Research On Automatic Apnea Recognition Algorithm Based On Nonlinear Dynamic Characteristics Of Sleep EE

Sleep apnea syndrome is a common sleep-disordered breathing.It not only severely reduces the patient’s sleep quality,but also greatly increases the risk of depression,hypertension,Alzheimer’s and other diseases.At present,the diagnosis of apnea mainly relies on polysomnography.However,due to the complexity of the operation,the high price and the high requirements for the equipment environment,it is difficult for polysomnography to be popularized at home.To this end,this study intends to use the method of nonlinear dynamics to analyze the electroencephalography(EEG),and adopt the artificial intelligence-based methods to realize the automatic detection of apnea,thereby simplifying the detection method of apnea.This study used sleep EEG data from two independent databases containing a total of 55 apnea patients.First,the EEG signal is divided into five sub-band signals according to the rhythm.Secondly,five nonlinear dynamic methods including correlation dimension,maximum Lyapunov exponent,detrending fluctuation analysis,Lempel-Ziv complexity and fuzzy entropy were applied to extract the characteristic parameters of EEG signals,and then statistical methods are adopted to test these features.The results showed that there were significant differences in the nonlinear dynamic characteristic parameters of EEG segments between apnea and normal breathing in specific frequency bands.On this basis,this research further established a machine learning classification model,and used the non-linear dynamic characteristics of EEG as the input vector to classify apnea events and normal respiratory events.The results showed that the five EEG nonlinear features all had a certain degree of recognition for apnea events.Among them,the method based on fuzzy entropy feature extraction and random forest classification model achieved the best apnea event recognition result.Its accuracy was higher than 93% in both two independent databases.The accuracy rates of correlation dimension,maximum Lyapunov index,detrending fluctuation analysis scaling index and Lempel-Ziv complexity are 84.24%,77.34%,85.53%and79.66%,respectively.In order to further improve the detection accuracy of apnea events,this study also established a deep learning network,based on fuzzy entropy feature parameters,achieved 97.65% and96.18% accuracy in the training set and test set,respectively.This study proved that the nonlinear dynamic index can effectively represent the changes of EEG characteristics caused by apnea,and can be used as the input of machine learning classification model to realize the automatic detection of apnea events.It provided new ideas for the diagnosis of apnea,and provided the research foundation and technical support for the realization of portable apnea detection equipment.