Classification of lung sounds based on linear prediction cepstral coefficients and support vector machine

The listening to the sounds of lungs is very important to know the detection and analysis of respiratory disorders. Physicians are not able to detect accurately lung sounds of patients. Many computer programs are conducted to help physicians in diagnosing lung diseases. In this paper, a robust classification method of lung sounds (i.e. polyphonic or stridor) is proposed. Features are extracted using Discrete Wavelet Transform (DWT) first. Secondly, linear prediction cepstral coefficients (LPCCs) are calculated. After that delta and delta-delta of LPCCs are extracted. Variance and kurtosis of LPCCs, delta LPCCs and delta-delta LPCCs are extracted as features of lung sounds. Classification of lung sounds is conducted using support vector machine (SVM). Training and testing data are chosen randomly from 42 subjects using cross-validation. Both numbers of testing and training subjects are 21. The obtained recognition percent is 95.24%. So, new classification algorithm is conducted between polyphonic and stridor sounds of lung sounds. The obtained recognition percent is the most.