Study On Feature Extraction Methods And Realization Of Data Acquisition System For Acupoint Potential

Acupoint potential contails a wealth of human pathological and physiological status information, and interpretation of these information is of great importance to meridian scientific research and clinical diagnosis application. The aim of acupoint potential characteristics research is to explore meridian specifity and relation between acupoint potential signal and human pathological and physiological status. Many achievements have been made in this research, Nevertheless, there are still some unsolved problems: On the one hand, the acupoint potential is a weak signal with nonlinearity and time-variability, which is easy to be affected by various factors such as human individual differences, environmental noise and measuring methods, so the accurate measurement of this signal is difficult. On the other hand, less effective and targeted processing methods for this signal have been presented, especially the acupoint potental feature extraction methods which can reflect the human pathological and physiological status effectively. To solve these problems, research works on realization of data acquisition system and feature extraction methods for acupoint potential have been done in this dissertation which can be summarized as follows:(1)In acupoint potential amplifier circuit design, traditional power frequency notch filters are not so convenient to adjust the center frequency, and can’t withstand power frequency fluctuations. In order to address the problem, a new power frequency notch filter based on the center frequency band is proposed in this dissertation. By optimizing the design of the center frequency parameter of the basic second-order notch filters, the ability of power frequency interference suppression within the scope of the power frequency source frequency fluctuation has been improved. Results show that this kind of notch filters has much more advantages than traditional ones.(2) Amplifier circuit modules’cascading sequence has an impact on anti-power line interference ability of the circuit. Aiming at the problem, a new cascading sequence optimization model is proposed, in which the cascading sequence acts as the independent variable, the intermediate level powerline interference equivalent magnification acts as the objective function, and the constraint is the allowed range of each module’s input voltage, then the solution algorithms is given. Results show that the given algorithm can solve the problem effectively.(3) Traditional signal source models can’t generate complex periodic waveform, which made the simulating results not so intuitive. To overcome its shortcoming, a modified SPICE model is presented to generate arbitrary periodical waveform. The model is composed of periodical oblique wave sources. For achieving high precision, the periodical oblique wave source is built by using fundamental pulse sources connected in series with matched parameters, and a nonuniform sampling algorithm based on the second derivative of sampling points is proposed to improve the simulating speed. Results show that the presented model can solve the concerned problems effectively.(4) For overcoming the contradiction between the increasement of sample frequency and multiple tasks such as acupoint potential data reading, data story and real-time curve display, an improvement for traditional multi-thread technology scheme based on the "producer-consumer" model is proposed. According to the time complexity analysis of key tasks in the process of data collection, the scheme overcomes the contradiction perfectly by more reasonable arrangement of multi-thead tasks and improves the maintainability of the program by utilizing the programming thought centered on the inventory balance.(5) From the aspects of threshold criterion and the rule of eliminating false positives, an improvement for traditional baseline polynomial fitting in removal of baseline drift is presented to avoid leak detection and false positives of fitting points. Experimental results validate the effect of such improvement; Spectrum difference among channels of a bio-electricity amplifier has an impact on signal recognition. In order to decrease the impact, an algorithm for calibrating spectrum difference among multiple channels is presented. The algorithm is based on expected and practical amplitude-frequency & phase-frequency property of channels. By utilizing Fourier Transform relationships between continuous signal and its sample sequence, the discrete calibration interval in frequency domain of the output signal is determined, and amplitude calibration factors and phase bias at each frequency point in the interval is estimated, then the calibration operation is implemented. Simulation and experimental results validate the effectiveness of this algorithm.(6) The feature extraction method for acupoint potential signal based on Power Spectrum Density (PSD) is proposed to describe acupoint "energy". Four different algorithms (Welch Algorithm, Levinson-Durbin Algorithm, Burg Algorithm, Improved Covariance Algorithm) were compared in estimating PSD of acupoint potential. The spectrum estimation influence of different data length and methods of using several criteria for model order determination are studied, and then several feature extraction methods such as Average Spectrum Energy on Effective Frequency Band (ASEEFB), Average Spectrum Energy on Low Frequency Band (ASELFB), and Spectrum Energy Ratio between Low and High Frequency Band (SERLHFB) based on PSD are proposed. The classification ability of the three features is explored and compared in the experiment which is designed for distinguishing between acupoint and non-acupoint, between the physiological status of health people and that of unhealthy people who have a cold. Results show that SERLHFB is more preferable than the other two features in reflecting the acupoint specificity and human pathological or physiological status. SERLHFB at acupoint is higher than that at nonacupoint, and SERLHFB for healthy group is higher than that for unhealthy (cold) group.(7) For distinguishing between "pass" and "block" status of meridian, a feature extraction method for two acupoints potential based on heartbeat energy was proposed. The extraction process for this feature includes the positioning section based on the waveform of acupoint potential and the analysis, computation of heartbeat energy, for which the optimization problem of structure elements in Top-Hat Transform for the R-wave location detection is studied and the optimum evaluation formulas of the length of structure elements is deduced; Then the heartbeat energy sequence is achieved on the basis of R-wave detection for original signal. Two energy representing methods such as Absolute Heartbeat Energy (AHE) and Percentage Heartbeat Energy (PHE) that describe the heartbeat energy of acupoint are compared with each other, and results show that PHE is relatively stable and shows a normal distribution. So the mean difference of percentage heartbeat energy (MDPHE) for two measurement points is used as the new feature. Experimental results show that the two classification problems (acupoint and non-acupoint, illness (cold) and health) are all separable on the feature, and the feature value of MDPHE of the two acupoints for unhealthy (cold) group is bigger than that for healthy group.