Single Channel Sleep Staging Based On Deep Learning

Sleep is an essential physiological activity.Studies have found that a good night’s sleep is important for health.Insomnia can produce different degrees of illness.Sleep is based on sleep staging.Traditionally,sleep staging is done manually by experts,but this method has drawbacks.In order to solve the problem of artificial sleep staging,we propose a method of automatic sleep staging based on deep learning.The Convolutional Neural Networks(CNN)and Long short term Memory Networks(LSTM)are involved in the model.Our model has 11 layers,including four convolution layers and two LSTM layers.The LSTM network captures the relationship between sleep periods before and after stage.The data are from sleep data from the American national physiological database.Due to the imbalance of data type,we conducted up-sampling operation.In addition,we have deleted some special data and data with uncertain staging.ln addition,we did not do any preprocessing of the original data.The overall classification accuracy of our model reached 80%,and the MF1 value reached 75%.This proves that deep learning is feasible for processing sequential medical data.Compared with the traditional staging model based on statistical rules.this model is simpler,more efficient and has better generalization performance.It is suitable for nonlinear and non-stationary eeg signal,and our work also provides reference for other similar physiological signal processing.