Regression-forests-based estimation of blood pressure using the pulse transit time obtained by facial photoplethysmogram

We investigated the estimation of blood pressure (BP) based on the pulse transit time (PTT) and regression forests and propose a noncontact PTT measurement system using microwave and image sensors. The PTT is the time difference between an electrocardiogram and photoplethysmogram recorded in a certain body region and considered a potential indicator of blood pressure. The aim of the present study was to measure the PTT remotely during daily activities, and the photoplethysmogram was thus measured in the facial region of the subject. To estimate the blood pressure precisely, not only the PTT but also other physical factors (i.e., age, height, weight, and body mass index) were considered. Contact-based experiments conducted on 75 subjects showed that the body mass index (BMI) contributes to the estimation of blood pressure. The absolute estimation error obtained using regression forests and the PTT and body mass index as explanatory variables (10 mmHg) was lower than that obtained using a conventional method of linear regression and the PTT (12 mmHg). Furthermore, results obtained using the proposed noncontact PTT measurement system correlated adequately with results obtained using a conventional contact-based system, demonstrating the potential of the remote estimation of blood pressure.