Detection of Morphological Characteristics of Atrial Fibrillation Using Semantic Segmentation

In this study, we propose a semantic segmentation-based atrial fibrillation (AF) detection method that learns and classifies the morphological characteristics of fibrillation waves (F-waves). F-waves are symptoms accompanying AF. The proposed method has higher localization performance for the target shape than existing methods; therefore, it can accurately detect when and how long AF occurs. We designed and experimented with two models: a binary classification model (BCM) that classifies only F-waves and a multiclass classification model (MCM) that additionally classifies P, QRS, and T information. We compared the performance of the method using the two proposed models and the rule-based algorithm using the RR interval, a major clinical feature of AF. The results show that the proposed method exhibits higher performance in AF event detection and localization, and the MCM-based method performed slightly better than the BCM-based method.