MVF-SleepNet: Multi-View Fusion Network for Sleep Stage Classification

Sleep stage classification is of great impor-tance in human health monitoring and disease diagnosing.Clinically, visual-inspected classifying sleep into differentstages is quite time consuming and highly relies on theexpertise of sleep specialists. Many automated models forsleep stage classification have been proposed in previ-ous studies but their performances still exist a gap to thereal clinical application. In this work, we propose a novelmulti-view fusion network named MVF-SleepNet based onmulti-modal physiological signals of electroencephalogra-phy (EEG), electrocardiography (ECG), electrooculography(EOG), and electromyography (EMG). To capture the rela-tionship representation among multi-modal physiologicalsignals, we construct two views of Time-frequency images(TF images) and Graph-learned graphs (GL graphs). Tolearn the spectral-temporal representation from sequen-tially timed TF images, the combination of VGG-16 and GRUnetworks is utilized. To learn the spatial-temporal represen-tation from sequentially timed GL graphs, the combinationof Chebyshev graph convolution and temporal convolu-tion networks is employed. Fusing the spectral-temporalrepresentation and spatial-temporal representation can fur-ther boost the performance of sleep stage classification. Alarge number of experiment results on the publicly availabledatasets of ISRUC-S1 and ISRUC-S3 show that the MVF-SleepNet achieves overall accuracy of 0.821,F(1)score of0.802 and Kappa of 0.768 on ISRUC-S1 dataset, and ac-curacy of 0.841,F(1)score of 0.828 and Kappa of 0.795 onISRUC-S3 dataset. The MVF-SleepNet achieves competitiveresults on both datasets of ISRUC-S1 and ISRUC-S3 forsleep stage classification compared to the state-of-the-artbaselines. The source code of MVF-SleepNet is available onGithub (https://github.com/YJPai65/MVF-SleepNet)