FTIR/ATR Spectroscopy Analysis Of Glucose And Total Cholesterol In Human Serum

Glucose and total cholesterol are important human biochemical indicators. They are measured by different chemical methods respectively now which require chemical reagents and produce pollution. In this paper, Fourier transform infrared (FTIR) spectroscopy with attenuated total reflection (ATR) techniques is applied to build up a reagent-free, direct, rapid and simultaneous quantification method of analyzing these indicators.Because human blood is a complex system with multi-component, the spectrum contain a variety of physical noises and interference outside meathured indicators when FTIR/ATR spectrum of sample are measured directly. To eliminate these noises and interferences while extracting indicators information effectively, chemometric methods must be researched such as samples selection of model and stability of model, spectral preprocessing modes, spectral waveband optimization and so on.Firstly, all the samples are divided into modeling set and test set independent modeling, all models are optimized in the modeling set. In order to get stable and reliable models, all results are obtained by multiple different divisions of calibration set and prediction set, and a new dividing method is proposed. Secondly, the spectral preprocessing and modeling platform is established based the Savitzky-Golay (SG) smoothing and partial least squares (PLS) to large-scale optimize SG Smooth modes and of PLS factors. Thirdly, based on the subtracted spectra between glucose aqueous solution and de-ionized water, absorption peaks are calculated in fingerprint area. Based on these absorption peaks and multiple linear regression (MLR) model, two discrete wavelength selection methods of serum glucose analysis are develped which are absorption peaks disturbance model and absorption peaks compensation model. Fourthly, continuous optimization model of serum glucose and total cholesterol are established by moving window partial least squares (MWPLS). On this basis, two new continuous wavelength selection model are developed which are compensation band model and multi-band combination model. Fifthly, absorbance and concentration of the sub-linear model is more reasonable considering the case of wide range of sample concentration. Chemical-segment combination model is established base on the sample concentration range. Finally, the independent test set is tested for model verification.The main results are as follows.(1) All 186 samples are divided into the modeling set (124 samples) and test set (62 samples). The whole spectrum of 4500-600 cm-1 internal cross-verification PLS model is constructed in the modeling set. The modeling samples are divided into calibration set (82 samples) and prediction set (42 samples) for a total of 50 times based on every forecast bias of internal cross-verification PLS model. All model optimization are base on these 50 divisions.(2) By extending smoothing modes from 117 to 406, plateform of SG smooth modes combining PLS factors are constructed, and the whole spectrum of 4500-600 cm-1 SG smoothing PLS model is calculated. The root mean square error of prediction of calibration set (C-RMSEP) and correlation coefficient of prediction of calibration set (C-RP) are 1.308 mmol/L and 0.782 respectively for serum glucose, and C-RMSEP and C-RP are 0.723 mmol/L and 0.823 respectively for serum total cholesterol. The results indicate that the optimal combination of SG smoothing model and the number of PLS factors can significantly improve the PLS model prediction.(3) Based on the subtracted spectra between glucose aqueous solution and de-ionized water, 5 absorption peaks:1150,1103,1078,1034,991(cm-1) are found in fingerprint area. Used these absorption peaks to establish absorption peaks disturbance model,the optimal wavelength combinations are 1140,1096,1084,1030,993 (cm-1), the corresponding C-RMSEP and C-RP are 1.164 mmol/L and 0.828 respectively. One wavelength is compensated based on peaks disturbance model,the optimal wavelength is 1063 cm-1, the corresponding C-RMSEP and C-RP are 0.981 mmol/L and 0.893 respectively.The optimal prediction effect of absorption peaks disturbance model and compensation model are obviously better than the one of the PLS model. The numbers of wavelengths adopted are only 5 and 6, and the complexity of the optimal model is reduced greatly.The results also provide a theoretical basis for design of small and portable human serum glucose spectrometer.(4) Compensation model based on MWPLS is adopted as analysis of serum glucose, the optimal band combination is 1389-908,2590-2544 (cm-1), the corresponding C-RMSEP and C-RP are 0.753 mmol/L and 0.931 respectively. Multi-band combination model based on MWPLS is adopted as analysis of serum total cholesterol, the optimal band combination is 2924-2642,1749-1466,1277-991 (cm-1), the corresponding C-RMSEP and C-RP are 0.356 mmol/L and 0.951 respectively. The optimal prediction effects are obviously better than the one of whole spectrum SG smoothing PLS model,and also superior to the traditional MWPLS model.(5) Analysis of serum glucose MWPLS model as an example, chemical-segment combination model is established base on the sample concentration range. Reference standards of human fasting blood glucose:less than 6.11 mmol/L for normal, greater than 7.00 mmol/L for high,3 models are established respectively.They are low concentrations (4.09-7.00 mmol/L) model, high concentrations (6.11-17.82 mmol/L) model, and all concentration (4.09-17.82 mmol/L) model. The optimal band are 1219-897,1165-910,1389-908 (cm-1) respectively, the corresponding C-RMSEP are 0.452,1.062,0.782 (mmol/L), the corresponding C-RP were 0.780, 0.908,0.925 respectively. Three integrated decision-making models for predicting values of samples are also established based on the above three models.(6) The prediction effect of serum glucose in test set is veryfied by chemical-segment combination model, the corresponding root mean square error of prediction of test set (T-RMSEP) and correlation coefficient of prediction of test set (T-RP) are 0.720 mmol/L and 0.953 respectively. The prediction effect of serum total cholesterol in test set is veryfied by multi-band combination model, the corresponding T-RMSEP and T-RP are 0.351 mmol/L and 0.955 respectively. The results showed that the prediction effects of serum glucose and total cholesterol are also very good.