Detection of abnormal lung sounds taking into account duration distribution for adventitious sounds

In this paper, we propose a novel method for distinguishing between normal lung sounds from healthy subjects and abnormal lung sounds containing adventitious sounds from patients. The spectral similarity of adventitious sounds and noises at auscultation makes it difficult to obtain a high accuracy of the abovementioned classification. However, there is a remarkable difference between the duration of noise sounds and that of adventitious sounds. In the proposed method, the duration of these sounds is described using a Gaussian/Gamma distribution. The spectral likelihood using hidden Markov models and the validity score of the duration of the noise/adventitious sounds are combined to derive the most likely acoustic segment sequence for each respiration. Our classification method achieved a higher classification rate of 90.0% between normal and abnormal lung sounds than the conventional method (classification rate: 88.1%). Our approach to the classification of healthy subjects and patient subjects using the proposed method also achieved a higher classification rate of 84.1%.