Contactless arterial stiffness estimation from face video using PINN

Abstract

Remote photoplethysmography (rPPG) enables non-contact physiological monitoring by capturing subtle variations in facial skin color caused by blood volume changes using standard video recordings. While rPPG has been widely explored for estimating vital parameters such as heart rate and respiratory rate, its application for assessing advanced vascular biomarkers, particularly arterial stiffness, remains relatively under-investigated. Arterial stiffness is an important indicator of vascular aging and cardiovascular risk and is traditionally measured using contact-based photoplethysmography (PPG) or specialized clinical devices. Investigate the feasibility of estimating arterial stiffness from facial videos using rPPG signals through waveform morphology analysis and physics-guided learning techniques. Specifically, physiologically meaningful features such as the Reflection Index (RI), systolic–diastolic peak characteristics, and pulse wave–related parameters are studied to evaluate their relationship with vascular compliance. Furthermore, a Physics-Informed Neural Network (PINN)-based framework is proposed to incorporate arterial blood-flow dynamics and pulse-wave reflection principles into the learning process, enabling more reliable estimation of vascular health indicators from remotely acquired signals. The proposed methodology aims to develop a scalable, contactless, and cost-effective framework for arterial stiffness estimation using facial video data. Experimental observations demonstrate that waveform-based rPPG features preserve important physiological characteristics similar to contact PPG signals and can serve as potential markers for vascular assessment. This work contributes toward advancing remote cardiovascular screening systems and supports the development of accessible telemedicine solutions for continuous and non-invasive health monitoring.