Classification of heart disease based on PCG signal using CNN

Abstract

Cardiovascular disease is the leading cause of death in the world, so early detection of heart conditions is very important. Detection related to cardiovascular disease can be conducted through the detection of heart signals interference, one of which is called phonocardiography. This study aims to classify heart disease based on phonocardiogram (PCG) signals using the convolutional neural networks (CNN). The study was initiated with signal preprocessing by cutting and normalizing the signal, followed by a continuous wavelet transformation process using a mother wavelet analytic morlet. The decomposition results are visualized using a scalogram, then the results are used as CNN input. In this study, the PCG signals used were classified into normal, angina pectoris (AP), congestive heart failure (CHF), and hypertensive heart disease (HHD). The total data used, classified into 80 training data and 20 testing data. The obtained model shows the level of accuracy, sensitivity, and diagnostic specificity of 100%, 100%, and 100% for training data, respectively, while the prediction results for testing data indicate the level of accuracy, sensitivity, and specificity of 85%, 80%, and 100%, respectively. This result proved to be better than the mother wavelet or other classifier methods, then the model was deployed into the graphical user interface (GUI).