Classification of Congenital Heart Disease by Svm-Mfcc Using Phonocardiograph Publisher



Attarodi G¹ ; Tareh A¹ ; Dabanloo NJ¹ ; Adeliansedehi A²
Source: Computing in Cardiology Published:2017

Abstract

In this paper, a new method is presented for nonlinear processing and classification of congenital heart valve-septum diseases in neonates. Two main groups of congenital heart diseases in neonates are aortic valve stenosis, and inter-ventricular septum puncture. Both diseases create harsh sounds in the heart's first sound area, S1. In this study, using a compilation of MFCC (Mel-frequency cepstral coefficient) and Auto Correlation methods, we separated the 1st sound range with high precision and thereafter managed to classify three groups of neonates: with normal sound, murmur sound resulted by VSD (ventricular septum defect), and murmur sound resulted by AS(aortic stenosis) using SVM(support vector machine) classifier equipped with RBF (radial-basis function) and Quadratic kernels. With regard to the data distribution in the feature space that was based on short term energy of 32-fold time intervals of wavelet transform's level 2 coefficients with db kernel, the SVM-Quadratic classifier managed to classify the three groups of foresaid neonates with 78% precision and SVM-RBF classifier with 96% precision. © 2017 IEEE Computer Society. All rights reserved.