Study On Blood Pressure Measurement Method Based On ECG And PPG Wave Signals

Hypertension is a chronic disease affecting the heart and vascular system,which can cause millions of deaths every year.The blood pressure(BP)waveform contains vivid cardiovascular information,which can reflect the surrounding vascular resistance and cardiac function.Therefore,accurate measurement of BP waveform is of great significance for the diagnosis and prevention of hypertension and the reduction of the incidence rate of cardiovascular and cerebrovascular diseases.Currently,BP measurement methods include direct and indirect methods.In clinics,the direct method can obtain accurate blood pressure data,but the invasive method increases the risk of complications and does not apply to the general population.The cuff-blocking sphygmomanometer is widely used in life to measure blood pressure non-invasively and intermittently,but even a few minutes of abnormal blood pressure will increase the risk of death and disease.Therefore,the ideal blood pressure measurement method should be non-invasive and continuous.Aiming at the problem of non-invasive continuous high-precision measurement of blood pressure,this paper proposes to build a blood pressure measurement model using a deep neural network;Considering the real-time problem of blood pressure measurement,the application of a time convolution network to establish a blood pressure measurement model was studied to improve running speed and accuracy.The main work and achievements of this study are as follows:(1)Research on effective methods for detecting R-wave in electrocardiogram signals and pulse wave peaks.This paper used electrocardiograms(ECG)and pulse wave signals(PPG)to measure BP.To improve the quality of input signals,the Butterworth filter and the integrated empirical mode decomposition method are used to filter ECG and PPG,respectively.At the same time,the variational mode decomposition combined with the wavelet transform method is used to detect the R wave of ECG;Using the sliding window search method to detect the peaks of PPG.Experiments show that the signal preprocessing operation can effectively filter out the noise and invalid data,while preserving the original information of the sample as much as possible,preparing for subsequent model learning.(2)Research the application of a two-stage attention-based recurrent neural network model in the field of blood pressure measurement.To address the problem of information loss when encoding long-term temporal data using sequence to sequence models,this study introduced an input attention mechanism in the encoding phase to enable the model to adaptively select features in the input signal that are strongly correlated with blood pressure.In the decoding phase,the temporal attention mechanism was introduced to improve the ability of the network to capture long-term temporal dependencies of the sequence.The introduction of the two mechanisms enables the model to take into account the effects of both characteristics and time on blood pressure.The effectiveness of this model was verified through experimental comparison with other blood pressure measurement methods.(3)Research the application of feature fusion time convolutional network model based on attention mechanism in the field of blood pressure measurement.To address the problem of slow model running speed caused by serial processing of ECG and PPG data in recurrent neural networks,a feature fusion time convolutional network is proposed to meet the needs of multi-dimensional variable input while improving the training speed of blood pressure measurement models.The model combines time convolution networks,empty causal convolutions,residual modules,and jump layer connection mechanisms to design independent submodules for detail feature extraction of the two variables.Then,attention mechanism is used to efficiently fuse the features proposed by the submodules,and finally,internal abstract features are further extracted through time convolutional network.The effectiveness of the proposed model was verified by comparing it with other traditional blood pressure measurement methods.This blood pressure measurement model can achieve high-precision real-time measurement of blood pressure waveforms.