Detecting beats in the ECG: A comparison of time domain and morphological features using Support Vector Machines and MultiLayer Perceptron

Aunsa Shah, Seral OzÅŸen, Abbas Shah

The ElectroCardioGraph (ECG) is the most widely used diagnostic test for determining heart related disease prognosis. This paper presents a comparison of two types of feature extraction methods and two types of classifiers for the detection of four types of heart beats in the ECG. The four types of heart beats considered in this work are Normal, Right Bundle Branch Block Beat, Left Bundle Branch Block Beat and the Premature Ventricular Contraction beat. The first set of features computed for each beat type are statistical in nature in the time domain and the second set of features are morphological in nature.  The values of the features in these two sets are then sent to two different classification algorithms, the Support Vector Machine (SVM) and the MultiLayer Perceptron (MLP) Neural Network. The classification results demonstrate that when comparing the chosen set of statistical and morphological features, the statistical values of each beat provide a higher detection accuracy for all beat types. Furthermore, it was also observed that when comparing the performance of the SVM and MLP algorithms for heart beat classification, the MLP was found to outperform SVM when using statistical features and when both feature sets were combined, however, the opposite was observed when only morphological features were used in which case, the SVM outperformed the MLP network.