Identification of a selective glucose spectral signature for noninvasive near infrared measurements with multivariate calibration approaches

Near infrared (NIR) spectroscopy is proposed as a noninvasive method for measuring in vivo glucose concentrations in individuals with diabetes. Such measurements involve transmitting a harmless band of NIR light through skin tissue and extracting glucose concentration from the resulting spectral information. Selectivity is a major requisite for such measurements because of the physical and chemical complexities of the in vivo skin matrix. The principal objective of the research detailed in this dissertation is to establish the selective basis for noninvasive NIR glucose measurements in living skin tissue.; Initial experiments focus on the measurement of glucose in a simple aqueous buffer matrix. Key issues addressed in this initial work include spectral range, net analyte signal (NAS), and optical path length. Results indicate superior results are obtained in the combination spectral range with an optical path length of 1.5 mm. The NAS is computed to provide the selective component of the glucose spectrum relative to instrumental background spectral variance.; Selectivity is then assessed for the measurement of glucose in a more complicated matrix composed of glucose, glucose-6-phosphate, and pyruvate. This matrix includes the principal species of the glycolytic pathway. Again, selective glucose measurements are demonstrated by using the partial least squares (PLS) algorithm and the NAS vector. Selectivity is further demonstrated by comparing the resulting calibration vectors and by using the principal component selectivity analysis.; Accurate glucose measurements are also demonstrated for direct measurements in undiluted cultivation medium for Tn5 insect cells. The impact of concentration correlations on the PLS calibration performance is examined and a comparison between PLS and NAS calibration vectors demonstrate the existence of glucose specific information in these models.; The distinguishing selective spectral signature for glucose is extracted from in vivo NIR spectra collected noninvasively from an animal model. An experimental procedure is developed to expose this NAS for glucose and these results permit real-time, absolute glucose measurements in living tissue. Similar results are described for urea, thereby demonstrating the generality of this approach. The concept of measuring glucose in one animal based on calibration information derived from a different animal is demonstrated.