Objective: Accurate blood pressure (BP) measurement is fundamental for hypertension management. However, standard-of-care (SOC) methods require validated devices, controlled environments, and trained personnel. Remote photoplethysmography (rPPG) offers a contactless, non-invasive alternative for BP estimation using facial video analysis. This study evaluated the accuracy and clinical applicability of the iSelfie Heart Health software for estimating systolic blood pressure (SBP) and diastolic blood pressure (DBP) compared with SOC hospital-grade devices.Design and method: This prospective, single-arm observational study was conducted at a specialized hypertension outpatient clinic in Brazil (November 2023–July 2024). Out of enrolled, 641 subjects met DVM and the inclusion/exclusion criteria. After a standardized 15-minute seated rest period, BP was measured using validated hospital-grade automated devices in accordance with office BP measurement guidelines. Participants then recorded a <1-minute facial video using the iSelfie web-based rPPG platform, accessed via a smartphone or computer camera (less than or equal to 5 MP resolution). The paired measurement sequence (SOC followed by rPPG) was repeated once, with each video acquisition performed within 60 seconds of the corresponding SOC measurement to ensure temporal alignment. Analyses were conducted in both uncalibrated and personalized calibration modes. Calibration employed an individual linear regression model derived from the first paired SOC–rPPG measurement, generating subject-specific correction parameters. The second paired measurement was used to evaluate calibrated performance. Accuracy was assessed using mean absolute error (MAE), standard deviation (SD) of the error, 95% margin of error, Bland–Altman agreement analysis, and cumulative error thresholds (less than or equal to 5, less than or equal to 10, less than or equal to 15 mmHg). A subgroup analysis was performed according to AAMI/ISO 81060-2 criteria, excluding participants with excessive reference BP variability. Results: In calibrated mode, rPPG-based blood pressure estimates demonstrated improved agreement with SOC measurements compared to the uncalibrated analysis, with consistent performance across systolic and diastolic ranges. Sensitivity and specificity for detecting hypertension were 84.6% and 93.4%, respectively. Conclusions: rPPG-based BP estimation demonstrated promising performance as an adjunct tool for ambulatory and home monitoring, supporting further longitudinal and real-world validation.
Sebba-Barroso et al. (Fri,) studied this question.