Classification Between Abnormal and Normal Respiration Through Observation Rate of Heart Sounds Within Lung Sounds

This paper proposes an effective classification method to differentiate between normal and abnormal lung sounds, which takes into account the detection level of heart sounds. Abnormal lung sounds frequently contain adventitious sounds; however, misclassification between heart sounds and adventitious sounds makes it difficult to achieve a high level of accuracy. Furthermore, the classification performance of conventional methods, which use the detection function of heart sounds, becomes worse for those lung sounds which contain a low level of heart sounds. To address this problem, our proposed method changes the classification method according to the detection rate of heart sounds, whereby if the rate was high, the heart-sound models in the HMM -based classification method were used. In addition to spectral information, temporal information of heart sounds and adventitious sounds were also used to obtain the rate more precisely. When using lung sounds from three auscultation points, the proposed method achieved a higher classification performance of 89.90% (between normal and abnormal respiration) compared to 88.7 % for the conventional method, which used the detection function of heart sounds. Our approach to the classification of healthy and unhealthy subjects also achieved a higher classification rate of 86.6%, compared to 83.1 % when using the conventional method having the detection function of heart sounds.