Analysis And Research Of Seismocardiogram Characteristics Based On Machine Learning

Continuous monitoring of heart activity is important to understand the health of the heart.Continuous monitoring of daily heart activity can help detect early signs of heart disease.Electrocardiogram is the gold standard in clinical diagnosis of cardiac status.Ecg signals record the electrical conduction activities of the heart,which cannot reflect the mechanical movement of the heart.Meanwhile,the placement of electrodes in the measurement process requires high requirements,which may fall off during long-term continuous monitoring and affect the measurement,and is not suitable for long-term daily heart monitoring.Seismocardiogram records the signals produced by the mechanical movement of the heart.Cardioseismogram is a noninvasive cardiac activity monitoring tool that can be used for daily,long-term measurements using inexpensive,high-precision accelerometers.However,the periodic characteristics of seismogram signal are too complicated and there is no authoritative definition,so it is difficult to apply the characteristics of seismogram signal in clinical application and diagnostic analysis.Therefore,in order to increase the feasibility of cardiac seismogram signal application in clinical diagnosis and daily monitoring,this paper designed an end-to-end machine learning model,which mapped and fitted the seismogram signal to the ecg signal collected simultaneously in the data set.The main research contents and conclusions are as follows:(a)This paper introduces the theory of cardiac electrical conduction and mechanical vibration,and studies the waveform characteristics of electrocardiogram and seismocardiogram,which lays a foundation for the understanding of data set.(b)The principle of sequence-to-sequence model in deep learning is studied,including convolutional neural network,recurrent neural network and attention mechanism.Two end-to-end deep learning sequence models were designed to fit the ecg signals to the corresponding ecg signals,and the models used in this study were determined by analyzing the fitting experiment results of the five-fold cross-validation method.(c)A comparative fitting experiment was conducted on the model with and without the attention mechanism,and the effectiveness of the attention mechanism in the model was proved by analyzing the experimental results and the fitting error index.(d)The heartbeat interval detection algorithm was used to extract the heartbeat interval sequence and the characteristic parameters of heart rate variability were extracted from the heartbeat interval sequence.The same method was used to extract the control index of ecg signal.The accuracy of the model fitting results was proved by analyzing the sensitivity and positive predictive value of the data center skip interval,and the consistency analysis of the heart rate specific parameters in the data using Bland-Altman plot proved that the fitted signal had a high correlation with the original ecg signal.It is proved that it is feasible to use the ecg signal to fit to the corresponding ecg signal and carry out characteristic analysis and research.