Multilayer Perceptron Network of ECG Peaks for Cardiac Abnormality Detection

Abstract

The inception of artificial neural networks (ANNs) was predicated on computational adaptations of human biology, specifically the fundamental principles underlying neurons. The feasibility of utilising ANN for diverse problem domains has been extensively investigated, with a particular emphasis on the domain of biomedical engineering. Applications of ANN are commonly employed in the fields of medicine and education for decision-making purposes. The ANNs employed in the present investigation were trained to identify cardiac anomalies by utilising a diverse set of reference data. The input parameters utilised for cardiac difficulties are commonly known as reference parameters, specifically pertaining to the amplitude and duration of the electrocardiogram (ECG) signal. The ECG complex is composed of three distinct components: the P peak, the QRS wave, and the T peak. The artificial neural network is provided with six input parameters, which are obtained by measuring the amplitude and length of each P peak, QRS wave, and T peak. The present study utilises a multilayer perceptron (MLP) as the structure for the ANN. This study examines the impact of the Tansig and Purelin activation functions on the structure of the MLP. All other networks were not as good as the MLP network, which got the best performance of 96.32% by using the BayR training method and the Tansig activation function.