Identification of Velcro rales based on Hilbert–Huang transform

• Based on HHT, many parameters having physical meaning were exacted.
• Features were exacted from lung sounds collected from an electronic stethoscope.
• It was proved that these parameters could be used for Velcro rales identification.
• Relatively high classification accuracy was obtained.

Velcro rales, as a kind of crackles, are relatively specific for lung fibrosis and usually the first clinical clue of interstitial lung disease (ILD). We proposed an automatic analytic tool based on Hilbert–Huang transform (HHT) for the computerized identification of Velcro rales. In particular, HHT was utilized to extract the energy weight in various frequency bands (EWFB) of crackles and to calculate the portion of crackles during late inspiration. Support vector machine (SVM) based on the HHT-derived measures was used to differentiate Velcro rales from other crackles. We found that there were significant differences in the extracted parameters between Velcro rales and other crackles, including EW75−200, EW500−1000 and the proportion of crackles that appeared during the late inspiration. The discrimination results obtained from SVM achieved a concordance rate up to 92.20%±1.80%92.20%±1.80% as confirmed by the diagnosis from experienced physicians. For practical purpose, the proposed approach may have potential applications to improve the sensitivity and accuracy of auscultation and conduct automatic ILD diagnose system.