An automatic ECG processing algorithm to identify patients prone to paroxysmal atrial fibrillation

Abstract

An algorithm to identify patients prone to paroxysmal atrial fibrillation (PAF) has been developed and evaluated using the PAF Prediction Challenge Database. The procedure is based on conventional electrocardiogram (ECG) signal pre-processing techniques for beat detection and classification, a correlation-based assessment of the P-wave morphology of both regular and premature heart beats of supraventricular origin, and a statistical test to calculate the PAF predictive parameter, i.e. the probability that a certain degree of P-wave variability is associated with potential triggers for PAF. This probability, finally, is used to differentiate between patients with and without PAF (screening) and to find out which of the two recordings of each patient immediately precedes the onset of PAF (prediction), respectively. The obtained diagnostic accuracies of 82% and 84%, respectively, indicate that this concept may be useful in terms of clinical PAF risk stratification.