心音信号的非高斯AR模型双谱分析

BISPECTRAL ANALYSIS OF HEART SOUND SIGNALS VIA NON-GAUSSIAN AR MODEL

本文利用高阶谱分析方法 ,提出以非高斯参数模型对心音信号进行双谱分析与分类。采用具有非高斯白噪声激励的参数模型对心音 (PCG)信号进行建模 ,由导出的基于累积量的三阶递推方程估计模型参数 ,并对心音信号进行参数化双谱估计。其次 ,对心音信号的非高斯AR模型的定阶进行讨论 ,提出采用双谱互相关法估计心音信号的模型阶次。另外 ,利用非高斯AR模型参数构成特征向量 ,对正常心音和病理性心音进行二类模式分类。最后 ,对实际心音信号进行分析 ,结果表明利用双谱技术分析心音信号是一种新的有效方法 ,有助于心音信号的定量分析并为某些心脏疾病的早期诊断提供新的辅助信息。

The heart sound signals were analyzed by applying the new signal processing tool: higher order spectral technique. Non Gaussian AR model was employed for the analysis and classification for PCG signals. First, AR model with non Gaussian white noise input was selected to fit the heart sound data. The third order recursion via cumulant and the parametric bispectral estimate for heart sound signals were also developed. Moreover, the non Gaussian AR model order determination was discussed. The method for AR model order selection based on bispectral cross correlation was proposed and applied to estimate the AR model order of the heart sound signals for parametric bispectral estimation. Finally, the characteristic parameters for pattern classification of heart sound signals were extracted by employing the parameters of the non Gaussian AR model. Both normal and pathological phonocardiograms were investigated for pattern classification. The results showed that the analysis and processing of the PCG signals based on bispectral technique were applicable and effective. The bispectral analysis of heart sound signals might be helpful to the quantitative analysis for the PCG signals and providing assistant information for early diagnosis of heart disease.