In this paper, we propose a new method for classifying patients with pulmonary emphysema and healthy subjects using lung sounds. Using conventional classification methods, every boundary between inspiratory and expiratory phases in successive respiratory sounds are detected manually prior to automatic classification. However, manual segmentation must be performed accurately and has therefore created significant obstacles in achieving automatic classification. In our proposed method, adequate boundaries are detected automatically in the classification process, based on the criterion of maximizing the difference between the acoustic likelihoods for a candidate with abnormal respiration and one with normal respiration. The proposed method achieved a classification rate of 83.9% between healthy subjects and patients. The reported rate was 1.3% greater than the rate achieved using the conventional method, which required manual phase-wise segmentation. Furthermore, the resulting rate was 2.2% higher than the rate obtained by the classification in which a lung sound sample was divided into phases of equal duration, indicating the effectiveness of the proposed method.

No abstract

Diagnosis of pulmonary emphysema by using a stethoscope is based on the common knowledge that abnormal respiratory (adventitious) sounds usually appear in patients with pulmonary emphysema. However, the spectral similarity between adventitious sounds and noises at auscultation makes highly accurate automatic detection of adventitious sounds difficult. In this paper, we have proposed a novel method for distinguishing between normal lung sounds in healthy subjects and abnormal sounds, including adventitious sounds in patients, taking into account the occurrence tendency of adventitious sounds and noises. According to our investigation results, adventitious sounds occur repeatedly in successive inspiratory/expiratory phases of patients. On the other hand, noise sounds mix at random in lung sounds of both patients and healthy subjects. In our method, the occurrence tendency of these sounds is described using Gaussian distribution of a random variable obtained by subtracting the acoustic likelihood for abnormal respiration from the likelihood for normal respiration. The spectral likelihood calculated using hidden Markov models and the validity score of the occurrence tendency of the adventitious/noise sounds are combined to derive the classification result. Our method achieved a higher classification rate of 94.1% between normal and abnormal lung sounds than that achieved using the conventional method (87.4%).