Paroxysmal atrial fibrillation prediction using Kalman Filter

In this paper, we proposed a method based on Kalman Filter for predicting the onset of paroxysmal atrial fibrillation (PAF) from the electrocardiogram (ECG) using clinical data available from the Computers in Cardiology (CinC) Challenge 2001. To predict PAF, we developed an algorithm based upon the number of atrial premature complexes (APCs) in the ECG. The algorithm detects classical isolated APCs by monitoring fidelity signals, which is defined here as a function of the innovation signal of Kalman filter, in vicinity of premature heartbeats and decides whether one beat is APC or not then predicts PAF, based on the number of APC. The challenge database consists of 56 pairs of 30-minute ECG segments that may or may not directly precede an episode of PAF. We used the learning set of the challenge database to optimize our algorithm. On the test set, it achieved 50 out of 56 for PAF prediction and thus predicted the onset of PAF more accurately than the methods reported at CinC challenge.