Non-Invasive Camera-Based Cardiac Output Monitoring via Facial Photoplethysmography: Advancing Hemodynamic Assessment
Abstract:
This study presents a novel non-contact approach for cardiac output (CO) monitoring via facial photoplethysmography (PPG), with a focus on advancing hemodynamic assessment during intravenous fluid bolus (FB) therapy in emergency settings. This study conducted at Odense University Hospital in Denmark, which involves four participants by capturing over 350 minutes of video data under clinical conditions. A MATLAB-based CO trend monitoring pipeline was developed, which comprises a robust two-stage process for dynamic region-of-interest (ROI) segmentation. Eight subregions of the forehead were selected based on skin detection and facial extraction using a deep learning-based face detection framework. Green (G) and Green-Red (G-R) PPG signal processing methods were employed to mitigate amplitude nonlinearity effects, ensuring accurate CO estimation. The proposed system demonstrated strong correlations between PPG-derived CO trends and ground truth values, achieving average correlation coefficients (r values) of 0.64 (G-R) and 0.59 (G) across all participants. Peak r values reached 0.82 and 0.87, respectively, for optimal ROIs and subjects. These findings highlight the potential of camera-based CO monitoring as a non-invasive, scalable solution for critical care and emergency applications, offering a transformative tool for hemodynamic evaluation in resource-constrained and timesensitive scenarios.
