Research on Feature Extraction and Classification of Pulse Waveform for Cholecystitis and Nephrotic Syndrome Diagnosis
|School||Harbin Institute of Technology|
|Course||Computer Science and Technology|
|Keywords||Pulse signal Feature extraction Best bases of WPT SVM|
Traditional Chinese Medicine (TCM) has been proved to be worthwhile and clinically valid over 2000 years. Traditional Chinese Pulse Diagnosis(TCPD) is the distinctive one of the four Traditional Chinese Medical Diagnoses. However, the doctors palpate pulses with fingertips and then to understand and judge the disease condition through their comprehension. Thus this method is limited to the doctor’s experiences, so the objectifying of TCPD is very necessary.Over the recent years, most of the studies focused on the possibilities of diagnosing cardiovascular diseases using pulse waveform and it had been proved feasible. Moreover, the pulse signal contains not only the physiological information of the heart but also that of some other organs such as gallbladder and kidney. If there are problems in these organs, the pulse signals will also be changed slightly. Hence, this paper further investigated the effects of cholecystitis and nephrotic syndrome on the pulse waveform.During the acquisition of pulse signals, the high frequency noise can be introduced by the interference of other device. We adopt db3 wavelet to eliminate the high frequency noise and spline method to remove the baseline wander of pulse waveform which caused by the respiration and body movements of the subjects. Moreover, the fluctuation of power could bring some pseudo-peaks into pulse waveform. We designed a method based on the proportion of peaks’ time to remove the pseudo-peaks which overcome the shortcomings of the original PPH algorithm.The feature extraction of pulse signals is important too. As a kind of physiological signal, experienced Chinese doctors analyze the structure of pulse image and give 14 features with the corresponding physiological significance. Based on those features, we extract 8 relative features from pulse signals and experimental results show that the relative features is more effective than original features. Moreover, pulse signal is non-stationary. Among the non-stationary signal processing technologies, wavelet transform and wavelet packet transform are most widely used. However, wavelet transform is always limited to the fixed decomposition form when it is used in feature extraction. So, we extract a set of energy features from normal people, cholecystitis patients and nephrotic syndrome patients based on the best basis of wavelet packet decomposition. The classification result shows that the energy features we extracted is more efficient than the energy features based on wavelet transform and wavelet packet transform.In order to classify the three sets of pulse signal, we adopt KNN and SVM to perform the classification experiment. Experimental results show that time features is more efficient in discriminating cholecystitis patients from normal people than energy features. But energy features provide a better capability when discriminating nephrotic patients from normal people and classifying the two sets of patients with different disease. Thus, we try to integrate the pulse information, including features and classifiers. At last the classification accuracy is up to 76.44%, which shows the feasibility of diagnosing these disease using pulse signals.