Research And Application Of Automatic Sleep Staging Based On Multimodality Data

Sleep is an important part of people’s life activities.The quality of sleep is closely related to people’s physical and mental health.Sleep staging is a technique to objectively analyze sleep quality,and it is also an important reference for the diagnosis and treatment of various sleep disorders related diseases.Artificial vision labeling sleep period is the gold standard of sleep staging,but sleep physicians need to spend a lot of time processing sleep data,and a reliable automatic sleep staging method is urgently needed,which can not only reduce the workload of sleep physicians,but also promote the development of portable sleep monitoring technology.At present,most of the methods to realize automatic sleep staging use a single EEG signal as data support,extract its features,and use the training model as the basis of sleep staging.However,a single data type has many limitations.Human sleep conversion is not only supported by EEG,but also can be used as favorable data for studying human sleep stage.Moreover,it also has limitations to focus solely on the features extracted from the data set,Most studies often ignore the characteristics of sleep itself.Sleep stages mainly face the classification of sleep stages,and there are some relationships between sleep stages.The pre and post state of sleep segments is an important reference for sleep doctors to label,and the transformation of sleep stages follows a certain law.In this context,this paper proposes an automatic sleep staging algorithm based on multimodal data,combined with feature learning and sequence analysis,which extracts relevant features based on traditional machine learning and deep learning methods respectively.On the one hand,it uses the multi-type data support of multimodal data to provide richer and valuable data support for the research algorithm,It focuses on how to use the time dependence of sleep state to improve the performance of the algorithm.The research of this paper is mainly divided into the following aspects:(1)Sleep staging based on traditional machine learning method:according to the characteristics of physiological signals in each sleep stage,the time-domain features,time-frequency features and nonlinear features are extracted from EEG,EOG and EMG channels,and the feature vector expression of each sleep segment is obtained.The gradientlifting tree(gbdt)algorithm is used to fit the data distribution and establish the pre classification model.Considering the importance of sequence analysis,conditional random field(CRF)is used to learn the transition law of state chain to revise the pre classification results and output more reasonable sequence staging results.(2)Sleep staging based on deep learning method:firstly,convolution neural network(DenseNet)with automatic learning features is constructed,which can effectively fuse time-frequency features with the superposition of convolution layers,and take compressed signal as low dimensional feature vector.Secondly,the long short term memory model(LSTM),which is good at dealing with time series dependence,and the improved Gru network are used to analyze the phase transition relationship,so that the accuracy of sleep staging can be further improved.In the first mock exam,the main points of this paper are as follows:first,compared with the single modality used in most sleep stages.For EEG signals,this study uses multimodal data combined with EEG,EEG and EMG as data support,which not only provides various types of data for the experiment,but also enriches the data characteristics,and provides favorable support for improving the accuracy of sleep staging.In addition,this study also relies on the premise of sleep feature classification,considering that the sleep stage conversion follows a certain law,extracts the sleep stage conversion law through the corresponding algorithm,and adjusts the staging results,so as to improve the accuracy.This study conducted experiments on two public data sets(sleep edfx and physionet2018).The experimental results show that the proposed method can achieve stable and reliable automatic staging,and the staging accuracy of sleep staging based on multimodal data is significantly improved compared with that of single EEG signal.Finally,adding sequence analysis to the model can significantly improve the accuracy of sleep staging.The combination of feature learning and sequence analysis is a practical method of sleep analysis.