Research On Continuous Blood Pressure Prediction Of PPG Signal Based On Cascaded U-NET Network

People’s irregular diet and life have led to the rising rate of staying up late and obesity,which has caused more and more cardiovascular diseases.In order to effectively prevent and treat blood pressure diseases,there is an increasing demand for accurate and continuous blood pressure measurement.In recent years,more and more scholars have begun to model and study blood pressure data,and establish blood pressure prediction models to predict human blood pressure values.From the prediction results,the continuous measurement of blood pressure based on PPG signals is feasible,but there are still The blood pressure prediction algorithm is not real-time enough and the prediction result is not accurate enough,which does not meet the international standard.How to achieve high-accuracy blood pressure prediction through PPG signal is a difficult problem that needs to be solved.The main improvements and innovations in this paper are as follows:(1)This paper proposes a mathematical method for PPG signal processing and a method for reducing the phase difference between PPG signal and blood pressure data.Targeted construction of data quality optimization and signal time alignment methods.Among them,in the process of signal acquisition,due to external light,human movement,breathing fluctuations and other reasons,it is inevitable that there will be noise in the signal,which has a great impact on the prediction accuracy.Before it is used for prediction,it is necessary to eliminate the baseline drift and high-frequency interference in the signal,and then judge the abnormal data caused by the measurement irregularity,and remove the data of no research significance;the R peak of the ECG is used as a reference to realize the time alignment of the PPG signal,Reduce time lag with blood pressure data.Reduce the influence of data quality and the phase difference between PPG signal and blood pressure data on the accuracy of blood pressure prediction,and the experiment verifies that this operation can effectively optimize the prediction effect.(2)This paper proposes a blood pressure prediction method based on PPG signal and blood pressure stiffness convolutional neural network.A fusion U-Net and MultiResUNet network model is proposed,and in order to improve the prediction accuracy of SBP,combined with the feature points of PPG,the feature value of blood vessel hardness is introduced.By comparing the experimental data with multiple standards,compared with the BHS standard,it has reached the A-level requirements.The average prediction accuracy of DBP is 13.33 percentage points higher than that of the A-level standard,and SBP is 3.67 percentage points higher.At the same time,it also meets the requirements of MAE<=5 and STD<=8 in the AAMI standard.(3)On the basis of the blood pressure prediction model,a portable continuous blood pressure prediction device based on reflective PPG acquisition and prediction is realized.The PPG signal is collected to remove noise and extract peaks and valleys.At the same time,it can also estimate the heart rate and blood oxygen signal.Finally,the experimental model predicts the blood pressure value and compares it with the measurement results of the electronic sphygmomanometer.The mean absolute error of DBP is 2.93mmHg,and the mean absolute error of SBP is 4.33mmHg.Obtained relatively good results.In summary,this paper proposes a cascade-based U-Net blood pressure prediction model,which improves the accuracy of blood pressure prediction by means of time alignment of PPG signals and the introduction of blood vessel stiffness features,and the prediction results also reach the BHS standard A level.and AAMI standards.At the same time,based on the experimental model,a device for PPG acquisition and continuous blood pressure measurement based on reflection is designed and implemented,which has high prediction effect and good practicability.