Detection of Atrial Fibrillation Using Contactless Facial Video Monitoring

Background: It is estimated that 33.5 million people in the world have developed atrial fibrillation (AF), and an estimated 30% of patients with AF are unaware of their diagnosis (silent AF).

Objective: The purpose of this study was to test a new technology for contactless detection of AF based on facial video recordings.

Methods: The proposed technique uses a camera to record an individual's face and extract the subtle beat-to-beat variations of skin color reflecting the cardiac pulsatile signal. In a group of adults referred for electrical cardioversion, we recorded the ECG and the video of the subjects' face before and after electrical cardioversion. We extracted the beat-to-beat pulse rates expressed as pulses per minute (ppm) from the videoplethysmographic (VPG) signal acquired using a standard web camera. We introduce a novel quantifier of pulse variability called the pulse harmonic strength (PHS) and report its ability to detect the presence of AF.

Results: Eleven subjects (8 male; age 65 ± 6 years) were included in the study. The VPG and ECG-based rates were statistically different between the AF and sinus rhythm periods: 72 ± 9 ppm vs 57 ± 7 ppm (P < .0001) for VPG and 80 ± 17 bpm vs 56 ± 7 bpm (P < .0001) for ECG signals. Among the 407 epochs of 15 seconds of synchronized ECG and VPG signals, PHS was associated with a 20% detection error rate, and the error rates of the automatic ECG-based measurements ranged between 17% and 29%.

Conclusion: Our preliminary results support the concept that contactless video-based monitoring of the human face for detection of abnormal pulse variability due to AF is feasible.