Detecting non-stationarity for auscultation signal of traditional Chinese medicine

The information about the nonstationarity of the auscultation signal is utilized in this paper to objectively and automatically identify healthy people and patients with qi-deficiency or yin-deficiency. In order to characterize the nonstationarity of the sound signal, the nonlinear cross-prediction method is used to extract features from the signal. A feature selection method based on conditional mutual information maximization criterion (CMIM) is implemented to find an optimal feature set. By means of the support vector machine (SVM) classifier, three common states (healthy, qi-deficiency and yin-deficiency) in traditional Chinese medicine are distinguished using the feature set, and a satisfactory classification accuracy of 80% is achieved in the experiment. In conclusion, the analysis based on the nonstationarity of the sound signal provides an alternative and outstanding approach to the objective auscultation of traditional Chinese medicine (TCM).