Paper Title
Classification of REM and NREM Sleep Stages with/without Sleep Disorder Breathing using Deep SDBnet
Abstract
The diagnosis of sleep-related disorders, such as Sleep Disorder Breathing (SDB) disease, highly depends on the
classification of sleep stages. Previous studies focus on sleep stages classification as the first step in detecting sleep-related
diseases. However, classification of sleep stages segments with or without a specific sleep-related disease is challenging. In
this study, we focus on the classification of REM and NREM sleep stages with or without SDB events. We propose in this
paper a novel deep learning architecture, named Deep SDBNet which comprises three deep learning networks based on
Convolutional Neuron Networks (CNN). All three deep learning networks are trained separately. The first two deep learning
networks extract important features from EEG, EMG and EOG signals while the third deep learning network extracts
important features from the concatenated features produced by the first two deep learning networks. Deep SDBNet classifies
the segments into four classes: Normal NREM, Normal REM, SDB NREM and SDB REM. To our knowledge, this is the
first study to propose an architecture for this classification using EEG, EMG and EOG signals. ISRUC-Sleep database is
used for evaluating the performance of our proposed model. The average accuracy, sensitivity, specificity and F1 score are
78.16%, 55.87%, 85.23%, and 54%, respectively.
Keywords - Obstructive Sleep Apnoea, Hypopnoea, CenteralApnoea, Mixed Apnoea, Sleep Disorder Breathing, EEG,
EMG, EOG, Deep Learning, Classification, Convolutional Neural Networks, REM NREM.