Pulmonary crackle feature extraction using tsallis entropy for automatic lung sound classification

摘要

pulmonary crackle sound is produced by an abnormality in the respiratory tract. Pulmonary crackle sound is one of lung sound that is discontinuous, short duration and appears on the inspiratory phase, expiratory phase or both. Various methods are used by researchers to detect crackle sound automatically, for example using entropy measurement. Tsallis entropy is a measure of the entropy that has nonextensivity property. Tsallis entropy is often used to measure rapidly changing signals. Crackle sound has both of properties, so hopefully, Tsallis entropy can be utilized as feature extraction techniques for pulmonary crackle sound. The test results showed the use of Tsallis entropy with nonextensivity order of q = 2, 3, and 4 produce the highest accuracy. Using MLP and 3fold crossvalidation, an accuracy of 95.35%, Sensitivity of 90.48%, and 100% Specificity are achieved. The advantage of this method is the fewer number of features produced and simple computation. Tests using data classes and the number of larger data required in future studies.