Sen, IpekWali, AmrouHassan, Salma Ahmed2024-07-182024-07-182022978-1-6654-5432-2https://doi.org/10.1109/TIPTEKNO56568.2022.9960198https://hdl.handle.net/11411/7796Medical Technologies Congress (TIPTEKNO) -- OCT 31-NOV 02, 2022 -- Antalya, TURKEYIn this study, the aim is to understand what happens to the pulmonary acoustic features due to smoking, with a particular interest on the distances to the non-smoker and chronic obstructive pulmonary disease (COPD) groups in the feature space. Three new measures are defined, the first two of which are to quantify the differences between the three groups and to sort the features with respect to their discriminative abilities. The third measure is defined to quantify the linear displacement of smokers' group away from the non-smokers towards COPD in the multivariate feature space formed by the selected features in the previous step. Among the features adopted, percentile frequencies f(50) and f(75), and the eighth mel-frequency cepstral coefficient (MFCC8) are selected by the first two measures in common. Using this set of features, it is found that early-inspiration, early-expiration, and the seventh microphone location are more useful to detect early signs of COPD. Recommending a reliable methodology for early detection of COPD in this manner requires increasing the data size, devising other measures, running additional tests, and interpreting the results clinically.eninfo:eu-repo/semantics/closedAccessSmokersNon-SmokersCopdPercentile FrequenciesMfccAr ModelRenyi EntropyStatistical DifferenceDistanceRespiratory SoundsLung SoundsDiagnosisDiseaseAsthmaConference Object2-s2.0-8514405633910.1109/TIPTEKNO56568.2022.9960198N/AN/AWOS:000903709700053