An efficient abnormal beat detection scheme from ECG signals using neural network and ensemble classifiers

This paper presents an investigation into the development of an efficient scheme to detect abnormal beat from lead II Electro Cardio Gram (ECG) signals. Firstly, a fast ECG feature extraction algorithm was proposed which could extract the locations, amplitudes waves and interval from lead II ECG signal. We then created 11 customized features based on the outputs of the feature extraction algorithm. Then, we used these 11 features to train an artificial neural network and an ensemble classifier respectively for detecting the abnormal ECG beats. Three manually annotated databases were used for training and testing our system: MIT-BIH Arrhythmia, QT and European ST-T database availed from Physionet databank. The results showed that for an abnormal beat detection, the neural network classifier had an overall accuracy of 98.73% and the ensemble classifier with AdaBoost had 99.40%. Using time domain processing approach, the proposed scheme reduced overall computational complexity as compared to the existing methods with an aim to deploy on the mobile devices in the future to promote early and instant abnormal ECG beat detection.