Research On Heart Rate Extraction And Electrocardiogram Waveform Reconstruction Based On Ballistocardiogram

Electrocardiogram(ECG)-based daily vital sign monitoring has long played an important role in the prognosis and treatment of cardiovascular disease.However,the monitoring needs to be operated by professionals in a specific place,and there are certain constraints on the human body.Due to the advantages of non-invasive,non-contact and long-term continuous monitoring,the monitoring based on Ballistocardiogram(BCG)has attracted more and more attention of researchers.Extracting the heart rate from the BCG signal is an important problem in clinical monitoring.However,the current heart rate extraction methods still have the characteristics of low accuracy,which hinders the wide clinical application of BCG signals.Therefore,this thesis proposes two heart rate estimation methods.One is based on Complete EEMD With Adaptive Noise(CEEMDAN),the other is based on the Adaptive Template Matching Algorithm(ATMA).In the former,we decompose the BCG signal by using CEEMDAN.By observing the characteristics of the seventh stage component and the BCG signal heart cycle have cycle consistency.The CEEMDAN method can also detect artifacts caused by factors such as body movement in the BCG signal.The latter extracts template of length 250 for each subject,and the correlation coefficient function curve is obtained by continuously shifting the template on the BCG signal,and then the template is updated.The peak detection algorithm is applied on the correlation coefficient function curve,and the heart rate extracted with the gold standard ECG signal is compared.The results show that the recall rates of the two heart rate detection algorithms proposed in this thesis are 96.72%and 99.79%,respectively;the precision rates are 97.31%and 99.91%,respectively.In the current cardiovascular clinical monitoring technology,ECG signal has been widely used and studied.Researchers can extract features corresponding to various types of abnormalities from the complexes of ECG signals.In this thesis,it is found that the ECG signal and the BCG signal have a certain degree of coupling in the time series.Benefiting from the characteristics of non-contact acquisition of BCG signals,if the characteristics of ECG signals can be extracted from BCG signals,the application of BCG in clinical monitoring will be greatly promoted.We proposed a novel and effective deep learning model to establish the mapping between the BCG signal and ECG signal.The deep learning model takes full use of the architecture of the famous U-Net and ResNet.The reconstructed ECG signal can reflect the characteristics of RR interval and QRS complex width and so on.These features play a very important role in the diagnosis of ECG diseases.After the feature mapping between the BCG signal and the ECG signal is established,the cardiovascular disease feature analysis can be performed only by measuring the BCG signal without relying on the ECG signal.In this way,an efficient diagnosis of cardiovascular disease based solely on the BCG signal can be achieved.