A Facial-Image-Based Blood Pressure Measurement System Without Calibration

Blood pressure (BP) is a major vital sign that is highly correlated with human health. However, for decades BP measurement has involved a cuff, which can be uncomfortable and even carry the risk of infection. The remote photoplethysmography (rPPG) technique has gained great success in heart rate monitoring for tackling these problems. Some studies integrate the pulse transit time (PTT) into the system and ask users put their hand beside their face. A calibration procedure is usually needed for such methods. Other methods measure the BP with several rPPG waveform features or the whole rPPG segment. However, these methods remain inconvenient, and few methods evaluate their algorithm with large enough datasets to be effective. Hence, we propose a calibration-free and facial-image-based blood pressure measurement system which takes a deep neural network as the backbone. Seven physiological indicators are derived and fed into the network together with three-channel rPPG signals. We also propose a novel strategy for training the models. For a fair and complete evaluation, we compile three massive datasets containing 1143 subjects and 2291 data. These datasets were collected under different scenarios: one from a school laboratory and the other two from two hospital departments. On the complete testing data, the proposed approach yields an MAE of 11.54 mmHg for SBP and 8.09 mmHg for DBP, outperforming previous studies.