Identification of atrial fibrillation episodes using a camera as contactless sensor

Identification of paroxysmal atrial fibrillation (AF) can be difficult and undiagnosed AF patients are at high risk of cardioembolic stroke or other complications associated with AF. The aim of this study is to analyze the video pho-toplethysmografic (vPPG) signal obtained from a videocamera to explore the possibility of discriminating AF from normal sinus rhythm (NSR) and other arrhythmias (ARR). We acquired 24 3-min long face-videos (8 for each rhythm) using an industrial camera. After preprocessing, vPPG signal was extracted using zero-phase component analysis. Diastolic minima were detected and inter-diastolic series obtained. The signals were characterized by time domain indexes, the sample entropy (SampEn); and the shape similarity index (ShapeSim). The time domain indexes and ShapeSim are significantly different when comparing the group of patients with AF or ARR to subjects in NSR. Sam-pEn is significantly higher in AF than in NSR and ARR. From the shape analysis, it can be noted that waves in NSR are more similar than in AF. These preliminary results show the capability of different indexes to capture differences among AF, ARR and NSR. Further studies will help in assessing the performance of the vPPG signal to screen general population.