Diagnosis of prosthetic heart valve using locality preserving kernel fisher discriminant analysis and local discriminant bases

Auscultation, a method to detect the condition of heart by examining the heart sounds, is widely used by cardiologists. Using artificial intelligence methods in auscultation to detect various heart diseases is increasing in present days. In this paper, we try to classify 5 different categories of mechanical artificial heart valve sounds. Considering that such classification task is highly nonlinear, a new feature extraction algorithm, which is based on locality preserving kernel Fisher discriminant analysis and local discriminant bases (LDB), is proposed to improve the classification accuracy. All the tests are carried on a dataset that consists of 271 heart sounds. When the features extracted by the proposed method are fed into a normal linear discriminant function based (LDF) classifier, the correct classification rates can reach up to 95.6%.