utilize spatial features, they neglect the capture of temporal information correlation. In this paper, a graph convolutional neural network model based on different time scales is proposed. Firstly, the standard convolutional neural network is utilized to extract features of two different time scales. Two functional connection matrix methods, namely adaptive learning and synchronous likelihood index, are employed to calculate the correlations between channels of different time scales, respectively. The spatiotemporal information is captured through the temporal attention mechanism and the spatial attention mechanism. Subsequently, the relevant spatial features are extracted through spatial graph convolution, and the temporal features are extracted through the convolutional network to learn the transition rules between sleep stages and obtain the classification results. Finally, the model is evaluated on the ISRUC-S3 sleep data in this paper. According to the experimental results, the performance of this model is comparable to that of other current models, with an accuracy of 0.847, a macro average F1 score of 0.818, and a kappa coefficient of 0.804. (Abstract)