Methods Of Wavelength Optimization For Near-infrared Spectroscopy And Its Application In Analysis Of Hyperlipidemia Indicators

Hyperlipidemia is the major cause of many serious diseases. Total cholesterol(TC) and triglyceride (TG) are two important indicators of hyperlipidemia for clinicaldetection. The clinical detection method involves the different chemical reagents fordifferent indicators and the requirement for operation by professionals. Therefore, analternative reagentless and simultaneous analytical method is essentially required. Inrecent years, studies for detecting indicators of hyperlipidemia with near infraredspectroscopy (NIR) have been reported, but the methods were not perfect, theprediction accuracy has not yet meet the requirement of clinical applications.Improving the accuracy of the model has been a hot research in this area. Theobjective was to reaserch new chemometric methods for the high precision of NIRdetection in hyperlipidemia indicators (TC, TG). Through a large of spectrumexperiments and computer experiments, the models of high signal to noise ratiowavelength and high precision quantitative were established, and provided technicalsupport for large population screening of lipid.Chemometrics methods of NIR spectroscopy about biomedicine analysis abovewere researched systemic, include modeling the new stability system, researchingcontinuous (improved moving window partial least squares, MW-PLS) wavelengthscreening methods, researching quasi-continuous (equidistant combination multiplelinear regression, EC-MLR) wavelength screening methods, researching discrete(friend wavelength combination) wavelength screening methods, and designing thecomputer algorithm platform for model experiments. NIR Spectroscopic analysis ofZinc in the soil as an example, the feasibility of the established methodology wasexamined. Further the methodology was applied to NIR Spectroscopic analysis ofhuman serum hyperlipidemia indicators (TC, TG). The major conclusion of the thesisincludes the following:1. Based on randomness, similarity and stability, the new modeling systemincluding calibration set, prediction set and validation set was established, and thereliable analytical model was builed. Based on mulit-division of calibration set andprediction set, improved MW-PLS method, EC-MLR method and friend wavelengthcombination method with stability were estabilshed. The stability and equivalent ofspectral wavelength selection were researched as new issues. The methods were successfully applied in NIR Spectroscopic analysis of Zinc in the soil.2. The continuous NIR wavelength was optimizal selected using improvedMW-PLS method for human serum hyperlipidemia indicators (TC, TG). Throughmodel experiments, comparing the effect of the average forecast (M_SEPAve) andpredicted stability (M_SEPStd) of all continuous wavebands, the optimal wavebandand the quivalent optimal wavebands were determined. The optimal waveband of TCwas1556~1852nm, the number of equivalent optimal wavebands was53; Theoptimal waveband of TG was1540~1858nm, the number of equivalent optimalwavebands was390; the common equivalent optimal waveband of TC and TG was1556~1852nm, which was as common information waveband of the two indicators.In the validation set, TC and TG were validated by PLS model in the commonequivalent optimal waveband; the validated effects of spectrum predicted values andclinical measured values (V_SEP, V_RP) were0.184mmol L-1,0.988(TC) and0.097mmol L-1,0.997(TG), respectively. Further sensitivity, specificity and overalldiscrimination rate of spectrum predicted values for hyperlipidemia diagnostic were94.8%,86.8%and91.7%, respectively.3. The quasi-continuous NIR wavelength was optimizal selected using proposedEC-MLR method for hyperlipidemia indicators. Through model experiments,comparing the effect of the average forecast (M_SEPAve) and predicted stability(M_SEPStd) of all quasi-continuous wavebands, the optimal wavelength combinationand the quivalent optimal wavelength combinations were determined. Thewavelengths of TC were1658,1670,1682,1694,1706,1718,1730,1742,1754,1766nm, the number of equivalent optimal wavelength combinations was29; Thewavelengths of TG were1670,1682,1694,1706,1718,1730,1742,1754,1766nm,the number of equivalent optimal wavelength combinations was22; the commonequivalent optimal wavelengths of TC and TG were1658,1670,1682,1694,1706,1718,1730,1742,1754,1766nm, which was as common wavelengths of the twoindicators. In the validation set, TC and TG was validated by MLR model in thecommon equivalent optimal wavelength combination; the validated effects ofspectrum predicted values and clinical measured values (V_SEP, V_RP) were0.231mmol L-1,0.983(TC) and0.139mmol L-1,0.993(TG), respectively. Furthersensitivity, specificity and overall discrimination rate of spectrum predicted valueswere94.8%,89.5%and92.7%, respectively. 4. The discrete NIR wavelength was optimizal selected using proposed friendwavelength method for hyperlipidemia indicators. Through model experiments andcomparing the effect of the average forecast (M_SEPAve) and predicted stability(M_SEPStd) of all discrete wavelengths, the optimal friend wavelength combinationwas determined. There were23wavelengths in optimal friend wavelengthcombination of TC; There were24wavelengths in optimal friend wavelengthcombination of TG. In the validation set, TC and TG was validated by PLS model intheir optimal friend wavelength combinations; the validated effects of spectrumpredicted values and clinical measured values (V_SEP, V_RP) were0.244mmol L-1,0.979(TC) and0.143mmol L-1,0.993(TG), respectively. Further sensitivity,specificity and overall discrimination rate of spectrum predicted values were94.8%,92.1%and93.8%, respectively.Continuous, quasi-continuous and discrete models of NIR spectroscopic analysiswere established for detecting human serum hyperlipidemia indicators (TC, TG). Theresults showed that the correlation and consistence between spectrum predicted valuesand clinical measured values were well, and the result precision was better than theprevious. Spectral prediction for hyperlipidemia diagnosis had high sensitivity,specificity and overall discrimination rate. The method does not require chemicalreagents, accurate, fast and provided an analysis of potential tools for large populationscreening of lipid, provides valuable reference for designing optical splitting systemof special spectrum analysis instruments.