Facial Multi-Region rPPG Extraction from Smartphone Videos with Comparative Analysis of Finger Videos for SpO₂...

Valuable insights into respiratory and cardiovascular health can be gained through continuous monitoring of blood oxygen saturation (SpO₂). Traditionally, SpO₂ is measured using contact-based pulse oximeters, which can be uncomfortable for prolonged use and may not be suitable for individuals with poor circulation in the extremities. Remote photoplethysmography (rPPG) has emerged as a non-contact technology for vital signs monitoring. However, the accuracy of rPPG is highly dependent on the selection of regions of interest (ROIs). This paper introduces a non-invasive system for SpO₂ measurement using rPPG signals extracted and analyzed from smartphone facial videos. The system incorporates data pre-processing, color magnification via Eulerian Video Magnification (EVM), and calibration techniques in order to rank ROIs based on accuracy in predicting SpO₂. Eight facial ROIs were compared to the fingertip. A custom dataset of synchronized facial and fingertip videos was created and validated against ground truth SpO₂ values from Finapres NOVA, a clinically validated commercial device. The analysis demonstrated that color magnification followed by calibration emerges as the best-performing method across all ROIs. The results also showed the superiority of specific facial ROIs over finger. The cheeks and forehead ROIs outperformed the fingertip with mean absolute error (MAE) and root mean square error (RMSE) of (0.578%, 0.701%) and (0.615%, 0.695%), respectively, compared to the finger's (0.902%, 1.108%). These results meet the standards set by the International Organization for Standardization (ISO) and the Food and Drug Administration (FDA).