Background Remote photoplethysmography (rPPG) is a noncontact optical measurement technique that can be used to detect blood volume changes in the microvascular bed of tissue, using a RGB camera. We developed a smartphone application that leverages rPPG in order to monitor the heart rate and respiratory rate of patients using a 30s video selfie. Due to the widespread availability of cameras in smartphones, tablets and laptops, such an application allows to automate noncontact physiological monitoring and recording, improving the quality and accessibility of this essential clinical information. Purpose This study aimed to evaluate the accuracy of such remote measurements compared with regular contact based measurement methods (pulse oximeter) in real-life clinical settings. Methods We recruited 443 patients who consulted the emergency department in 4 French hospitals. After obtaining written informed consent, the patients filled in a questionnaire collecting demographic information including age, gender and the Fitzpatrick skin phototype. Each patient was asked to sit in front of the tablet, with the reference pulse oximeter installed on their finger. All measurements were 30s long. When starting the measurement, the medical staff started to count the patient's respiratory cycles. After 15s, the first COBOX result was available, and then updated every second. At the end of the 30s, the medical staff reported the number of respiratory cycles which were doubled to obtain a respiratory rate in cycles per minute (cpm). Results The mean age was 44 ± 18 years old, 45% was male. A total of 6311 pairs of values of heart rate taken by COBOX (HRCOBOX) and by the reference pulse oximeter (HRreference) and 453 pairs of values of respiratory rate taken by COBOX (RRCOBOX) and by the reference manual count (RRreference) were collected. The proportion of heart rates > 100bpm (beats per minute) was 4.4% and of respiratory rates > 20cpm was 22.9%. The number of insufficient quality measurements was 838 for heart rate (13%). The medical staff could not count reliably the respiratory rate in 20 sessions (4.4%). 30 sessions (6.9%) had a measurement of insufficient quality. For the heart rate, the Pearson correlation between COBOX and the reference was 97.2%. The root mean square error (RMSE) was 3.1bpm. For the respiratory rate, the Pearson correlation between COBOX and the reference was 78.3%. The RMSE was 2.9cpm. Conclusion Our work aimed to evaluate the accuracy of such remote measurements compared with existing contact point measurement methods in real-life clinical settings. Our results described a very good accuracy and correlation between the rPPG system and the gold standard, thus paving the way for more precise and more accessible remote patient monitoring. Such a technology has the potential to be transformative for telemedicine, especially in medical deserts where the access to both physicians and medical devices is lacking.< Réduire