Heart sounds analysis using wavelets responses and support vector machines

Over the last decade, computerized heart screening techniques have been increasingly receiving attention. In general, one can say that such techniques can be categorized as: with, or without the so-called Electrocardiogram (ECG) signal. Considering this latter strategy, we devote this paper with the intention to design an algorithm that provides with heart sounds known as Phonocardiograms (PGC) investigation for further definition of the present pathology if any. A novel algorithm for heart sounds segmentation is also presented. The decision making is accomplished by means of support vector machines (SVM) classifier which is fed by characteristic features extracted from PCGs basing on wavelet filter banks coefficients so that PCG signals are classified into five classes: normal heart sound (NHS), aortic stenosis (AS), aortic insufficiency (Al) mitral stenosis (MS), and mitral insufficiency (MI). The SVM was trained on a low-dimensional feature space, and tested on relatively a big dataset in order to show its generalization capability.