Premature atrial complexes detection using the Fisher Linear Discriminant

Currently, no reliable method exists to detect premature atrial complexes (PAC). The detection of PACs is clinically essential to predict supraventricular tachycardia, postoperative atrial fibrillation and paroxysmal atrial fibrillation. We propose an algorithm for intra-class classification that includes an analysis of the R-R time series. In the pre-processing phase, we used Butter worth filters to remove the baseline wander and the other noise. In the feature extraction phase, we detected the RR interval duration and the distance between the occurrence of P wave and T wave. Using these features we applied Fisherpsilas Linear Discriminant to create a criterion that can be used for classification. Combining pre-processing, feature extraction and Fisherpsilas Linear Discriminant we succeed in separating Normal and PAC beats with 99% Accuracy.