Reconstruction Of Central Arterial Pressure Signal Based On Long Short-term Memory Network

Cardiovascular diseases are one of the important factors threatening residents' health.Therefore,the monitoring and diagnosis of human cardiovascular health is particularly important.Compared with the traditional peripheral arterial pressure,central arterial pressure has a higher correlation with the occurrence of many cardiovascular diseases,which can provide an important basis for diagnosis and disease prevention.However,the existing central arterial pressure detection methods have the problems of low accuracy,poor generalization and high cost,which limits its clinical application and popularization.In this paper,a new method of central arterial pressure signal reconstruction using deep learning is studied,a central arterial pressure signal reconstruction model based on long-term and short-term memory network is proposed,and a noninvasive central arterial pressure detection system is designed.The main work of this paper is as follows:(1)Reconstruction of central arterial pressure based on CNN-BiLSTM.In this paper,a central arterial pressure reconstruction model(CNN-BiLSTM)based on convolutional neural network and bidirectional long short-term memory network is proposed.The model has strong learning ability of global and local features at the same time,and can realize the good reconstruction of central arterial pressure.Compared with two traditional methods and four deep learning models,the experimental results show that CNN-BiLSTM model achieves the best reconstruction effect of central arterial pressure waveform(MAE:2.18 ± 0.13 mmhg,RMSE: 2.95 ± 0.16mmhg).At the same time,the reconstruction effect of central artery systolic pressure(RMSE: 3.34 ± 0.91mmhg)and diastolic pressure(RMSE: 2.41 ± 0.18mmhg)was also excellent.(2)Reconstruction of central arterial pressure based on CBi-SAN.This paper proposes a neural network reconstruction mechanism based on CNN-BiLSTM and self-attentione(CBi-SAN).Based on CNN-BiLSTM,the model further strengthens the learning ability of local features and effectively improves the reconstruction effect of the model on central arterial pressure waveform and important parameters.The experimental results show that compared with CNN-BiLSTM and transformer neural network,CBi-SAN model not only improves the waveform reconstruction effect(MAE: 2.23 ± 0.11 mmhg,RMSE: 2.21 ± 0.07mmhg),but also improves the detection effect of central artery systolic blood pressure(RMSE: 2.94 ± 0.48mmhg)and diastolic blood pressure(RMSE:1.96 ± 0.06mmhg).(3)Study on noninvasive central arterial pressure detection system.Aiming at the problems of high cost,low accuracy and difficult to guarantee the generalization ability of the existing central arterial pressure detection equipment,a noninvasive central arterial pressure detection system based on CBi-SAN model is built in this paper.The system includes two parts: hardware acquisition and software processing.The hardware acquisition part realizes the acquisition of radial artery blood pressure,PPG signal and PPG signal,and transmits them to the host through serial communication.The software processing part realizes the data reading,and completes the radial artery pressure waveform reconstruction based on PPG signal by using CBi-SAN model,and finally realizes the noninvasive detection of central artery pressure.