The Relevance of Auscultation Site in Healthy versus Restrictive Pulmonary Disease Classification

Abstract

The aim of this study was to perform a quantitative analysis to assess the relevance of the auscultation site on the chest wall in classifying restrictive lung diseases versus the healthy class. Accordingly, 41 features were calculated from 14-channel pulmonary sounds acquired on the posterior chest wall of 20 healthy subjects and 20 subjects with restrictive lung disease (5 with pneumonia and 15 with interstitial lung disease, ILD). These features were then used to train Bayes, decision tree (DT), and support vector machine (SVM) classifiers in leave-one-subject-out scheme. The two symmetric sites at the bases of the right and left lungs achieved the highest subject-level F1 scores of 0.974 and 0.976, respectively, while also maintaining the smallest deviation in flow cycle-level accuracies across subjects. These results were obtained using the DT classifier with f(50) and MFCC5.