Abnormal-respiration detection by considering correlation of observation of adventitious sounds

We propose a classification method to distinguish between normal and abnormal respiration by considering the correlation of the observation frequencies of adventitious sounds between auscultation points. This method is based on the fact that adventitious sounds are frequently observed in lung sounds from multiple points. We use the product of the correlation score and the abnormality score, which indicates the likelihood that a candidate is abnormal, of lung sounds from different points. When using lung sounds from eight points, the proposed method achieved a higher classification performance of 92.0% between normal and abnormal respiration compared with the baseline method not considering the other lung sounds, which achieved a performance of 84.1%. Our approach to the classification of healthy subjects and patients also achieved a higher classification rate of 90.8%.