Automatic classification of Cardiac Ischemia through Principle Component Analysis and Artificial Neural Network

Abstract

Accurate detection of cardiac ischemia is very crucial stage of Electrocardiogram (ECG) for diagnosis and treatment of cardiac arrhythmia. ECG when done manually can take a lot of time if there is a large amount of data available which is not feasible given the crucial nature of the disease. Due to this issue we need automatic classification for the processing and analysis of the ECG data. This research proposes usage of ECG features and classify those features through Artificial Neural Network (ANN) for detection of normal and abnormal heart signals. Feature set of different features based on time intervals and different frequency bands is to be extracted through QRS complex, Fast Fourier Transform, Discrete Wavelet Transform and Wigner-Ville Transform. Furthermore Principle Component Analysis (PCA) is used for reduction of features without losing any information extracted from the dataset. Those PCA reduced features are classified through ANN into normal and ischemic heart signals. The proposed model provides accuracy in the range of 80 percent. The results shows feasibility of the mechanism in order to develop an application which can automatically detect heart arrhythmia on the basis of ECG signal. This work is primarily focused on the classification of cardiac abnormalities on the basis of both (morphological and spectral features) through PCA and ANN. To the best of our knowledge this is a novel model. This model is applied on Physiobank dataset which is globally used for research purposes and it provides satisfactory results.