Biomedical Research

Spectral or temporal features employed in computerized lung sound analysis should be statistically significant and capable to distinguish the subjects accurately into respective disease categories, regardless of their age, gender and demography. In this article the spectral features of five different classes of lung sounds; stridors, wheezes bronchial, vesicular and crackles are estimated via an automated method and evaluated for their statistical significance using Analysis of Variance (ANOVA) and Fisher’s Class Separability Measure (FCSM). The spectral features included in this study are median frequency, dominant frequency, maximum frequency, spectral roll off and spectral centroid. The maximum and dominant frequencies and spectral centroid are identified directly from the lung sound spectra. The median frequency and spectral roll off are computed from the Power Spectral Density (PSD) estimate using an analytical method. Before computing the spectrum, the lung sound specimens are preconditioned with offset elimination and normalization. The normalized specimen is windowed with Hanning window to suppress the ripples induced in the spectrum during the computation of Fast Fourier Transform (FFT). The pre-processing, estimation of the spectral features and their statistical evaluation are performed in Matlab®. P-values of 0.0386, 0.7508, 0.0197, 0.055 and 0.6979 were observed at a confidence level of 0.05, for dominant frequency, maximum frequency and median frequency, spectral roll off and spectral centroid, respectively. The values of FCSM are 0.1242, 0.0192, 0.1498, 0.1112 and 0.0222, respectively and in compliance with ANOVA. The median frequency comparatively is more significant than the other four. It is capable of discriminating the stridors and crackles.

Author(s): Nishi Shahnaj Haider, Justin Joseph, Periyasamy R
Abstract | Full-Text | PDF

Share this