Automatic Sleep Staging Study Of PSG Based On Multi-Scale Salient Features

According to the World Health Organization,an increasing number of people are suffering from sleep-related disorders.These sleep-related disorders seriously affect people’s quality of life,emotional stability,memory and concentration in their daily lives.Therefore,real-time monitoring of patients’ sleep and timely treatment of sleep-related disorders are crucial to people’s physical and mental health.And sleep staging is an important part in assessing sleep quality and assisting in the diagnosis of sleep-induced psychiatric and neurological disorders.Manual staging of sleep stages is a very time-consuming,tedious,boring and easily influenced by subjectivity.Therefore,in order to improve the efficiency of sleep stage classification,it is of great clinical value to propose a highly accurate automatic sleep staging algorithm.In recent years,researchers have started to use techniques related to deep learning to perform automatic sleep staging.Although,many studies have achieved a high classification accuracy.However,there are still some problems that have not been well solved,such as how to capture the multiscale salient features of different sleep stages from physiological signals,how to learn the transition rules from one sleep stage to another from physiological signals of different sleep stages,how to exploit the complementarity of multibranch physiological data for sleep stage staging,how to solve the problem of class imbalance in the sleep data set,and how to reduce the model parameters etc.In order to solve the above problems,this paper proposes a series of automatic sleep staging models based on deep learning for automatic staging of sleep stages.The specific research work in this thesis is as follows:(1)Based on the EEG and EOG channels in PSG,an automatic sleep staging network with a two-branch class U2 structure is proposed in this thesis.The multi-scale salient features in EEG and EOG are extracted by the U2 structure,and the noise in the salient features is filtered using the residual shrinkage module,and finally,in order to improve the robustness of the model,the multi-branch fusion attention module is used to make the model learn the transition rules of sleep stages while effectively fusing EEG and EOG features.The results show that the model effectively improves the accuracy of automatic sleep classification.(2)To extract multiscale salient waveforms,a deep neural network consisting of a Ustructure with a multiscale feature extraction module(MSE)and a convolutional attention module(CBAM)is proposed in this thesis to extract multiscale salient waveforms from a single-channel EEG signal.The U-structure framework with MSE is used to extract multiscale salient features from the original EEG signal.After that,CBAM is used to focus on the significant changes of the signal between different sleep stages and then learn the transition rules between successive sleep stages.In addition,a class-adaptive weighted cross-entropy loss function is proposed to solve the class imbalance problem.Experiments on three public datasets show that the sleep stage classification accuracy of the proposed model is much better than the state-of-the-art results available today compared to existing methods.The proposed model is very promising to completely replace human experts for sleep staging.(3)In order to reduce the model parameters,the automatic sleep staging model can be widely used.In this thesis,a lightweight automatic sleep staging network is proposed.In the proposed model,a multiscale feature pyramid model with fewer parameters is used to extract salient features at different scales in the EEG signal.The joint attention module is used to focus on the global information in the multi-scale features and to learn the sleep transition stages.And convolution is used in the joint attention module instead of the fully connected layer in the traditional attention mechanism to reduce the model parameters.Finally,the validity of the model is verified on two datasets,and the results show that the proposed model not only has high classification accuracy but also has much smaller number of parameters than other methods.It helps to apply automatic sleep staging from large instruments in the clinic to portable devices in daily life.