Measuring Vital Signs via Video: Heart Rate and Blood Pressure

Abstract

Heart rate and blood pressure are important vital signs in medicine. Enabling self-measurement of heart rate and blood pressure using only a smartphone would greatly expand access to this health information. The goal of this project is to establish a foundation for such a smartphone-based system by addressing the following questions: Can we estimate the fingertip contact surface area from a smartphone video recorded with the fingertip supporting the weight of the smartphone? Does this contact surface area vary periodically over time, at the same frequency as the heart rate? Can it be used to improve the accuracy of photoplethysmographic (PPG) heart rate measurement, which relies solely on green color intensity? Can this contact surface area also be used to estimate systolic blood pressure?

To address these questions, we recorded a dataset of 786 fingertip videos with ground truth heart rate and blood pressure measurements from 62 consented participants. We estimated the contact surface area of the fingertip as an ellipse in each video frame, as well as its average green color intensity. We compared the accuracies of seven algorithm variations for estimating heart rate using these signals. We also tested the relationship between the inverse of the average fingertip contact surface area over time and the ground truth systolic blood pressure. We demonstrate that in a minority of participants the area signal did enable accurate estimation of systolic blood pressure; however, for the majority of participants noise in the area signal prevented accurate blood pressure estimation. We believe we now understand the source of this noise and could minimize or prevent it in the future.

The contributions of this project are twofold. First, we propose and demonstrate an inverse relationship between fingertip contact surface area supporting a constant weight, here a smartphone, and blood pressure in the fingertip. Second, we provide an algorithm for estimating fingertip contact surface area in each frame of a smartphone video, and show that this signal can be used to improve the accuracy of smartphone-based heart rate measurement over purely PPG approaches. We also provide a proof of principle that, if noise is managed, this relationship can be used to measure systolic blood pressure from fingertip videos. Finally, we note patterns that may reflect clues to other parameters of cardiovascular health, such as blood vessel elasticity, in our data. This work is a first step towards improved video-based heart rate and blood pressure measurement, and perhaps more.