Research On Heart Beat Extraction And Sleep Apnea Event Monitoring Based On BCG Signals

Sleep apnea is a sleep related disease that significantly affects human health,and in recent years,the population affected by this disease has continued to expand.In the diagnosis of sleep disorders,the gold standard is polysomnography(PSG).However,PSG is expensive and uncomfortable.In order to simplify PSG,researchers have conducted single channel or multi-channel sleep apnea event detection studies using ECG signals,respiratory signals,SPO2,etc.The ballistocardiograms(BCG)signal has received widespread attention and research due to its advantages such as non-invasive and noncontact collection.Using BCG signals for detecting sleep apnea events is suitable for long-term supervision in households.To achieve monitoring of sleep apnea events based on BCG signals,this thesis proposes two methods.One is the end-to-end convolutional neural network method,which can adaptively extract features from BCG signals and classify them.In the experiment,a sliding window with a step size of 1 second was used to segment signal segments,achieving the recognition of sleep apnea events per second.The accuracy,precision,recall and F-score of this method is 87.81%,99.83%,87.87%and 93.43%.Although deep learning methods have high accuracy,their interpretability is not strong.Clinical studies have shown that HRV(heart rate variability)is associated with sleep conditions.Therefore,this thesis also proposes a machine learning based on HRV.This method requires HRV features to detect sleep apnea using traditional machine learning methods.In order to improve the accuracy of estimating heartbeat from BCG signals,this thesis innovatively proposes a method for automatically labeling the heartbeat position of BCG signals,and establishes an end-to-end classification model based on U-net++network for BCG signals to heartbeat sequences.The accuracy and precision of heartbeat detection is 99.34% and 98.31%.In the end,the accuracy,precision and recall of this method detecting sleep apnea events is 78.85%,80.30%,95.28%.