Prediction of Paroxysmal Atrial Fibrillation using Empirical Mode Decomposition and RR intervals

In this paper, we proposed a method based on time-frequency dependent features extracted from Intrinsic Mode Functions (IMFs) and physiological feature such as the number of premature beats (PBs) to predict the onset of Paroxysmal Atrial Fibrillation (PAF) by using electrocardiogram (ECG) signal. To extract IMFs, we used Empirical Mode Decomposition (EMD). In order to predict PAF, we used variance of IMFs of signals, the area under the absolute of IMF curves and the number of PBs, since increasing of all of these parameters are a clear sign of PAF occurrence. We used clinical database which was provided for the 2001 Computer in Cardiology Challenge (CinC). The test set of this database consist of 28 pairs of 30-minute ECG segments that may or may not directly precede an episode of PAF. We used the training set of this database to optimize our algorithm. By applying our method on test set, we manage to predict 25 out of 28 pairs correctly which is 11 percent better than the challenge winner result.