| It is noted that recently the incidence of sleep-related diseases has increased quite significantly. Long-term sleep disorder will lower the bodys immunity and result in various diseases, which is seriously harmful for human health. Polysomnography is the gold standard for evaluating the quality of the whole night's sleep and also is the most essential part in the treatment of sleep disorders. However, this method is difficult to apply in the daily sleep monitoring at home because of its strong specialization and complexity.This paper proposes a sleep monitoring method based on multiple physiological parameters, involving two aspects: sleep staging and sleep apnea event detection, and aims at providing theoretical base and methods for the development of wearable sleep monitoring devices applied in family monitoring. The content of this paper includes are as follows.1. A sleep staging algorithm based on single lead EEG signal has been implemented. First, EEG features of all patients in MIT-BIH Polysomnographic database are extracted. Second, by utilizing stepwise regression analysis, 16 effective features are selected. Finally, distance discriminant analysis with linear kernel is to classify each sleep stage and the average accuracy is 81.7%.2. A sleep staging method based on heart rate variability is proposed. Discrete wavelet transform analysis and detrended fluctuation analysis are first to extract features of heart rate variability signal. To test the significance and feasibility of the features extracted by detrended fluctuation analysis which remains unclear so far, statistical analysis is performed. At last, combining two groups' features, support vector machine method is implemented to realize sleep stage classification with an average accuracy of 61.2%.3. An anti-motion interference algorithm of blood oxygen saturation is presented. The algorithm consists of three parts: 1) de-trended analysis, 2) fast Fourier transform, 3) post processing analysis. De-trended analysis aims at calculating DC component by separating DC component and AC component. For the part of fast Fourier transform, AC component is obtained. Finally, post processing analysis, which includes standard averaging and singularity elimination, further improves the accuracy of Sp O2 calculation. Movement experiment demonstrates that anti-motion interference algorithm put here stands out in both mean deviation and variance, compared with the other two algorithms commonly used in the time and frequency domains.4. An algorithm for real-time detection of sleep apnea events based on blood oxygen saturation is put forward. In order to recognize each sleep apnea event in time, begin-end times are required which are labeled by observing respiratory signal. Next, according to the characteristics of every 5 seconds signal slope, fragments that may occur sleep apnea event are warned. Further, the real-time detection of each sleep apnea event was completed by combining with the threshold of standard deviation. As a result, the average accuracy is above 90%, within less than 20 seconds delay.5. Changes of EEG during sleep apnea are analyzed. Whether EEG makes changes before sleep apnea event is explored through event-related potential and event-related(de)synchronization, using samples of the MIT-BIH Polysomnographic Database. Normalized standard deviation analysis and three-time normalized standard deviation test are futher performed to label abnormal points of EEG signal. Results show that EEG signal have changed before sleep apnea events, and the changes in each frequency band are in a certain order.In the present paper, two aspects of sleep monitoring, sleep staging and sleep apnea event detection, are studied based on EEG, heart rate variability and blood oxygen saturation. The algorithms proposed here and other findings meet the requirements of wearable devices whether in terms of computational capacity or performance. They also supply a technical support for daily sleep monitoring at home. |
PDF Full Text Request