Lung sounds classification using spectrogram's first order statistics features

Abstract

Lung sounds can indicate a person's health condition. Lung sounds are generated from the air flow in the respiratory tract. Various of signal processing techniques are used for lung sounds analysis to reduce the subjectivity of the lung sound analysis. In this study, we propose lung sound signal analysis using first order statistic texture analysis on the spectrogram. The mean, variance, skewness, kurtosis, and entropy are used as features of each lung sound. These features are analyzed using K-NN with two methods of distance measurement. The proposed method achieves an accuracy of 96.3% for 81 data.