Neural wavelet analysis of life threatening ventricular arrhythmias

Abstract

One of the most important task which can be implemented by an automatic monitor of cardiac arrhythmias is the reliable detection of those arrhythmias which represent a serious threat to the patient. Among these, ventricular arrhythmias occupy a primary place, and in particular ventricular fibrillation (VF), ventricular tachycardia (VT) and atrial fibrillation (AF) because of the haemodynamic deterioration which they produce. Consequently, interest has arisen in the development of algorithms which could be transferred easily to a microprocessor system. We use the backpropagation training (BP) algorithm on wavelet transformed results to classify the three mentioned arrhythmias. The BP algorithm perform the gradient descent search to reduce the mean square error between the actual output of the network and the desired output through the adjustments of weights. The results show significant improvements in the sensitivity (95%) and specificity (92%) over previous studies.