A new approach for daily life Blood-Pressure estimation using smart watch

Abstract

Hypertension has become a major factor affecting people’s health. Timely monitoring and prevention of hypertension are of great significance. Most blood-pressure monitoring devices have redundant accessories and complex operations, which are not suitable for daily life applications. To solve these problems, we designed an application-level smart watch that can calculate the PAT feature using simultaneously collected ECG and PPG signals. By fusing the time–frequency-domain features of PPG and the correlation matrix between pulses, the information related to blood pressure in physiological signals was comprehensively extracted. Most of unstable signal segments can be eliminated and the impact of signal disturbances on overall accuracy will be reduced. This paper proposed two blood-pressure estimation modes: a two-step calib-free mode and a personal calibration mode based on transfer learning. The results on our self-established dataset (36 normotensive subjects and 40 hypertensive subjects) indicated that the two-step calib-free mode could estimate blood pressure with accuracy of 0.23 ± 6.1 mmHg and 0.63 ± 9.35 mmHg for DBP and SBP respectively, which was better than previous calib-free methods. Through the personal calibration mode, the estimation accuracy reached 0.02 ± 5.94 mmHg for DBP and 0.3 ± 7.69 mmHg for SBP, which complied with the Association for the Advancement of Medical Instrumentation standard.