computer methods and programs in biomedicine 94 ( 2 0 0 9 ) 223–231 journal homepage: www.intl.elsevierhealth.com/journals/cmpb PVC discrimination using the QRS power spectrum and self-organizing maps M.L. Talbi * , A. Charef Laboratoire de traitement du signal, Département d’électronique, Université Mentouri de Constantine, Constantine 25000, Algeria article info Article history: Received 5 February 2008 Received in revised form 27 December 2008 Accepted 30 December 2008 Keywords: ECG QRS Power spectrum PVC SOM abstract This paper deals with the discrimination of premature ventricular contraction (PVC) arrhyth- mia using the fractal behavior of the power spectrum density of the QRS complexes. The linear interpolation of the QRS complex power spectrum density in Bode diagram in two different frequency intervals gives two straight lines with two different slopes. The scatter plot of one slope versus the other shows that there exists two distinct regions which rep- resent the normal beats and the PVC beats. Therefore the PVC beats are classified using a self-organizing map fed by the two slopes of the QRS complex power spectrum. The MIT/BIH arrhythmia database is then used to evaluate the usefulness of the proposed method in the discrimination of the premature ventricular contraction (PVC) arrhythmia. The results have indicated that the method has achieved 82.71% of sensitivity and 88.06% of specificity over 46 records from the MIT-BIH arrhythmia database. © 2009 Elsevier Ireland Ltd. All rights reserved. 1. Introduction Cardiovascular diseases are still a major cause of mortality around the world. A report published by the World Health Organization states that an estimated of 17 million people die from cardiovascular diseases each year [1]. Arrhythmias represent one of the serious heart diseases and ventricular arrhythmias are the most life threatening. Premature ven- tricular contraction (PVC) is an arrhythmia caused by the existence of ectopic centers in the ventricles that changes the path propagation of the activation front and leads to genera- tion of QRS complexes with wide and bizarre waveforms. The PVC waveforms can also be uniform or multiform for the same patient besides they represent a lot of variations from patient to another. Many studies have shown that PVCs, when asso- ciated with other heart diseases such myocardial infarction, can be linked to increased mortality [2,3], consequently their immediate detection and treatment is essential for patients with heart diseases. Hence an automatic detection and a quick ∗ Corresponding author. Tel.: +213 661293451. E-mail address: mltalbi@yahoo.fr (M.L. Talbi). and reliable identification and classification of these condi- tions constitute a challenge for a cardiovascular diagnostic system and a considerable importance in critical care. In the last decades, significant amount of research work for automatic detection and classification of PVC beats have been done. Some methods are simples they have been developed for the discrimination between normal and PVC beats only and some are more complex methods, they have developed to clas- sify several arrhythmias into different classes or clusters at the same time. Because of our proposed approach is validated in the MIT-BIH database [4] only some of the recent methods using this database have been considered for description and comparison. Wieben et al. [5] have developed a classifier based on filter bank features and decision trees. The algorithm has achieved a sensitivity of 85.3% and a positive predictivity of 85.2%. Using only 14 records of the MIT-BIH database, the classi- fier based on neural networks presented by Al-Nashash [6] has achieved a sensitivity of 98.1% and a positive predictiv- 0169-2607/$ – see front matter © 2009 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.cmpb.2008.12.009