Automatic Classification of Respiratory Sounds Phenotypes in COPD Exacerbations

During the last few years, different COPD phenotypes are being defined. Despite being one of the basic characteristics of exacerbations, studies on changes and peculiarities of respiratory sounds during an exacerbation episode have been barely studied. A computerized analysis of respiratory sounds recorded in patients hospitalized because of acute respiratory symptoms was performed. It was aimed to be applied in the classification of COPD exacerbations only using the auscultation data registered after the admission. The analyzed exacerbations were classified into two categories according to the initial conditions and the evolution of the exacerbation in terms of its acoustic characteristics. Multi-parametric analysis using features extracted in the time-frequency domain was applied and a RBF network was trained and validated for classifying. Based on the cross-validation results, sensitivity of 78.4% and specificity of 81.3% were achieved. The proposed method could contribute to extend the knowledge of respiratory sounds during COPD exacerbations and to provide additional information of the disease as a basis for improving the impact on the patient.