The analysis and classification of phonocardiogram based on higher-order spectra

Abstract

This paper investigates the application of a non-Gaussian AR model and parametric bispectral estimation in analyzing normal and pathological heart sound signals. The non-Gaussian AR model of PCG signals (phonocardiogram) is used to detect quadratic nonlinear interactions and to classify the two patterns of phonocardiograms in terms of the parametric bispectral estimate. The bispectral cross-correlation is proposed for the order determination of the model. Real PCG data are implemented to show that the quadratic nonlinearity exists in both normal and clinical heart sounds. It was found that parametric bispectral techniques are effective and useful tools in analyzing PCG and other biomedical signals, such as EMG, ECG and EEG.