An automatic neural-network based SVT/VT classification system

Abstract

Describes a novel automatic ECG rhythm analysis system for the problem of classifying between normal sinus rhythm (NSR), supraventricular tachycardia (SVT) and ventricular tachycardia (VT). The system comprises two stages-a preprocessing stage and a neural network based classification stage. The preprocessing stage performs feature vector extraction from multi-leaded ECG sources. Key temporal (morphological), spatial (inter-lead) and spectral (frequency) features are used to form the feature vectors. The neural network classifier comprises a multi-layer perceptron trained using the backpropagation algorithm. By fusing features from the spectral and temporal domains, 100% classification is again possible.<>