Classification of single-lead ECG signal to normal and abnormal subjects using histogram-based features

This paper presents an inter-patient classification method for classifying single-lead Electrocardiography (ECG) signals into normal and abnormal classes using histogram-based features. This method is composed of two stages; namely, the lead selection stage and the classification stage. In the lead selection stage, a small set of normal and abnormal ECG signals are selected and subsequent histograms are estimated for each class. The discriminative bins in each lead are specified using the Two-sample Kolmogorov-Smirnov test function. Based on these bins, the lead's importance is estimated and the most important lead is selected. These bins of the most important lead are used as feature indices in the classification stage. Three classifiers; Naïve Bayes (NB), Simple Cart (CART), and Naïve Bayes Multinomial (NBM) methods were used in this study. The proposed approach was evaluated using 104 subjects (52 normal and 52 abnormal) from the Physikalisch-TechnischeBundesanstalt (PTB) dataset and achieved a considerably high sensitivity of 96%, a specificity of 92%, and an accuracy of 94% using the CART classifier system.