Predicting spontaneous termination of atrial fibrillation based on the RR interval

It is important to characterize conditions under which atrial fibrillation (AF) is likely to terminate spontaneously. A novel method is proposed here. Eleven features are first extracted to characterize RR interval and Poincaré plot from a statistical viewpoint and a geometric viewpoint respectively. Then sequential forward search (SFS) algorithm is utilized for feature selection. Finally, a fuzzy support vector machine (FSVM) with a new fuzzy membership is applied for AF termination prediction. The method is studied with an AF database of electrocardiogram (ECG) recordings provided by PhysioNet for the Cardiology Challenge 2004. It is divided into a training set and two testing sets (A and B). Experiment results show that 100 per cent of testing set A and 100 per cent of testing set B are correctly classified, together with 92.3 per cent of non-terminating and soon-terminating AF correctly classified. It demonstrates that the proposed method can predict spontaneous termination of AF effectively.