Fractal Dimension for Lung Sound Classification in Multiscale Scheme

Abstract

Lung sound is a biological signal with the information of respiratory system health. Health lung sound can be differentiated from other pathological sounds by auscultation. This difference can be objectively analyzed by a number of digital signal processing techniques. One method in analyzing the lung sound is signal complexity analysis using fractal dimension. To improve the accuracy of lung sound classification, Fractal Dimension (FD) is calculated in the multiscale signal using the coarse-grained procedure. The combination of FD and multiscale process generates the more comprehensive information of lung sound. This study used seven types of FD and three types of the classifier. The result showed that Petrosian C in signal with the scale of 1-5 and SVM with fine Gaussian kernel had the highest accuracy of 99% for five classes of lung sound data. The proposed method can be used as an alternative method for computerized lung sound analysis to assist the doctors in the early diagnosis of lung disease.