Classification of Respiratory Sounds using Wavelet Transform and Neural network

Abstract

Respiratory sound contains information of lung condition which helps in the diagnosis of lung diseases. Stethoscope is the traditional method used to obtain this information but it depends on the physician experience and hearing. To avoid this limitation and to make optimum benefit of the respiratory sound information a computer aided diagnosis system was built. The respiratory sound signals were divided into segments each contains one inspiratory and expiratory cycle, wavelet transform (WT) was used for analysis, features were obtained from its coefficients and finally classifying using artificial neural network (ANN) to normal sound and abnormal sound and classifying the abnormal sound to crackle and wheeze. The accuracy of classification between normal and abnormal was 95.7% and for classification between crackle and wheeze was 98.1%.