Detail Investigation on Success of the Filters and Classification Algorithms for Determining Pneumonia Disease

Abstract

Today, as a result of the widespread use of electronic stethoscopes, lung sounds can be recorded and transferred to a computer. Thus, it was possible to analyze the lung sounds. Healthy lung sounds are called normal sounds, and unhealthy sounds are called abnormal sounds. In this study, lung sounds taken from the Kaggle site were studied. Normal and three types of abnormal sounds were selected among lung sounds. Abnormal sounds are belonging to patients with pneumonia, bronchiectasis, and COPD. Pneumonia sounds were tried to be distinct from other sounds. In the presence of pneumonia, crackles are heard in addition to normal lung sounds. For detecting crackles, first of all, three different filters were designed Butterworth, Chebyshev, and Elliptic bandpass filters by using MATLAB. Wavelet Transform was applied to the filtered sounds and features were extracted from the obtained subbands. These features were classified as normal-abnormal and 'yes or no pneumonia' with the help of K-nearest neighbors and Support Vector Machines. According to the results obtained, the filters and classification methods were compared.