Research On Sleep Staging Method Based On Deep Learning

Whether it is in scientific research or people’s cognition,it is well known that good sleep is fundamental for maintaining good physical and mental health and good quality of life.However,due to many factors such as aging and mental pressure,sleep disorders such as sleep apnea,parasomnias,and hypersomnia,affect about 40 percent of Chinese people according to a Chinese survey.The focus of this paper is to put forward different solutions to the problems that may arise in the analysis of sleep model under different conditions,and improve the traditional sleep staging model,so as to get a relatively better model for solving different problems.The main work of this thesis is as follows:(1)In this paper,a dual-modal and multi-scale deep neural network system using Electroencephalogram(EEG)and Electrocardiograph(ECG)signals is proposed for sleep staging in an end-to-end way.The proposed network adopts a dual-modal(one for EEG and one for ECG)and multi-scale structure with its basic block being convolutional module.The dual-modal structure is designed to combine the merits from two different signals for a more robust sleep staging.The multi-scale structure is adopted to utilize features at different scales of EEG signals,which has been found to be very important in characterizing the sleep states.The extracted features from EEG and ECG signals are fused after several full connection operations and then inputted into a classifier for sleep staging.The performance of the proposed network on sleep staging was evaluated on the public MIT-BIH polysomnography dataset.The experimental results indicated averaged accuracy of 97.97%,98.84%,88.80%,and 80.40% for distinguishing between ‘deep sleep vs.light sleep’,‘rapid eye movement stage(REM)vs.non-rapid eye movement stage(NREM)’,‘sleep vs.wake’,and ‘wake vs.deep sleep vs.light sleep vs.REM’ respectively.(2)Currently,most deep-learning-based sleep staging system relies heavily on a large number of labeled physiological signals.However,sleep-related data,such as polysommography(PSG),are often manually labeled by one or more than one professional experts with much effort.Meanwhile,due to physiological differences that existed among different subjects,how to boost the performance of trained models on an unseen dataset is still an open issue.In order to improve the performance of the model on invisible datasets,this paper borrows the prior knowledge from labeled datasets to train unlabeled or a few labeled datasets through unsupervised domain adaptation.To overcome the problem of insufficient labeled data for training robust sleep staging systems,this study aims to investigate the training of an unlabeled target sleep dataset from a labeled source sleep dataset in a deep learning framework,which integrates a conditional and collaborative adversarial domain adaptation module.To facilitate the network to learn domain-invariant features,a domain classifier is deployed for each feature extraction block at different scale.The input to the domain classifier at different level is the multilinear mapping of the sleep stage prediction vector and the corresponding feature vector at this level.It is assumed that the feedback of the class information provided by the network into the domain classifier can be beneficial to help the network to reduce the feature distribution distance between different domains.The effectiveness of the proposed method is verified by using public Sleep-EDF dataset.Compared to other domain adaptation approaches,the proposed approaches can provide better sleep staging performance in different model transferring tasks.