Recent progress and technical advances in catheter ablation have dramatically improved the success rate and safety of pulmonary vein isolation (PVI) for atrial fibrillation (AF). AF recurrences occur predominantly because of reconnections in previously isolated pulmonary veins. The aim of this study is to analyze an acute pulmonary vein reconnection (PVR) using graph convolutional networks (GCN). The subjects of the patient population survey were 30 patients (20 males and 10 females), who underwent ablation for AF from May 2019 to March 2020. The average age of patients is 65.3 years old (from 53 to 89 years old). The targeted atrial fibrillation is paroxysmal atrial fibrillation (PAF), persistent atrial fibrillation (PeAF), and chronic atrial fibrillation (CAF). The target samples included 11 cases of PVR (spontaneous PV-LA reconnection or dormant conduction (DC)) and 19 cases of non-PVR. The feature parameters analyzed in this study are as follows: Location X, Y, Z, Duration time (DT), Average force (A-Force), Max temperature (MT), Max power (MP), Base impedance (BI), Impedance drop (ΔImp), Force time integral (FTI), Ablation index (AI), Unipolar (Uni), Bipolar (Bi). We conducted a comparative study of predicting the AF recurrences using GCN and other popular machine learning methods. As a result, we found that GCN has better prediction accuracy than the other methods. Its prediction accuracy still needs to be increased. However, in this field, our study is the first trial of integrating the location of the cauterization points and the various indices with GCN. It is expected that GCN will be very useful in predicting the AF recurrences during radiofrequency catheter ablation (RFCA).
抄録全体を表示