Blood Pressure Prediction Based On ECG And PPG Synchronous Signals

With the progress and development of society,under the influence of the fast-paced era background and living environment,many people do not pay attention to their own diet and daily routine,which leads to the emergence of a series of common high metabolic rate diseases.Now the three high problems tend to be younger.Among them,high pressure blood is the "invisible killer" of human health.Elevated blood pressure increases the risk of developing other organ diseases.Therefore,the monitoring and research of human blood pressure is of great significance.At present,blood pressure detection equipment on the market is mainly divided into two methods: non-invasive blood pressure measurement and invasive blood pressure measurement.Invasive blood pressure measurement involves placing a catheter in a person’s heart or blood vessel cavity,and a probe that detects blood flow and shock.The probe can feel the flow and impact of blood,and blood pressure can be measured in real time through an external pressure sensor device,which can be read in time,and is relatively accurate and reliable.However,its disadvantage is that it is traumatic and has the risk of infection,and its use can only be limited to hospitals.The non-invasive blood pressure measurement method is relative to the invasive blood pressure measurement method.The non-invasive blood pressure detection devices commonly found on the market include electronic sphygmomanometers and mercury sphygmomanometers.Primarily by inflating the cannula,the pressure sensor senses oscillations during the delivery of the tube,allowing it to detect blood pressure within a given time.Although this method is very common and convenient,it will be affected by the ambient temperature during daily use.Peripheral vasoconstriction at low temperatures can cause inaccurate measurements.This topic is an application research of regression prediction of blood pressure based on the electrocardiogram ECG signal and photovolume PPG signal collected by smart wearable devices.It is a method to realize non-invasive real-time detection and monitoring of blood pressure.Firstly,the smart wearable device is used to collect photoplethysmography PPG signal and electrocardiogram ECG signal in real time and synchronously.The collected signals are filtered and denoised.The baseline drift of ECG signal and PPG signal was then removed.For the PPG signal,the first-order derivative and the second-order derivative function are combined to realize the accurate identification and positioning of the five feature points.Then,based on the accurately identified trough points as the division points,the period division and normalization of the continuous PPG signal are realized.Corresponding ECG signal waveform and PPG signal waveform to blood pressure data is inseparable from feature extraction.Therefore,a series of time domain features are extracted from the obtained single-cycle photoplethysmography,such as: amplitude ratio feature,time Scale features and area-related features,etc.Then,the spectral amplitude and phase correlation information of the PPG signal is obtained through the fast Fourier transform,and this part of the information is extracted as the frequency domain feature.In this project,the pulse arrival time PAT feature is specially added,that is,the ECG signal is used as the proximal timing reference and the PPG signal is used as the remote timing reference in the same cycle,from the R peak of the ECG signal to the photoplethysmographic pulse wave.The time elapsed for the maximum value of the first derivative of the PPG signal.Finally,the extracted features are used for data cleaning and feature selection.Then it is sent to the conventional machine learning for the prediction experiment of systolic blood pressure and diastolic blood pressure of blood pressure.In order to improve the precision and accuracy of prediction,the idea of ensemble learning is adopted.The results are fed into random forest and Light GBM models to train and fit.According to the results,the blood pressure prediction model based on ensemble learning is better than conventional machine learning.This paper finally realizes the prediction of systolic and diastolic blood pressure without parameter pre-fitting.