ABSTRACT
Ventricular-based arrhythmias are a major cause of heart attacks and sudden cardiac deaths. Accurate detection, classification and prediction of these arrhythmias play a key role for the earlier prediction of heart attacks. The objective of the proposed method is to detect the non-visible (µV) level T-Wave Alterations (TWAs) which is the major cause of sudden cardiac death. Extracted morphological and statistical features from the denoised ECG signal are used to classify six different chronic ventricular arrhythmias. The performance of the classifier is validated with a supervised neural network (NN): Probabilistic Neural Network (PNN), Extreme Learning Machine (ELM), Self-adaptive Resource Allocation Network (SRAN) Classifier and Back Propagation Neural Network (BPNN). Among the four different NNs, the results in terms of accuracy, i.e., 99.8%, obtained clearly indicate a high degree of agreement with the PNN classifier with improved reliability and decreases structural complexity.
