Neck-focused Remote Photoplethysmography (rPPG): A comparative study using clinical data and the PyVHR framework

Remote Photoplethysmography (rPPG) promises to turn digital cameras into medical devices with the measurement of heart rates, oxygen saturation and the diagnosis arrhythmias already demonstrated. The face-centric nature of current rPPG techniques limits open-datasets from including subjects with clinically-relevant cardiorespiratory conditions without sharing private medical information. The neck, with few identifiable characteristics, is well suited to overcoming this limitation, as it serves as a region of interest (ROI) for pulse detection during jugular venous examination, a common clinical technique. rPPG techniques hinge on the ability to accurately measure heart rates. This paper aims to motivate the development of neck-based rPPG technology by first showing that comparable heart-rate measurement accuracy can be achieved on the neck using rPPG algorithms developed for the face, and then investigating how subject posture affects measurement accuracy. We do this by using a version of the PyVHR framework adapted to process both necks and faces and using 20 different rPPG approaches. We introduce a novel metric, Heart-Rate Within Tolerance (HRWT) as a more intuitive metric for understanding the overall reliability of an rPPG approach. We found comparable heart rate measurement accuracy on a neck-centric dataset, compared with 5 face-centric datasets, with differences in mean absolute percentage error of 1.79 ± 4.93% BPM between the neck and faces. Neck videos were also found to have an improved HRWT, outperforming facial videos by an average of 15%. For neck-centric videos, subject posture was found to decrease performance as the subject moved from a supine, to seated position.