A Blood Pressure Prediction Method Based on Imaging Photoplethysmography in combination with Machine Learning

This paper proposes a non-contact blood pressure implement (NCBP) system based on imaging photoplethysmography (IPPG) The system collects facial videos through a webcam under ambient light, and extracts pulse wave signals from the videos by means of IPPG technology. From the signals (also called IPPG signals), we extracted 26 features for estimating blood pressure (BP), and trained them through four machine learning algorithms. Finally, we selected the most accurate model for blood pressure prediction. By experimenting on 191 volunteers and comparing four models, support vector regression (SVR) is the best model for predicting blood pressure. The results of SVR are that the standard deviation (STD) and mean absolute error (MAE) of systolic blood pressure (SBP) are 3.35 mmHg, 9.97 mmHg, and those of diastolic blood pressure (DBP) are 2.58 mmHg, 7.59 mmHg respectively. We conclude that through our proposed system based on IPPG technology, blood pressure can be accurately predicted in a non-contact way. In addition, this paper proposes two new methods, the region of interest (ROI) selection method based on colormaps and robust peak extraction method, which solve the key steps in IPPG technology. Finally, we discussed the influence of light intensity on the experiment, and simplified the NCBP experimental device. The system has the potential of replacing the traditional cuff-based sphygmomanometers, and has guiding significance to the future development of blood pressure measurement devices.

Automated summary

This paper introduces a non-contact blood pressure measurement system based on imaging photoplethysmography (IPPG) technology, which utilizes machine learning algorithms to train extracted pulse wave signals and selects the most accurate model for blood pressure prediction.