Detection of Abnormal Respirations Considering the Occurrence Timing of Adventitious Sounds

Adventitious sounds, abnormal lung noises, are heard in a person with pulmonary disease. Thus, we aimed to automatically detect abnormal sounds from auscultatory sounds. For this purpose, we expressed the acoustic features of normal lung sounds for healthy subjects and abnormal lung sounds for patients using two type of stochastic models and distinguished between normal and abnormal lung sounds. Although the classification rate achieved was 86.53%, issues arose due to the acoustic similarity between background noise and heart sounds with adventitious sounds, leading to their misidentification as abnormal lung sounds. Therefore, we focus on the occurrence timing of adventitious sounds during both expiration or inspiration. In this study, we propose a method to accurately discriminate between normal lung and abnormal lung sounds by considering the occurrence tendency of adventitious sounds during the respirations. Classification experiments confirmed that the classification rate was increased (91.12%). The results indicate the effectiveness of the proposed method.