A Matlab Based Toolbox For Development Of Infrared Spectra Model And Its Application In Traditional Chinese Medicine

A Matlab based toolbox for development of infrared spectra model was programmed in this dissertation, and infrared calibration models of two different research systems were developed conveniently using the toolbox, thus the content of Cu in TCM Lycium barbarum and Phe content in dried blood spot(DBS) of newborn feet could be determined rapidly.On the basis of SE(Software Engineering) mindset, five modules were included in the toolbox, which are spectra file save/load integrated module, preprocessing module(preprocessing of spectra and object data), spectra optimizing module, model developing module, and model evaluation module. Moreover, the toolbox provided a visual, intelligible, easy operating Graphical User Interface(GUI), which is easily to operate for users even those who without any experience on infrared spectra model developing. With this toolbox, the medical researchers can easily and comfortably handle the model development. Thus, their workload and labor will be considerably reduced.(1) In the first, basic principle and procedure of quantitative calibration model developing was introduced, which include: preprocessing methods as normalization, smoothing and derivation; methods of optimizing spectral bands such as Uninformative Variables Elimination(UVE), consensus Uninformative Variables Elimination(c UVE), Backward interval Partial Least Squares(Bi PLS), Wavelet Transformation(WT) and Random Forest(RF); multivariate calibration methods including Partial Least Squares(PLS), consensus Partial Least Squares(c PLS), Support Vector Machine(SVM), Least Squares Support Vector Machine(LS-SVM), Back Propagation Artificial Neural Network(BP-ANN), Radial Basis Function Artificial Neural Network(RBF-ANN) and RF. Besides, the GUI design and programming realization method of the infrared spectra model developing toolbox with spectra file save/load integrated module, preprocessing module, spectra optimizing module, model developing module, and model evaluation module were introduced.Infrared calibration models of two different research systems were developed conveniently using the toolbox, which demonstrated the process of using and the advantage of the toolbox. Application 1: Quantitative calibration model developing of content of Cu in TCM lycia(Lycium barbarum). Raw spectra of lycia were opened in the toolbox, and was smoothed(1D9S), content data were logged in the same time. Then the spectral band was optimized using Bi PLS, in which 320 data points were selected(compressibility 18%), and PLS quantitative calibration model was developed with principal component number = 7, resulting Relative Coefficient(R), Root Mean Square Error of Prediction(RMSEP) and Mean Relative Error(MRE) were 0.92, 0.51 and 0.06 respectively. Application 2: Quantitative calibration model developing of content of Phe in DBS of newborn feet. The spectra data were compressed by wavelet transformation with wavelet function bior2.8, then the model was developed using RF, in which tree number = 1500, mtry = 26, 5 times, the result showed that R, RMSEP, MRE, Accuracy(Acc), Sensitivity(Sens) and Specificity(Spec) were 0.82, 142.09, 0.40, 94.31, 88.33 and 100 respectively. Moreover, 1537 data points were selected using c UVE, and were calculated in c PLS, the c UVE-c PLS model showed that R, RMSEP, MRE, Acc, Sens and Spec were 0.91, 96.80, 0.38, 93.14, 85.83 and 100 respectively. In addition, Bi PLS-c PLS model was also developed, and the optimal intervel number was 22, which correspond with antecedent researches. In this section, loading, integration, and saving of spectra file, preprocessing, spectral band optimization and model developing were demonstrated using the data above, showed that the toolbox was user-friendly, brief and clear and easily to operate, provided a simple tool for TCM IR spectra quantitative calibration model developing.