Non-contact blood pressure estimation by imaging photoplethysmography

This contribution is dedicated to the non-contact, calibration-free estimation of blood pressure (BP) using imaging photoplethysmography (iPPG). Central aims of our investigation are (1) to assess the quality of continuous BP estimation by iPPG in general, (2) to quantify the impact of different processing strategies on the quality of continuous BP estimation and (3) to deepen the understanding on the function and malfunction of continuous BP estimation. Within this work, we pursue a feature-based approach to continuously estimate BP considering different variants for video processing and signal formation. For training and testing, we use own experimental data covering 43 healthy subjects, who participated in an experimental study featuring different stimuli. Within a leave-one-subject-out cross validation, we yield minimum median mean absolute errors (MAE) over all subjects down to 12.8 mmHg for systolic blood pressure (SBP), 8.81 mmHg for diastolic blood pressure (DBP) and 7.14 mmHg for pulse pressure (PP). Maximum median correlation coefficients between BP estimates and reference BP were 0.25, 0.70 and 0.79 for SBP, DBP and PP. Our results are in line with results of comparable works from the literature. Within our contribution, we demonstrate the efficiency of an advanced video processing by Level Sets as well as the positive impact of color channel combinations. Improvements in MAE for DBP and PP compared to a baseline estimator as well as high correlation coefficients for DBP and PP indicate the potential for BP estimation using iPPG. However, absolute errors are too high illustrating the need for further developments.Clinical relevance - The non-contact continuous acquisition of BP using iPPG is of high interest for clinical and out-of-hospital applications. To date, the number of available works and the basic knowledge, e.g. with respect to the general applicability, impairing factors and optimal processing strategies, is limited. Accordingly, basic research on the matter is needed to enable a future use. Our study makes a contribution in this regard by evaluating different processing methods for their applicability to estimate BP using iPPG.