Classification of lung sounds using wavelet-based neural networks

Abstract

Computer-based systems for diagnosing diseases have been widely used in various areas of medicine in the last decades and similar studies have been performed to parameterize and increase the reliability of lung-sound based diagnosis by using computational techniques. In this study, two types of classifiers, namely wavelet-based neural network and conventional artificial neural network (ANN), are used and compared for the classification of healthy and two-class athological lung sounds which are acquired using two microphones on the chest wall along with the air-flow signal. The inputs of classifiers are organized using two different methods, 'even-odd partitioning' and 'leave-one-out'. The lung sound signals belonging to inspiratory or expiratory phases are divided into thirty segments with 25% overlapping. In wavelet-based classifiers, the signals belonging to segments are decomposed to five levels using wavelet transforms and the reconstructed signal at each level is represented by AR parameters at the input of the network along with a volume constant indicating the sub-phases (early, mid, and late) of the respiratory cycle. The outputs of five networks belonging to five octaves are later combined to determine the performance of the classifier with respect to the frequency intervals used. For the ANN, the AR parameters obtained from the segments and the volume constant are used as inputs for the network. The classifiers operate on the respiration phases separately and a comparison between the results of the two phases indicates that expiration is more useful in diagnosis.