Higher-order statistics: a robust vehicle for diagnostic assessment and characterisation of lung sounds

Abstract

The use of higher-order statistics for diagnostic assessment and characterisation of lung sounds is presented in this article. The parametric approach of bispectrum estimation, which is a third-order spectrum, based on a non-Gaussian white noise driven autoregressive (AR) model, reveals information about lung sounds that is not contained in the ordinary power spectrum, such as the degree of nonlinearity and deviations from normality. Characterisation of source and transmission of lung sounds is achieved using an AR model based on third-order statistics. Furthermore, harmonic analysis of lung sounds is combined with the bicoherence index in order to obtain information regarding possible quadratic phase coupling among harmonic components of musical lung sounds. Experiments have shown that higher-order statistics can offer reliable evaluation of lung sounds characteristics, since their general properties and robustness in noiseless or noisy environments (lung sounds contaminated with additive symmetrical noise, e.g., Gaussian) proved to have superior advantages in objective analysis of pulmonary dysfunction.