| In traditional Chinese medicine (TCM) pulse diagnosis has been successfully usedfor thousands of years. In traditional Chinese pulse diagnosis, doctors put theirfingers on human wrist to feel pulse. In view of pulse signal containing importantinformation of human organ, the doctor could use this information to determine thepatient’s physical condition. But the accuracy of pulse diagnosis completely dependson the experience and skills of the doctor. In the diagnosis of the same patientdifferent doctors maybe give different diagnostic results. Therefore it is urgent todevelop a standardized and objective method for pulse diagnosis. In recent years,although a lot of researchers objectively make research for pulse diagnosis usingmodern equipments and methods and get some research, there still are manyproblems to be solved.In the paper, pulse signals are collected by ultrasound equipment that thecollected signals are called as Doppler ultrasound pulse signals. We will focus onquality evaluation, feature extraction and classification methods of Dopplerultrasound pulse signal for the objective study of pulse diagnosis.In the objective study of pulse diagnosis, how to accurately and effectivelyacquire pulse signal is the most critical work. However, in current acquisition deviceof pulse signal it is not with the function of quality evaluation, which leads to alarge number of low quality of pulse signals are collected for clinical diseasediagnosis. Thus second-order differential sample entropy-based a real-time qualityevaluation mothed of signal is proposed for Doppler ultrasound pulse signal in thispaper. First, the wavelet transform is adopted for removing the high frequency noiseand low-frequency baseline wander of Doppler ultrasound pulse signals. Then, bycalculating the second-order differential and sample entropy of Doppler ultrasoundpulse signal the real-time evaluation method is carried out. The results show thatthis method could reduce the loss of sample and efficiently acquire high-qualityDoppler ultrasound pulse signals.In order to make full use of Doppler ultrasound pulse signal for the analysis andclassification of disease, in this paper multiscale sample entropy-based method,TWED distance-based method and multiple kernel learning-based method ofcombination of heterogeneous features for Doppler ultrasound pulse signalclassification are respectively proposed for the analysis of Doppler ultrasound pulsesignal.In the classification research of multiscale sample entropy-based method first thevalue of sample entropy is calculated with different parameters, i.e., τ, m and r, that the calculated value could constitute a multidimensional vector. In virtue of theconstructed redundant multidimensional vector we proposed multiple linearsubspace learning methods to extract useful information and eliminate redundantinformation from multidimensional vector. The vector with dimensionalityreduction is input into the SVM classifier for the classification of healthy people andpatients. Experimental results show that the proposed method is efficient indiagnosis of disease.In the classification research of TWED distance-based mothed we proposed aclassification method with non-feature extraction that combined with1NN classifier1NN-TWED classifier is constructed for the classification of Doppler ultrasoundpulse signal. In the classification method it introduces a stiffness parameter tocontrol the elasticity of the metric in the time domain, and thus it is more flexiblefor Doppler ultrasound pulse signal matching and classifying.In the classification research of multiple kernel learning-based method ofcombination of heterogeneous features for Doppler ultrasound pulse signalclassification, first it respectively selects the appropriate kernel function forheterogeneous features of Doppler ultrasound pulse signal. Then SimpleMKL isadopted to integrate the features of Doppler ultrasound pulse signal that deletesredundancy features and make use of the valuable features for signal classification.The experimental results show that in virue of the enhanced information theproposed method could increase the classification accuracy and classification speed.Finally, by adopting Doppler ultrasound pulse signal for the diagnosis of thedisease related with flow velocity and blood viscosity the classification results showthe performance of Doppler ultrasound pulse signal in disease diagnosis. |
