Diagnosis of aortic valve stenosis based on PCG signal using wavelet packet decomposition (WPD) and parametric models

Our aim in this study was to diagnose aortic valve stenosis from PCG signals using methods of converting the wavelet packet and statistical parameters. For categorization of three subcategories, K-nearest neighbor and multi-layer perceptron were used. After applying the proposed method on the PCG signals, the expected results were obtained. The highest classification accuracy for the features that were combined with the principal component analysis method and given to the nonlinear kernel classifier with a non-linear kernel of order 4 was 94.48% (without the principal component analysis method) and 98.38% (by the principal component analysis method). Therefore, the results of the classification show that our proposed method has been able to distinguish with the high ability of the phonocardiogram signals of a person with aortic valve stenosis and a healthy person. The accuracy of the multilayer perceptron and the nearest neighbor of K were also found to be 98% in detecting the stenosis of the aortic valve from the other. In fact, after reducing the number of features with PCA (composition of features), the classification accuracy of the diagnosis of aortic valve stenosis was increased by 5.13% and 2.96%, respectively, using the KNN and MLP classifications. Also, after reducing the number of features with the PCA (feature combination), the SVM classifier with a nonlinear kernel was 2.91% and for the linear kernel increased by 15.42%.