Abstract
A nonlinear digital filter system is proposed for automatic extraction of crackles which are discontinuous adventitious sounds in lung sounds. This system is composed of two nonlinear digital filters; one is a stationary-nonstationary separating filter, and the other is a width-discriminating filter. The former separates nonstationary signals from stationary ones in the lung sounds, using the prediction error to the input lung sound signal. If the prediction error is small enough, the lung sound is considered to be stationary, but if the error is large, a nonstationary signal is considered to occur and the nonstationary part is separated. The latter, the width-discriminating filter, performs signal extraction by considering the signal wave form of the crackles, simply realized by logical algebra. This filter extracts an impulsive signal, which is a small-width wave, and its succeeding waves; such wave form is typical of that of crackles. First, crackles are roughly separated by the stationary-nonstationary separating filter as nonstationary signals, and then are more precisely extracted from the nonstationay output by the width-discriminating filter. Some examples of processing actual lung sound data by this system show its high performance.
