Sleep Stage and Sleep Disorder Detection using Spatial and Temporal Attention based Capsule Network

Sleep is referred as a cyclic process which is the most important part in human life. The sleep disorder and its stages are observed from time waveforms such as Electroencephalogram (EEG), Electrocardiograms (ECG), and Electromyograms (EMG) while diagnosing patients. However, the existing Deep Learning (DL) based neural network used for multi-modal sleep stage and sleep disorder detection has limitations in decision making that results in inaccurate detection of sleep stage. To overcome this limitation, a Spatial and Temporal Attention based Capsule Network (STA-CapsNet) is proposed for sleep stage and sleep disorder detecting for multimodal waveforms. The proposed CapsNet model with spatial and temporal attention mechanism focuses on informative regions that helps in making decision and classifies the sleep stages accurately. Firstly, the time waveforms are acquired and preprocessed to convert the signals into Red Green Blue (RGB) images for precise classification of sleep disorders. These RGB images are fed to proposed STA-CapsNet model for classifying the six stages of sleep correctly. Experimental results of the proposed STA-CapsNet method achieved accuracy of ${9 6. 8 9 \%}$ which is higher than the existing approaches such as Convolutional Neural Network (CNN) and Visual Geometry Group-16 (VGG-16).