Developing time-frequency features for prediction of the recurrence of atrial fibrillation after electrical cardioversion therapy

External electrical cardioversion has been used as a therapeutic option to terminate atrial fibrillation (AF) and restore sinus rhythm (SR). However, identifying patients who would benefit from this therapy is still an active area of research. In this study, we develop new time-frequency features to characterize the atrial activity (AA) and to predict the success of electrical cardioversion therapy by identifying the AF patients who will maintain SR in the long term. New features are extracted from the surface AA using a matching pursuit (MP) decomposition with various combinations of wavelet families. The performance of the features is validated using a dataset of AF patients who underwent electrical cardioversion therapy. Results indicate that the developed features are significantly (p-value <; 0.05) correlated with SR maintenance which suggests that the MP decomposition captures detailed morphological information of AA that may potentially be used to guide the therapy of AF patients.