Clustering and classification applications for ventricular focal cardiac arrhytmia source localization

Catheter-based approaches used in determining the regions of abnormal conduction causing rhythm disturbances on the heart have recently become popular due to their minimal invasive nature. Another noninvasive method, body surface potential mapping (BSPM), strives for staying not only as a research tool but becoming a clinically applicable approach.In this thesis, feasibility and performance of clustering and classification approaches in the localization of the source of focal ventricular arrhythmias using catheter-based or BSMP-based techniques have been investigated. For this purpose, the measurement points on the heart surface have been clustered by Kohonen self organizing maps (SOM) and fuzzy C-means (FCM). Finally, back-propagation, probabilistic neural networks (PNN), discriminant analysis (DA), k nearest neigbor (kNN), support vector machines classification approaches have been studied in determination of the cluster (region of the heart) from which the arrhythmic beat originates. In the first phase of study, outer surface (epicardial) mapping have been performed using a subset of activation times that coming from catheters inside coronary veins. In the second stage, the simulation data set based on Aliev-Panfilov model that was developed for the mathematical modelling of the electrical activity of the heart was used for the determining the source of the arrhythmias on both inner and outer surfaces of the heart. the third stage, the feature vectors have been selected using the Karhunen-Loeve Transform (KLT) and QRS-QRST integral calculations on the body surface potential maps. Then these vectors have been used for determining epicardial sources. As a result of this study, new methods that detect the source of the ventricular arrhythmias have been developed.