Remote Blood Pressure Measurement Through Facial PPG Signals

Blood pressure (BP) is a measure of the force exerted by blood against the walls of arteries as the heart pumps it throughout the body. Today, isolated patients require regular and precise blood pressure measurements. The current sphygmomanometers demand a contact-based measurement setup using cuffs, which increases the risk of infection and skin sensitivity. The dataset comprises videos of 142 participants engaged in three activities. These activities were recorded using an iPhone 11 camera mounted on a tripod, and the videos underwent preprocessing techniques that includes background removal, face detection, selection of region of interest (ROI), and isolation of RGB channels. Motion artifacts removal is done by Richardson Lucy Algorithm - a blind deconvolution method which restores the image with high PSNR, BRISQUE score and edge sharpness. Extraction of ROI by UNET segmentation is done using rectangular shaped mask which has given good predictions with 95.02% accuracy. For non-contact BP measurement, Photoplethysmography (PPG) signals are extracted from each channel of the video. Two parameters, Pulse Transit Time (PTT) and Pulse Amplitude Ratio (PAR), are extracted from each dicrotic notches of PPG signal. Gradient Boosting Machine algorithm estimates both systolic and diastolic pressure with a minimum Root mean square error (RMSE) of 0.09 and an accuracy mean of 99.08%, outperforming the Multiple Linear Regression, Random Forest Classifier and Support Vector Machine algorithms.