Deep Convolutional Network Method for Automatic Sleep Stage Classification Based on Neurophysiological Signals

The accurate interpretation of sleep stages has a very important significance in the diagnosis of sleep disorders and the assessment of sleep health. The visual inspection on sleep staging required qualified skill and enough clinical experience. Usually, the visual inspection on one's overnight sleep recording takes 1 2 hours. The automatic sleep stage interpretation can reduce the laborious task of visual inspection. In this study, a deep convolutional network model was developed for automatic sleep stage classification based on neurophysiological signals. The residual module is utilized to increase the depth of the network to extract the multi-level features of the sleep stages. The long-short term memory (LSTM) is used to learn the sleep transition mechanism during sleep process. 20-fold cross validation experiment was performed. The results showed that the developed model achieved an accuracy of 81.0 and 73.6 of the macro-averaging F1-score (MF1).