QRS morphological analysis using two layered self-organizing map for heartbeat classification

QRS morphological analysis in Holter electrocardiography has been developed using correlation coefficient methods. However, the accuracy of automated classification for QRS complexes, does not fully satisfy the clinical needs. In this paper, we propose a two-layered classification using self-organizing map (SOM). In the first layer, each beat is divided in sections. The average level, height, amplitude of the peak, maximum and minimum slope are calculated as the characteristics of the section. By learning these characteristics in the first SOM, the sections are classified in qualitative attributes. In the second layer, QRS complexes are reconstructed as a line of the qualitative attributes and classified by the second SOM. We evaluated our method using MIT-BIH arrhythmia database and compared it with the accuracy of a standard cross correlation coefficient method. The classification error rate of the correlation coefficient method and proposed method is 0.75% and 0.41% respectively. We confirmed that the accuracy in our method for the QRS complex analysis has significantly improved.