Methods for detection of real time ventricular arrhythmias using hybrid features of ecg signals

Abstract

newline Ventricular arrhythmias (VAs) such as ventricular tachycardia (VT) and ventricular newlinefibrillation (VF) are the most life-threatening cardiac arrhythmias and they often cause newlinesudden cardiac death. These high-risk cardiac arrhythmias, if identified timely, can prevent newlinethe sudden death with an application of a defibrillator. Hence a computer-aided detection newlinesystem is essential for precise detection of VAs in clinical practice. In this thesis, different newlinemethods have been implemented for pre-processing, signal decomposition, feature extraction, newlineand classification of ventricular arrhythmias using ECG signals. The desired ECG signals newlinehave been acquired from CUDB and VFDB databases of PhysioNet repository. Signals are newlinede-noised with digital filters and the filtered signals are decomposed with techniques such as newlinediscrete wavelet transform (DWT), ensemble empirical mode decomposition (EEMD) and newlinevariational mode decomposition (VMD). A set of 24 time-frequency based features has been newlineextracted and ranked in gain ratio attribute evaluation in order to improve the detection newlineaccuracy. newlineThe feature set was classified by two different classifiers i.e. support vector machine newline(SVM) and decision tree (C4.5) algorithm for precise detection of normal sinus rhythm newline(NSR), VT, and VF rhythms. Initially, a set of thirteen time-frequency based fea