Digital Health -> Digital Health D-PO01 - Featured Poster Session (ID 11) Poster

D-PO01-064 - Ai-based Method For Monitoring Pulse Rate Using Facial Videoplethysmography Recorded With Mobile Devices (ID 904)


Background: Facial videoplethysmography (VPG) is a novel technology for monitoring pulse rate (PR) using the video camera of smart devices. The primary advantage of VPG technology is to enable long-term intermittent monitoring of pulse rate without the burden of using wearable devices.
Objective: We evaluated the reliability and accuracy of self-acquired PR by cardiac patients using video selfies from their smart tablets in an uncontrolled environment.
Methods: Cardiac patients were enrolled after successful cardioversion or ablation for atrial fibrillation. We provided each subject with a smart tablet for a maximum of 14 days while wearing an ECG patch. We asked the subjects to perform 30-sec. video selfies twice a day using an application extracting VPG signals from facial recordings. The average PR and heart rate (HR) values were extracted from the synchronized VPG and ECG signals, respectively. Machine learning was trained to reject VPG recordings associated with an error ≥10% in reference to HR using a 30/70% split of the data (validation based on 30%).
Results: Sixty subjects (47m/13f, 65±8 yrs) recorded 880 video-based PR in sinus rhythm from June 2018 to May 2019. Subjects wore the ECG patch for 11 days on average (ranging from 1 to 15 days). The recorded HR varied between 40 and 122 bpm. Random Forest model was trained to reject measurements with an error >10% between VPG and ECG rates. Bland Altman applied to the validation set revealed a mean difference between PR and HR of 0.3±9.8 bpm while rejecting 33% of the VPG signals for low signal quality.
Conclusion: We developed a contactless monitoring method enabling smart devices to measure PR without the need for patients to use a dedicated device.