Classification of Segmented ECG Signals Using Multi-Layer Perceptron Classification Algorithm

Abstract

Heart disease is a condition which affects the cardiovascular system. Heart disease is the major global cause for the increase in the death ratio. Research related to heart disease prediction reflects that heart disease can be identified in earlier stage as it is predictable. ElectroCardioGram (ECG) is the primary, basic and painless method for monitoring and detecting the functioning of the heart. This paper is an attempt to retrieve ECG signals and remove unnecessary noises, segment the signals and classify the segmented signals. This classification will pave way to detect the heart disease in the earlier stage itself. The recorded ECG signals are extracted and preprocessed using Butterchev algorithm and the heart peaks are identified in the first stage. After identifying the Heart peaks the signals are segmented which represents the elements of ECG Signal such as P,Q,R,S,T and U. The temporal, spectral and fiducial ECG features are extracted from the segmented ECG signals for classification purpose. The extracted feature vectors are utilized to classify the signals. Multi-Layer Perception (MLP) algorithm is used to classify the extracted ECG features into five primary classes based on gender. Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) Arrhythmia database and Noise Stress database are used for this implementation and the classification classes are classified based on the given dataset parameters. Performance metrics such as accuracy, specificity and sensitivity are computed to identify the performance of the classification algorithm. This MLP based classification serves as the source to classify the segmented ECG signals which can be enhanced with deep learning methodology.