Multi-channel acoustic mapping of respiratory system based on adventitious sound source separation

The aim of this study is to localize the pathological compartments of the lung based on the characteristics of the modified pulmonary sounds. Sound signals recorded simultaneously at more than one point on the posterior chest wall are subjected to source separation methods in order to separate sound components associated with the disease. For this study, Basic Independent Component Analysis (BICA), Separation By Autocovariances (SBA) and Convolutive Blind Source Separation (CBSS), out of various source separation methods, are adopted and used. The measure proposed to find the most similar among the extracted independent components to the true source is kurtosis and Kullback-Liebler (K-L) distance. After the similarities between measured signals and the chosen component are located as coefficients onto the recording points on the chest wall and thereby formed matrix is visually improved via interpolation, a two-dimensional map is obtained, although with low resolution, showing estimated pathology source location and audibility distribution of the adventitious sound component around this source. This study is intended to be a pioneer to studies on acoustic-based respiratory system imaging which can function as an alternative to computer chest tomography in some necessary circumstances, and complementary in others.