Applied Research On The Technology Of Near-infrared Spectrum In Medicine Testing

Our country's medicine industry develops with each passing day. Now,thousands of western medicines and two thousand Chinese traditional medicinesare produced in China, Which makes our country become the second producingarea in quantity. In order to minish medical treatment accident, to improvedomestic medicine's credit and make medicine industry develop placidly,medicine's quality and effective component should be strictly controlled whenmedicine industry rapidly develop. Not long ago, State monitoring andadministering bureau for food and medicine announced that national monitoringbranch for medicine will startup 'plan of enhancing natinal state medicine standard'in 2005. This plan's main idea is to make national technology of measuring formedicine reach to international advanced level and to realize the national medicinestandard connects with international forerunner. This paper's intention is to trieshard to find out fast, simple and accurate method for measuring medicine.This paper introduces the technology of Near Infrared Spectroscopy (NIR) tomeasure effective component for medicine. The technology of NIR possesscharacteristic of not destroying sample, no pollution, on-line measure, saving timeand saving labor. NIR contains various information because every matter andcharacteristic can bring absorbance in NIR spectrum area. The technology has beenapplied in extensive fields such as agriculture, industry, medicine and health care.NIR spectrum belongs to weak information spectrum area because this areaincludes information of sum of fundamental frequencies and product offundamental frequencies in which plenty of information wraps. So the informationcan't be recognized in the traditional way. Recognizing information is the key ofNIR technology for NIR. This paper emphasizes on recognizing and analyzing NIRin process of medicine measure. We take example for several NIR of western medicines and Chinesetraditional medicines to study NIR data processing method with chemistrymetrology such as wavelet, Partial Least Square Regression (PLS) and ArtificialNeural Network (ANN) Frist, this paper improved on arithmetic of wavelet threshold filtering andarithmetic of spatial correlation filtering and applied them into process of NIR. VCYinqiao tablets are Chinese traditional medicines, which have complex componentand weak spectral signal. When we filtered their NIR by arithmetic of waveletthreshold filtering, the signal that has near scope with noise scope would be filtered.When we filtered their NIR by arithmetic of spatial correlation, a little noiseswould be reserved. In order to raise NIR forecasting precision, this paper brings upbetterment aim to shortage for this two filtering. For arithmetic of waveletthreshold filtering, we introduce slim adjustment gene according to waveletcharacteristic whose coefficient of signals and noises spread in different scale(Signal's wavelet coefficient augments with scale's augment and noise's waveletcoefficient diminishes with scale augment.). Threshold with slim adjustment genecan change with different characteristic of signal and noise, which benefit tosignal's passing and noise's filtering. For the arithmetic of spatial correlationfiltering, noise is imported when mistakenly estimating halt iteration conditionoccurs. We approach correct halt iteration condition with adaptive filter, whichgreatly minish noise.This paper process aether NIR came from computer simulation with two betterwavelet filtering methods and compare the result to result based on two traditionalwavelet filtering, Better wavelet threshold filtering raised gain of Signal NoiseRatio to 1.1dB and better arithmetic of spatial correlation filtering raised gain ofSignal Noise Ratio to 1.6dB. We also applied two better filtering method to preprocess for VC Yinqiao tablet NIR, after that, forecasted components ofacetaminophen and vitamin C with Principal Component Regression. For result bybetter wavelet threshold filtering method, the regression of coefficient ofacetaminophen is 0.994 and average absolute error is 0.2227, the regression ofcoefficient of vitamin C is 0.996 and average absolute error is 0.1957. For result bybetter wavelet spatial correlation filtering method, the regression of coefficient ofacetaminophen is 0.993 and average absolute error is 0.2246, the regression ofcoefficient of vitamin C is 0.994 and average absolute error is 0.1962. Theexperiment and result of data processing show that two better wavelet filteringmethods can enhance prediction precision. For apply wavelet filtering into preprocess of NIR