Session P86.4

Detecting and Discriminating Premature Atrial and Ventricular Contractions: Application to Prediction of Paroxysmal Atrial Fibrillation

A Ghaffari, MR Homaeinezhad*, M Akraminia, M Atarod, Y Ahmadi

K. N. Toosi University of Technology
Tehran, Iran

In this study, a new method is addressed to predict the onset of paroxysmal atrial fibrillation based on comprehensive analysis of RR-tachogram and classified P-waves. First, using Trous discrete wavelet transform and applying appropriate pre-processing on the second or third scales of the transformation, electrocardiogram (ECG) signal is completely delineated. Then, using suitable simulated features, detected P-waves are classified to five categories namely as "deep minus", "weak minus", "zero vicinity", "weak positive" and "strong positive". To classify P-waves, an appropriate measure is calculated for each wave based on corresponding third scale signal and a quantity is attributed to each wave. Next, using a proper Kernel Machine (Kernel Fisher Discriminant), P-waves are classified. Accordingly, the detected QRS complexes are classified into "Normal", "PVC" and "PAC" complexes and local and global bigeminy test is then conducted on PACs. Investigations show that the numbers of PACs, atrial bigeminy or trigeminy accompanied by paroxysmal atrial tachycardia are proper indicators for prediction of PAF. The proposed algorithm is finally applied to PhysioNet PAF Database (CinC Challenge 2001) and up to now, values of 92.8% and 94.3% are attained for sensitivity and positive prediction, respectively.

(Abstract Control Number: 78)