Study On A Portable SAS Automatic Detection System Based On Hybrid Convolutional Neural Network

Sleep Apnea Syndrome(SAS),often called "sleeping killer",is a sleep disorder disease which seriously affects the sleep quality and health of public.As the golden standard for SAS clinical diagnosis,polysomnography(PSG)has the disadvantages of complicated operation and high price,which makes the diagnosis and census of SAS diseases difficult to popularize.So far,the diagnosis rate of SAS is less than 20%.This article focuses on the development of a SAS automatic detection system,which can realize convenient detection and is an auxiliary tool for undisturbed sleep.According to the research ideas of problem proposing,system modeling,and research on related theories,realization of system,simulation and data testing,we conduct an in-depth research on SAS detection,data preprocessing and the methods of automatic detection model establishment.The innovative work is as follows:Firstly,in view of the hardware problems of SAS monitoring equipment,such as cumbersome wearing,easy allergy,and uncomfortable monitoring,a pulse oximeter based on a wraparound oximeter was designed in this study to realize the functions of photoplethysmography(PPG)acquisition,preprocessing and wireless transmission.Secondly,due to the PPG signal is weak and susceptible to various noises,we propose a PPG signal denoising method based on double median filtering by analyzing the main noise characteristics of PPG signal.To evaluate the denoising effect,we use the database provided by Physionet and the data collected by our self-developed pulse oximeter.Furthermore,we compare the proposed method with traditional wavelet denoising algorithm,and the comparison result shows that the method has high reliability and effectiveness,which lays a foundation for improving the accuracy of the subsequent automatic detection model.Thirdly,aiming at the high-dimensional data generated in sleep-respiratory monitoring,the PCA-based and PRV-based feature extraction methods were studied for dimensionality reduction.Accurate PPG signal peak detection is crucial to the PRV-based feature extraction.We,therefore,propose a peak detection method based on quadratic spline wavelet modulus maximum algorithm.The detection method guarantees the accuracy of the peak-to-peak sequence of the PPG signal and improves the effectiveness of the PRV-based feature extraction method.Fourthly,a method for constructing a GACNN-based SAS automatic classification model is proposed.We study the establishment and optimization of shallow modeling methods of BP,and the deep modeling methods of DBN,GACNN,and establish an automatic classification model using the PPG signal obtained from the database and the measured samples.Furthermore,we used the 10-fold cross-validation method to perform a comparative experiment on the AH automatic classification models.Finally,we conducted the experiments to evaluate the SAS automatic detection system.By using the PPG data collected by the PSG and the pulse oximeter,the effectiveness of the GASNN-based SAS automatic detection system was preliminarily verified.The GACNN-based SAS automatic detection system studied in this paper has the advantages of simple signal acquisition,non-sensitization of electrodes,easy to wear,and low disturbance.And the GACNN classification model has good predictive power and generalization ability.This study can provide technical support for rapid screening and early detection of SAS,and contribute to improve medical efficiency,which has academic meaning and application value.