Identification of recurring wavefront propagation patterns in atrial fibrillation using basis pursuit

High density contact electrogram data of atrial fibrillation (AF) contain detailed information on recurring activation patterns and dominant signaling pathways. Current methods to analyze these patterns and pathways rely mainly on supervised atrial deflection annotation and wave reconstruction. In this study, we developed a new algorithm to automatically identify recurring patterns and dominant pathways without the need for annotation. A sparse multivariate autoregression model was estimated on short segments of synchronous unipolar electrograms to extract the dominant interactions between electrograms at different recording electrodes. Sparsity of the electrode interaction matrices at several time-lags was maximized by applying a distance-weighted basis pursuit algorithm. Dominant interactions were identified by computing the mean interaction matrix over a number of consecutive time segments. The algorithm was evaluated on high-density recordings with 234 electrodes and 2.4mm electrode spacing in the left and right atrial free wall of a goat model of AF. The method was able to identify relevant patterns of AF, including wave trains, repetitive breakthrough waves and rotating wave activity.