Radar-based blood pressure estimation using multiple features

Abstract

This paper presents a non-contact blood pressure measurement model based on the random forest algorithm and arterial pulse waveform detected by radar. After the radar signal is pre-processed with filtering and smoothing methods, feature parameters of arterial pulse waves are automatically extracted, and correlation analysis is conducted to further explore the relationship between feature parameters and blood pressure. Then, a blood pressure regression model based on the random forest is established. Compared with the reference blood pressure obtained by a sphygmomanometer, the DBP error of this model is $0.22 \pm 3.85\ \text{mmHg}$ (Mean Difference ± Standard Deviation), and the SBP error is $2.52 \pm 6.73\text{mmHg}$ (Mean Difference ± Standard Deviation), which proves this method can effectively measure blood pressure by using a single radar in a non-contact state.