Lung sound classification using wavelet transform and entropy to detect lung abnormality

Abstract

Lung sounds provide essential information about the health of the lungs and respiratory tract. They have unique and distinguishable patterns associated with the abnormalities in these organs. Many studies attempted to develop various methods to classify lung sounds automatically. Wavelet transform is one of the approaches widely utilized for physiological signal analysis. Commonly, wavelet in feature extraction is used to break down the lung sounds into several sub-bands before calculating some parameters. This study used five lung sound classes obtained from various sources. Furthermore, the wavelet analysis process was carried out using Discrete Wavelet Transform (DWT) and Wavelet Package Decomposition (WPD) analysis and entropy calculation as feature extraction. In the DWT process, the highest accuracy obtained was 97.98% using Permutation Entropy (PE), Renyi Entropy (RE), and Spectral Entropy (SEN). In WPD, the best accuracy achieved is 98.99 % when 8 sub-bands and RE are used. These results are relatively competitive compared with previous studies using the wavelet method with the same datasets.