Classification of Cardiac Arrhythmias Based on Wavelet Transform and Neural Networks

Abstract

Cardiac diseases are one of the most common causes of death, killing millions of people worldwide each year. However, they can be effectively controlled by early diagnosis. Electrocardiograph is the most important and powerful reference tool used to diagnosis and treatment of heart diseases, it represents the electrical activity of the heart and contains vital information about its rhythmic characteristics. This study was built to design computationally efficient models and an intelligent tool for diagnosis of ECG abnormalities with high accuracy while reducing the complexity, cost, and response time of the system and contributing to solve the problem of lacking of physician in rural area. The ECG signals obtained from MIT-BIH database; forty signals used for training and testing. De-noising processing was applied to power line interference and baseline wanders to facilitate accurate detection of features. Symlet wavelet transforms was selected as mother wavelet to feature extraction and pattern recognition Neural Net Works was used as classifier. The extracted features consisted twenty four features, it contain both morphological and statistical features of the signals, and is used as an inputs pattern recognition neural net work with ten hidden neuron layer, which identify normal ECG and four different types of arrhythmias, which are Paced, Atrial Premature, Ventricle Premature contraction and Left Block Bundle Branch. The performance of the proposed method has been evaluated in terms of accuracy, sensitivity and specificity. The accuracy of classification for each class are normal class is 100%, paced class is 90%, AP class is 73%, PVC class is 95% and for LBBB class is 100%.