Identification of Spatiotemporal Dispersion Electrograms in Persistent Atrial Fibrillation Ablation Using Maximal Voltage Absolute Values

Abstract

Atrial fibrillation (AF) is a sustained arrhythmia whose mechanisms are still largely unknown. A recent patient-tailored AF ablation therapy is based on the use of a multipolar mapping catheter called PentaRay. This new protocol targets areas of spatiotemporal dispersion (STD) in the atria as potential AF drivers. However, interventional cardiologists localize STD sites visually through the observation of intracardiac electrograms (EGMs). The present work aims to automatically characterize ablation sites in STD-based ablation. Recent research suggests that the distribution of the time series of maximal voltage absolute values at any of the PentaRay bipoles (VAVp) is affected by the STD pattern. Motivated by this finding, we consider VAVp as a key feature for STD identification. To our knowledge, this work applies for the first time statistical analysis and machine learning (ML) tools to automatically identify STD areas based on VAVp time series. Experiments are first conducted on synthetic data to quantify the effect of STD pattern characteristics (number of delayed leads, fractionation degree and number of fractionated leads) on engineered features of the VAVp time series like kurtosis, showing promising results. Then these features are tested on a real dataset of 23082 multichannel EGM signals from 16 different persistent AF patients. Statistical features like kurtosis and distribution (histogram) of VAVp values are extracted and fed to supervised ML classifiers, but no significant dissimilarity is obtained between the two categories. The classification of raw VAVp time series is finally conducted using ML tools like a shallow convolutional neural network combined with cross validation and data augmentation, reaching AUC values of 96%.