Classification of electrocardiogram (ECG) signals plays an important role in diagnoses of heart diseases. An accurate ECG classification is a challenging problem. This paper presents a survey of ECG classification into arrhythmia types. Early and accurate detection of arrhythmia types is important in detecting heart diseases and choosing appropriate treatment for a patient. Different classifiers are available for ECG classification. Amongst all classifiers, artificial neural networks (ANNs) have become very popular and most widely used for ECG classification. This paper discusses the issues involved in ECG classification and presents a detailed survey of preprocessing techniques, ECG databases, feature extraction techniques, ANN based classifiers, and performance measures to address the mentioned issues. Furthermore, for each surveyed paper, our paper also presents detailed analysis of input beat selection and output of the classifiers.

No abstract

An arrhythmia is an abnormality in the heart rhythm, or heartbeat pattern. ECG beats can be classified into six different arrhythmia beat types (left bundle branch block, right bundle branch block, paced beats, premature ventricular contradiction, atrial premature beats, and normal rhythm). Early and accurate detection of arrhythmia types is important in detecting heart diseases and choosing appropriate treatment for a patient. This paper proposes a combination of Particle Swarm Optimisation (PSO) and Feed Forward Neural Network (FFNN) for ECG beat classification. We have used MIT-BIH arrhythmia database for data collection and prepared three different datasets. Features, such as R peak sample number and QRS complex, are extracted using Pan-Tompkins algorithm. The extracted features are used as inputs to three different classifiers: Multi-Layer Perceptron Neural Network (MLPNN), Support Vector Machine (SVM), and PSO-FFNN. Results show high classification accuracy of over 97% with either of these three classifiers. The performance comparison of these classifiers is carried out using three measures: sensitivity, specificity, and accuracy. The results suggest that PSO-FFNN shows slightly better performance than MLPNN and SVM in terms of accuracy on all datasets.