Spectral processing tools to improve and characterize near infrared spectroscopic measurements of glucose in complex biological matrixes

Water displacement factors for alanine, ascorbate, glucose, lactate, triacetin and urea were determined. Molar absorptivities for water and these six biological molecules were determined in the first overtone and combination spectral ranges.; Analytical performance of the first overtone and combination spectral ranges was compared using eighty uniquely prepared aqueous mixtures of alanine, ascorbate, glucose, lactate, triacetin and urea. Partial least squares (PLS) calibration models were generated and optimized for each component. Calibration models in these two spectral regions were compared in terms of standard errors of calibration and prediction and mean percent error. Results indicate models from combination spectra out perform those from first overtone spectra.; Source information for the calibration model was studied through a pure component selectivity analysis, comparison between regression vectors and net analyte signal vectors. Figures of merit such as selectivity, sensitivity and limit of detection for each calibration model were characterized through these net analyte signal vectors.; The effect of various path lengths and temperatures on near infrared (NIR) spectra was studied. A path length prediction algorithm was used to determine the path length for unknown samples. A glucose PLS calibration model was generated with combination spectra and a path length correction method was applied to compensate for the influence of path length on model prediction results. A temperature calibration model was proposed based on temperature-induced spectroscopic variation of water. Finally, a universal glucose calibration model was evaluated to demonstrate acceptable compensation for variations in temperature and path length.; Glucose calibration models were generated with NIR spectra from whole bovine blood. Spectra were collected in the first overtone and combination spectral regions. A glucose calibration model was generated using computed spectra with and without glucose. An apparent working glucose calibration model was obtained from a glucose-free data set. Correlation between the concentrations of glucose and fat is responsible for the apparent working glucose calibration model. Evidence is provided, however, that PLS models generated from whole blood spectra are based on glucose-specific information with no contribution from dissolute fat. This result suggests that an effective validation method is necessary for the application of NIR spectroscopy for noninvasive glucose measurements.