Detection of adventitious lung sounds using entropy features and a 2-D threshold setting

The presence of adventitious lung sounds, such as the wheezing sound, is an indication of possible respiratory disorders. Many algorithms have been proposed in the literature for the detection of adventitious lung sounds but they involve the use of sophisticated pattern recognition techniques which are complex and are hence not suitable for use in wearable personal devices. While a recent work reported in the literature uses a small number of features and a simple threshold based algorithm for wheeze detection, it is not designed for use when there are more than two signal types to be detected. This paper investigates the problem of automatic detection of four types of lung sounds namely, stridor, wheeze, crackle, and normal lung sounds. Specifically, we propose a computationally efficient detection method that involves the use of only two entropy features and a two-dimensional threshold setting. The proposed method has been tested with 45 samples and promising preliminary results in detection accuracy have been obtained. The method has also been tested to be robust against additive white Gaussian noise added artificially to the test samples.