Advances in near-infrared glucose monitoring using pure component selectivity analysis for model characterization and a novel digital micromirror array spectrometer

Near infrared (NIR) spectroscopy is being developed for the measurement of glucose in human subjects for the treatment and management of diabetes. The attractive features of NIR spectroscopy are the nondestructive nature of the measurement and the unique spectral signature of glucose in living tissue. Methods are developed to improve calibration robustness and measurement reliability.; The pure component selectivity analysis (PCSA) is introduced and evaluated as a method to characterize the selectivity of a multivariate calibration model. Simulated spectra are generated for a series of ternary mixtures of glucose, lactate, and maltose with different levels of concentration correlations between glucose and either lactate or maltose. The PCSA of calibration models generated with different levels of concentration correlation demonstrate the utility of this method to identify and quantify such correlations.; Path length optimization is a critical element to define functional noninvasive calibration models in both simple and complex sample matrixes. A method is proposed and characterized for establishing the ideal optical path length for partial least squares calibration models of glucose in a simple buffer system. Path length optimization is guided by a signal-to-noise ratio term that is based on the net analyte signal (S/N)nas. Predicted optimal path lengths exhibit improved agreement with experimental findings compared to alternative methods reported in the literature. Predicted optimal path lengths are 1.55 and 6.13 mm for models based on combination and first overtone NIR spectra, respectively.; A novel solid-state spectrometer system is also developed for collecting NIR first overtone spectra. The centerpiece of this spectrometer is a digital micro-mirror array device (DMD) that permits programmed selection of transmitted wavelengths. Analytical performance of our first generation spectrometer is characterized by collecting and analyzing spectra based on the Hadamard transformation as well as Gaussian shaped bandpass filters. Analytical measurements are demonstrated with both types of spectral measurements in a simple set of glucose/lactate binary mixtures. Hadamard spectra result in calibration models with standard errors of 1.3 and 0.7 mM for glucose and lactate, respectively, while filter data provided measurement errors of 1.59 and 1.38 mM for glucose and lactate, respectively.