Multi-channel classification of respiratory sounds.

In this study, respiratory sounds of pathological and healthy subjects were analyzed via frequency spectrum and AR model parameters with a view to construct a diagnostic aid based on auscultation. Each subject is represented by 14 channels of respiratory sound data of a single respiration cycle. Two reference libraries, pathological and healthy, were built based on multi-channel respiratory sound data for each channel and for each respiration phase, inspiration and expiration, separately. A multi-channel classification algorithm using K nearest neighbor (k-NN) classification method was designed. Performances of the two classifiers using spectral feature set corresponding to quantile frequencies and 6th order AR model coefficients on inspiration and expiration phases are compared.