Studies On Nondestructive Quantitative Pharmaceutical Analysis Using NIR Spectroscopy Technique

In recent years, near infrared (NIR) spectroscopy has made rapid progress as a powerful analytical tool. NIR spectroscopy analysis technique is an efficient, simple, nondestructive and no contamination method that has been used in chemical analysis in diverse fields, such as agriculture, food, petrochemical, textile and pharmaceutical industries etc. NIR has made lots of pharmaceutical analysts put their hearts in it, its widely prospect in this field is attractive.The NIR spectral region is generally defined as the wavelength range from 780nm～2500nm. It is customarily divided into two ranges, short-wavelength NIR spectroscopy (780nm～1100nm) and NIR spectroscopy (1100nm～2500nm). The spectra in this region are dominated by the absorption for overtones and combinations of fundamental vibrational modes correspond mainly to O-H,C-H and N-H groups in compound. However, NIR spectroscopy has some disadvantages resulting in weak, partly overlapped, non-specific bands, so it is hard to interpret of the data only depend on the conventional analytical method. In order to obtain the exact result in the qualitative and quantitative analysis, some data processing methods must be applied. The modern NIR technique is the outcome of NIR spectroscopy combined with the chemometrics method. Combined with the strong information process ability of chemometrics, NIR spectroscopy can fully demonstrate its rapidness, precision and nondestructive analysis which has great potential in these application.The chemometrics methods such as three linear regression modeling methods, multiple linear regression (MLR), principal component regression (PCR) and partial least squares (PLS) are commonly used. However, the linear methods possess some deficiencies such as modeling of data sets containing strong nonlinear relationships, whereas there always exists nonlinear mapping between the spectra data and concentration of the component, so when they were used to deal with the nonlinear system, these methods always make some higher prediction errors. In quantitative analysis, artificial neural networks (ANNs) are more and more widely applied during the past several years. The main advantage of ANN is their anti-jamming, anti-noise and robust nonlinear transfer ability. It has been demonstrated that it is possible to obtain excellent results in multivariate calibration problems by using ANN.The conventional determination of active compounds content in pharmaceutical analysis often calls for using various analytical techniques, including chromatographies (GC, HPLC) and spectroscopies (IR, UV). However, these methods usually require dissolving the samples, separating them and determining their ingredients, so it not only time-consuming, but also destroy the samples, and cause some contamination. In this paper the active components of four different drugs were determined by using NIR spectroscopy combined with radial basis function (RBF) neural network, principal component analysis-radial basis function (PCA-RBF) network, genetic algorithm-radial basis function (GA-RBF) network and adaptive neuron-fuzzy inference system (ANFIS) model, respectively. The results show that the optimal models that designed by using such methods were convenient and expeditious in the determination.The main research content of this dissertation involves:1. Different pretreated techniques were applied to preprocess the original spectra. Original spectra contain not only the information from the components of the samples, but many noises from any aspects. These noisy signals can interfere in spectral information. Therefore, the original spectra must be pretreated. The general pretreated methods such as standard normal variate (SNV), multiplicative scatter correction (MSC), first-derivation and second-derivationwere studied and discussed detailedly in this paper. For different samples, the selection of the optimum pretreated method might be different.2. Both RBF model and PLS model based on short-wave NIR spectroscopy of compound erythromycin ethylsuccinate powder drug were established. Through the comparation of RBF and PLS multivariate calibrations, we can see that, the best result was obtained when RBF model was applied. It shows that PLS calibration results in the poor predicted models due to its overlapped bands, complicated absorptance, and nonlinearity. And RBF method can result in the robust models due to its anti-jamming, anti-noise and nonlinear transfer ability. So the proposed RBF method based on short-wave NIR is more valuable and economical for quantitative analysis than PLS model.3. The principal component analysis (PCA) method was added to the RBF algorithm, i.e., the scores of the principal component analysis of the response data of the calibration mixture were used as an input layer. The result of PCA-RBF model in determination compound amoxicillin powder drug was satisfied. The use of scores reduced input nodes, so training time of the network was shortened, and some of the noise was removed, thus enhancing the chemical information in the spectra. PCR and PLS multivariate calibrations are also used, which are compared with PCA-RBF neural networks, both RMSE and R of PCR and PLS models were worse than those of PCA-RBF models, so this method is better completely .4. Genetic algorithm (GA) was added to the RBF networks in determination of compound erythromycin ethylsuccinate powder drug and thiamphenicol powder drug on NIR spectroscopy,and it was used to search the centers and width of the RBF functions, so as to make certain of the structure of the networks. In addition, the application of wavelength selection by genetic algorithms is used in GA-RBF models for determination of compound thiamphenicol powder drug, and the applications of wavelength selection with genetic algorithms is systematically studied, which enhance prediction accuracy and improve the robustness of the calibration model. Furthermore, the Akaike's information criterion (AIC) was applied, which provides a compromise between network performance and network complexity is used to evaluate the fitness of individual networks. Therefore the GA-RBF networks have a better generalization performance and simpler network structure. Experimental results demonstrate that the proposed GA-RBF method based on NIR spectral data is a valuable tool for quantitative analysis.5. This study was first applied the ANFIS in nondestructive determination of solid pharmaceutical samples. Furthermore, the PCA technique is applied to extraction relevant features from the lots of spectral data in order to reduce the input variables of the ANFIS. The generated scores of the principal components (PCs) subsequently were used as the input variables of the ANFIS instead of the spectra data and constitute the principal component analysis-adaptive neuron-fuzzy inference system (PCA-ANFIS) model. Experiment results show that the PCA-ANFIS model was efficient for nondestructive determination of solid pharmaceutical samples.Through the research of four different drugs, we know that the methods of RBF, PCA-RBF, GA-RBF and ANFIS for nondestructive quantitative analysis of solid pharmaceutical samples on NIR spectroscopy can get the accurate results, and they have widely prospect and applied value in the pharmaceutical analytical field.