Study On Quantitative Calibration Model For IR Spectroscopy And It’s Application In Medical And Pharmaceutical Detection

Having advantages of fast analysis, easy preprocessing, non-destructive and pollution-free procedure, Fourier Transform Infrared Spectroscopy- Attenuated Total Reflection(FTIR/ATR) is widely used in medicine and food industry. However, it is an indirect measurement method, whose performance depends on mathematical models. Therefore, the improvement of predictive capability and stability of the model is the key point in the research of this method. In this dissertation, various Chemometrics methods were adopted in constructing the calibration models of FTIR-ATR, which were then applied for rapid screening of neonatal phenylketonuria(PKU) and rapid detection of copper content in traditional Chinese medicine Lycium barbarum.1. Screening of PKU: Spectrum data of dried blood spot(DBS) samples were collected using infrared spectroscopy(IR), taking air as blank(background). The raw spectra were analyzed with several preprocessing methods, and the 9-point smoothing coupled with first differential(1D9S) method was finally selected to develop the quantitative calibration model. Several calibration models were developed by PLS, c PLS, UVE-c PLS, SVM, GA-LSSVM and RF-RBF-ANN respectively, and were evaluated by correlation(R), root mean square error of cross-validation(RMSEC), root mean square error of prediction(RMSEP), mean relative error(MRE) and predictive accuracy(Acc). The Phe quantitative calibration model developed by UVE-c PLS method was found to be superior to other methods, with R, RMSEP, MRE and Accbeing 0.88, 102.61, 0.31 and 100, respectively. Quantitative calibration model for Phe/Tyrosine(Tyr) ratio was also developed in the same way.The raw spectra and the first differential(1D) spectra were selected to develop models, which were developed by SVM, UVE-c PLS and RBF-ANN, and by SVM, UVE-c PLS, GA-LSSVM, RBF-ANN and RF-RBF-ANN respectively. The results showed that UVE-c PLS method was found to be superior to other methods, with R, RMSEP, MRE and Accbeing 0.80, 4.93, 0.30 and 100 2.Determination of the content of Cu in Lycium barbarum: The raw spectra, the 1D5 S spectra and the vector normalization(VN) spectra were selected to develop the calibration models, by GA-LSSVM, SVM, RBF-ANN, RF-RBF-ANN and UVE-c PLS respectively, and were evaluated by R, RMSEP, MRE and Acc. The results showed that, the performances of the models preprocessed and developed by VN was better than the other two, and the RBF-ANN and UVE-c PLS models were optimal, with R, RMSEP and MRE were being 0.87, 0.57, 0.02 and 0.88, 0.51, 0.05, respectively.