Crackle detection and classification based on matched waveletanalysis

Abstract

Crackles have an explosive pattern in the time domain, with a rapid onset and a short duration. The timing, repeatability and shape of crackles are important parameters for diagnosis. Therefore, automatic detection of crackles and their classification have important clinical value. Since crackles have a general characteristic shape, it is obvious that wavelet analysis can be exploited to detect crackles and to classify them. In this paper, we present a new method for crackle detection which is based on a `matched' wavelet transform. We first model crackles as a mathematical function. Then we design a matched wavelet based on this model. Applying a soft threshold to the results of the continuous wavelet transform to suppress noise further, the optimal scale can be obtained. Crackles can be detected based on the envelope of the signal at an optimal scale, and can be classified based on the energy distribution with scale. Theory, methods and experimental results are given in detail in this paper.