Classification of ECG arryhtmia beats with artifical neural networks

Abstract

Electrocardiography (ECG) is very useful noninvasive imaging method of the heart's electrical activity. Based on these recordings, a wide range or heart conditions can be diagnosed. These conditions may vary from minor to life threatening ones. Therefore, the scientists started to work on automatic systems that would detect any kind od abnormalities in the heart's electrical activity. These autome-ated systems are expected to help patients monitor themselves or the clinicians monitor their patients for any kind of abnormalities. With the help of these automated systems, there is a big contribution to early, quick and efficient diagnose of the heart diseases. Based on this need, this thesis presents an automated arrhythmia detection system. The classification of beats is performed in a Graphical User Interface, namely Patient Monitoring GUI. Based on the user's selection, the GUI displays the type of beats that flow on the screen. In the background, the GUI uses an Artificial Neural Network (ANN) trained to classify the 7 different types of arrhythmias. During the training process of ANN's the ECG recordings from MIT BIH Arrhythmia database are used as references. The arrhythmia samples are extracted from the database and preprocessed to create input sets to rain ANNs. The Fourier Transforms of a predefined window of signals were taken as a feature extraction method. The training was performed in multiple steps in order to obtain best performing ANN that will be finally used by the Patient Monitoring GUI. The training of the ANNs was performed by using the Neural Network Toolbox in Matlab 2008b and the results were recorded to track the difference between the training attempts. The overall success rate of the best performing ANN was measured as 80%.|Keywords: Artificial Neural Network, ECG, Classification, Neural Network Toolbox, Fourier Transform, Arrhythmia.