Investigating into segmentation methods for diagnosis of respiratory diseases using adventitious respiratory sounds.

Respiratory condition has received a great amount of attention nowadays since respiratory diseases recently become the globally leading causes of death. Traditionally, stethoscope is applied in early diagnosis but it requires clinician with extensive training experience to provide accurate diagnosis. Accordingly, a subjective and fast diagnosing solution of respiratory diseases is highly demanded. Adventitious respiratory sounds (ARSs), such as crackle, are mainly concerned during diagnosis since they are indication of various respiratory diseases. Therefore, the characteristics of crackle are informative and valuable regarding to develop a computerised approach for pathology-based diagnosis. In this work, we propose a framework combining random forest classifier and Empirical Mode Decomposition (EMD) method focusing on a multi-classification task of identifying subjects in 6 respiratory conditions (healthy, bronchiectasis, bronchiolitis, COPD, pneumonia and URTI). Specifically, 14 combinations of respiratory sound segments were compared and we found segmentation plays an important role in classifying different respiratory conditions. The classifier with best performance (accuracy = 0.88, precision = 0.91, recall = 0.87, specificity = 0.91, F1-score = 0.81) was trained with features extracted from the combination of early inspiratory phase and entire inspiratory phase. To our best knowledge, we are the first to address the challenging multi-classification problem.