ECG feature extraction and classification for Arrhythmia using wavelet & Scaled Conjugate-Back Propagation Neural Network

Abstract

Electrocardiogram (ECG) plays an imperative role in heart disease diagnostics, Arrhythmia is a cardiological disorder with prevalence of the world's population..The purpose of this research to discusses the electrocardiogram analysis because this problems concerning health issue which encourage the present research. The main objective of our research is to analyze the acquired ECG signals using signal processing tools such as wavelet transform and Neuro classifier, classify them.. Total 62 ECG data subjects were analyzed. These data were grouped in two classes i.e, Normal class, and Arrhythmia class respectively. In order to achieve this we have applied a back-propagation based neural network classifier. DWT coefficients are used to extract the relevant information(statistical feature) from the ECG input data which are Energy, Variance, power Spectrul Density, Mean and Standard Deviation and morphological feature from the ECG input data which are mean and standard deviation of Magnitude of P,Q,R,S,T peak & PQ, QR,RS,ST,PR,QRS interval & RR interval. Then the extracted features data is analyzed and classified using Adaptive Neuro System (ANNS) as a Neuro classifier. The proposed algorithm is implemented and also tested in MATLAB software. The ECG signal are being selected and tested from PhysioNet Database using MIT-BIH Arrhythmia Database, and & Normal Sinus Rhythm (NSR) database. 14 subjects from Normal set and 48 subjects from arrhythmia set were analyzed for feature extraction and classification and data were divided in training, testing and validation of proposed algorithm. The ANNS system successfully classifies the Normal, Arrhythmia signal with the rate of overall accuracy is 98.4%. The analysis system also can achieved the sensitivity up to 100% for normal class and 97.9% for arrhythmia class, respectively for each class tested