Development of Rhythm-based and Morphology-based Algorithm for Atrial Fibrillation Detection From Single Lead ECG Signal

Atrial Fibrillation or AF is the most common atrial arrhythmia and often occurs especially in the elderly people. Since AF is associated with an increased risk for stroke, regular AF monitoring is recommended for the elderly as well as the patients to reduce the incidence of stroke. In this study, the subsequent two methods of AF detection have been. The first one is a low complexity algorithm with rhythm-based for embedding into portable ECG devices. The second one is morphology-based method using a transferred deep learning with fine-tuning. The performances of the low complexity algorithm are 100% of sensitivity and 86.67% of specificity, respectively while the performances of the deep learning algorithm are 96.97% of sensitivity and 100% of specificity, respectively. From these results, the combination between rhythm-based and morphology-based algorithm can offer the highest sensitivity for screening of AF and highest specificity for separating Normal Sinus Rhythm (NSR) from AF. These developed methods will be useful for improving the performance of AF detection utilized in healthcare system.