Low computation heartbeat classification based on ECG using artificial neural networks

Abstract

Electrocardiography (ECG) is an essential non-invasive tool for classification and diagnosing a cardiac arrhythmia. Many methods using ECG signals for the classification of different types or classes of heartbeats are proposed. These classes are started from normal beats or abnormal beats and reached many types based on the classification methods and the variability of database beat types. A low computation method with high accuracy is a challenge for many researchers in this field that can be implemented in a real-time monitoring system. The proposed classification algorithm is a low-computation, real-time, and high-performance technique based on artificial neural networks (ANN). In this work, the three stages algorithm is designed and evaluated for low computation applications. The first stage is filtering the original ECG signal, features extraction for QRS, and QRS-detection. In the second stage, the new features are extracted from RR- interval and added to the same QRS-detection features. Finally, A simple ANN is designed to classify the beats for two classes (normal and abnormal) based on the QRS shape used for detection and the RR-interval features. The algorithm is evaluated using MIT-BIH Arrhythmia Database and MATLAB software. The accuracy of 98.97%, recall of 99.42%, and high precision of 99.13% are promising results for a low computation algorithm. It is a new approach for real-time applications with low resources and high efficiency compared with the other classifier approaches.