Noninvasive One-Year Ablation Outcome Prediction for Paroxysmal Atrial Fibrillation Using Trajectories of Activation From Body Surface Potential Maps

Abstract

Almost 40% of paroxysmal atrial fibrillation (AF) patients experience arrhythmia recurrence within a year after initial ablation success. The rich spatiotemporal information provided by body surface potential maps (BSPMs) can reveal AF dynamics. We hypothesised that the dipole direction of the heart during AF can be traced by the centroid trajectory of the principal "activated" electrode patch from the BSPM, where an electrode is defined as "activated" when its signal exhibits a local peak. This hypothesis was first verified using simulated and patient data, indicating that the trajectory has a high correlation with atrial electrical activity. The trajectory was then used as a spatiotemporal feature to predict one-year AF recurrence (22 negative and 23 positive) after ablation among 45 paroxysmal AF patients. The trajectories were segmented according to AF cycles for prediction in a multiple instance classification framework, using a Gaussian mixture regression (GMR) and a linear support vector machine (SVM) with L1 penalty for classification. A leave-one-out test showed 0.73 accuracy, 0.70 sensitivity and 0.77 specificity, and the area under the curve (AUC) of the receiver operating characteristic (ROC) as 0.84. The work suggests that with the proposed trajectory extracted from the BSPM, the prediction for paroxysmal AF ablation follow-up could be improved.