more effectively, discussedthe influence of decomposed times of wavelet, ranks of wavelet and choose ofwavelet base to wavelet filtering for NIR of Shanghajinwei tablet and blood andpresented principle of confirm of decomposed times of wavelet, choose of ranks ofwavelet and wavelet Second, present a novel section Partial Least Square Regression based onresearch on traditional PLS. Mensurated the component of glucose in glucoseliquor, discussed the effects of three preprocess methods which are smoothness,derivative coefficient and base line revising. Results show that preprocessesmethod of smoothness can enhance prediction precision for glucose liquor's NIR.In addiction, the best NIR wave band should be choosed in order to enhanceforecasting precision by PLS. The result for glucose liquor processing indicates thatobvious error appeared in small concentration section by traditional PLS.Traditional PLS can't be applied in prediction of glucose concentration. This paper presents another NIR Multi-component analysis method: sectionPartial Least Square Regression. First, this method divides concentration range oftraining samples into some sub-ranges and respectively computes PLS correlationmodel in every sub-range with the sub-range's training samples. Then, we classifyprediction samples according to its concentration sub-range with clustering analysisand judges which sub-range the prediction sample belongs to. Last, compute theconcentration of prediction component with the PLS correlation model of thesub-range according to clustering analysis. Thus, the model's applied range andprediction precision is raised at the same time. We used this section PLS to predictglucose concentration of glucose liquid, its regression of coefficient was 0.99, itsRMSEC reached to 0.19 and average absolute error is 1.06. Compared to result ofliterature 104, this method evidently enhances prediction precision and offersanother method for analysis of NIR. Third, researched application of BP ANN and linear ANN in the quantitativeanalysis of NIR, enhanced stability of BP ANN with Bates standardization,enhanced precision of NIR with linear ANN, discussed parameter's adjustment ofANN. it exist the shortcoming of poor stability and bad reappearance to apply BPANN to quantitative analysis of NIR. This paper solves this problem and enhancesstability of BP ANN by Bates standardization. We used BP ANN and Batesstandardization to mensurate content of rutin and vitamin C of compound rutintablet and received the regression of coefficient of rutin, average absolute error ofall samples and average absolute error of prediction samples respectively is 0.99,0.374 and 0.459, received the regression of coefficient of vitamin C, averageabsolute error of all samples and average absolute error of prediction samplesrespectively is 0.99,0.478and 0.589. The result shows that BP ANN and Batesstandardization can enhance analysis precision in NIR quantitative analysis. Linear ANN is not as good as BP ANN in nonlinear emendation because ofits linear concealed passing function. But, it inexistence partial least extremum andproblem of flat because of its error curve of paraboloid type, moreover, linear ANNcan possess zero error and less error when it is used to linear model's prediction.Consequently, the stability, convergence and prediction precision of linear model inlinear ANN excelled BP ANN. For compared vary processing method, we predictacetaminophen of Anjiahuangmin capsule for children with PLS, BP ANN andlinear ANN, the result shows that the regression of coefficients respectively is0.994, 0.992 and 0.996, the average absolute error of prediction samplesrespectively is 1.021, 1.051 and 0.630. We can draw a conclusion from aboveexperiment and process, for prediction of acetaminophen of Anjiahuangmincapsule for children, linear ANN excelled BP ANN and PLS and BP ANN is asgood as PLS. Whether for BP ANN or for linear ANN, we should attach Parameter adjustingof ANN importance to prediction precision. There are pivotal parameters such asthe number of inputting point of ANN, the number of concealed layer nerve cell ofANN, average square error aim of ANN and learning times of ANN in parameteradjusting of ANN. To choose these parameters, we should consider factor ofvarious facet. For example, when we ensure the number of concealed layer nervecell of ANN, precision of ANN will diminish if the number of concealed layernerve cell of ANN is lack, the extensive ability will become weak when the numberof concealed layer nerve cell of ANN is excessive, which demand us to considertwo facet factor when adjust parameter of ANN. So, we should consider forcasting...