Machine Learning Method for Continuous Noninvasive Blood Pressure Detection Based on Random Forest

Xiaohui Chen,Shuyang Yu,Yongfang Zhang,Bin Sun,Fangfang Chu

doi:10.1109/access.2021.3062033

Xiaohui Chen, Shuyang Yu + Show 3 more

Open Access

https://doi.org/10.1109/access.2021.3062033

Copy DOI

Abstract

In order to reduce the influence of differences in human characteristics on the blood pressure prediction model and further improve the accuracy of blood pressure prediction, this paper establishes support vector machine regression model and random forest regression model for accurate blood pressure measurement. First, the photoelectric method is used to obtain the photoelectric plethysmography signal (PPG) and ECG signals from people of different ages, and the blood pressure value is roughly estimated based on the high-quality physiological signals and the vascular elastic cavity model; then the human body characteristics are used as the input parameters of the blood pressure prediction model, and the model parameters are used to find the best parameter combination to improve the prediction performance of the model; finally, through a lot of training and learning, the best blood pressure prediction model is selected to achieve accurate measurement of blood pressure values. It has been verified by experiments that the average absolute error of diastolic and systolic blood pressure based on the random forest optimization model meets the standard of less than 5mmHg formulated by AAMI (American Medical Instrument Promotion Association), which is better consistent with the method of mercury sphygmomanometer, and has more excellent performance than support vector machine regression model under the same conditions.

Highlights

Blood pressure is one of the important indicators to measure human health, and its abnormal fluctuation will bring serious harm to human body [1], [2]
Chen et al.: Machine Learning Method for Continuous Noninvasive Blood Pressure Detection Based on Random Forest model by using deep neural network and combining human physiological characteristics, and the prediction result of this method is obviously better than that of BP neural network method, but its running speed is slow and it is not suitable for real-time monitoring. In response to these problems, this paper establishes a support vector machine and a random forest regression model to predict blood pressure to improve the accuracy of blood pressure measurement and the generality of the model
Compared with the traditional model based on PTT blood pressure prediction, the prediction accuracy of the random forest regression model for systolic blood pressure is increased by 68%, and the diastolic blood pressure is increased by 52%

Summary

INTRODUCTION

Blood pressure is one of the important indicators to measure human health, and its abnormal fluctuation will bring serious harm to human body [1], [2]. X. Chen et al.: Machine Learning Method for Continuous Noninvasive Blood Pressure Detection Based on Random Forest model by using deep neural network and combining human physiological characteristics, and the prediction result of this method is obviously better than that of BP neural network method, but its running speed is slow and it is not suitable for real-time monitoring. Chen et al.: Machine Learning Method for Continuous Noninvasive Blood Pressure Detection Based on Random Forest model by using deep neural network and combining human physiological characteristics, and the prediction result of this method is obviously better than that of BP neural network method, but its running speed is slow and it is not suitable for real-time monitoring In response to these problems, this paper establishes a support vector machine and a random forest regression model to predict blood pressure to improve the accuracy of blood pressure measurement and the generality of the model. Where, PTT is the pulse wave conduction time, z, v is constant, K is the characteristic value of pulse wave, T is the pulse cycle, Td is the time of the diastolic period of the descending pulse wave, m1 and m2 are the constant

DATA NORMALIZATION

BP PREDICTION MODEL BASED ON SUPPORT VECTOR MACHINE REGRESSION

Findings

CONCLUSION