A new method to detect crackles in respiratory sounds.

In this paper, an automatic method to detect and analyze crackles in digitised respiratory sounds is presented. The method is based on two steps: (1) a threshold (T) value is applied to the first derivative absolute value (FDAV) of lung sound to locate the "zone of interest" and (2) in this zone a crackle is detected if certain conditions are verified. The first derivative (FD) is evaluated by means of a derivative-smoothing filter, preserving areas under the spectral lines of the signal (moment zero), its mean position in time (first moment) and its spectral line width (second moment). The conditions to verify step 2 concern the following: the number and height of the peaks of FDAV and their distances from the starting point of the crackle, within a temporal window TW. This method shows good performances as an automatic detector (sensitivity 84% and specificity 89%), and is specifically designed to find the starting point of the crackle.