Research And Applications Of Entropy Measures And Compressed Sensing In Sleep Quality Evaluation

Sleep is a complex physiological process,and the level of sleep quality directly affects the health of human body.With the deepening research of sleep,people's cognition of sleep and the evaluation methods of sleep quality have also made great progress.However,sleep is still far from totally known,and there are a lot of problems that need to be solved or keep unclear in the physiological mechanism of sleep and the diagnosis and treatment of sleep diseases.In this context,this paper focuses on the study of sleep quality and works on three main issues,including sleep staging,brain dysfunction of patients with obstructive sleep apnea and sleep Electrocardiogram(ECG).In this paper,we firstly summarized the theoretical basis of Entropy measures and Compressed Sensing,based on which we presented our researches to the three issues.For the first one,this paper studied an automatic sleep staging method based on Entropy measures and Support Vector Machine(SVM),and then a new sleep staging entropy measure was proposed,called tensor approximate entropy,which effectively improved the accuracy of sleep staging.For the second issue,this paper proposed a method to evaluate the hemispheric dominance based on fuzzy entropy and lateralization index,we found and confirmed the phenomenon of lateralization of brain function in patients with obstructive sleep apnea.For the third issue,this paper proposed a new sleep ECG sampling compression method based on Compressed Sensing,which provides an effective solution for the big data acquisition and transmission of sleep ECG.In conclusion,the main works of this study are listed as follows:(1)Several Entropy measures that were commonly used in nonlinear analysis of physiological signals in recent years were reviewed in details,including approximate entropy,sample entropy,fuzzy entropy and fuzzy measure entropy.At the same time,the mathematical model of Compressed Sensing was given,and the applications of Compressed Sensing in EEG and ECG were also discussed.(2)An automatic sleep staging method based on Entropy measures and SVM was studied in this paper.Three Entropy measures,including sample entropy,fuzzy entropy and fuzzy measure entropy were extracted from Electroencephalogram(EEG)and Electrooculogram(EOG)signals,respectively.A multi-classification method based on One-Versus-Rest support vector machine was designed to classify sleep stages in two modes:independent/non-independent samples training and testing.The experimental results showed that this method had obvious advantages in both accuracy and consistency under the same conditions.(3)A new sleep staging entropy measure was proposed,called tensor approximate entropy.By organizing physiological signals into the form of tensors,the physiological state of signal source can be more accurately simulated.The experimental results showed that tensor approximate entropy had good consistency and discrimination ability.In the sleep staging task,the statistical test results showed that the proposed tensor approximate entropy had significant differences among different sleep stages,and compared with traditional Entropy measures,tensor approximate entropy showed a higher classification accuracy.(4)Aiming at the study of brain dysfunction in obstructive sleep apnea,a new method based on fuzzy entropy and lateralization index(LI)was proposed to evaluate the hemispheric dominance.We found and confirmed the phenomenon of brain function lateralization in patients with obstructive sleep apnea.Considering that the traditional LI can only consider a single brain activity evaluation index,we improved LI by proposing a new enhanced lateralization index(ELI),which was applied to clinical polysomnography monitoring data to verify the reliability of this index in evaluating the asymmetry of brain functions(5)A new method of sampling,compression and reconstruction of sleep ECG signals based on Compressed Sensing was studied in this paper.Firstly,several typical factors that may affect the reconstruction accuracy and efficiency of ECG signals were systematically explored.On this basis,this paper proposed a Compressed Sensing method for two-dimensional ECG signal processing.Considering the quasi periodic characteristics of ECG signals,we can enlarge the sparsity of ECG signals to a greater extent by cutting and reconstituting ECG beats into two-dimensional images.The experimental results showed that the ECG reconstruction accuracy can meet the clinical needs with a lower sampling compression rate with this model.