In vitro model for human noninvasive blood glucose measurements

An in vitro model is developed which accurately simulates the spectral features of human webbing tissue in the near infrared (NIR) spectral range between 7000 and 5000 cm{dollar}sp{lcub}-1{rcub}.{dollar} This model is used to answer several key questions about a human noninvasive blood glucose measurement.; The in vitro model is prepared by combining sequential layers of animal tissue and buffered solution inside the sample compartment of a FT-NIR spectrometer. A two component model of fat and water is found to adequately simulate the spectral features of human webbing tissue. Spectra from different human subjects can be simulated by adjusting the thicknesses of the layers.; The effect of aqueous layer thickness and spectral noise upon prediction errors from partial least squares (PLS) models is studied. Six data sets of the in vitro model are collected consisting of all combinations of two aqueous layer thicknesses and three source intensities. PLS models are optimized for each data set and prediction errors are found. Plots of prediction error versus RMS noise for 100% lines are presented that can be used to estimate model performance from a spectrometer configuration or subject without a lengthy noninvasive data collection. Scattering particles are added to the aqueous layer and prediction errors are compared to the models without scattering.; A data set is collected using the model without glucose in the aqueous layer. This phantom data set is used to study three important unknowns. The phenomenon of seemingly valid PLS models when there is no analyte is investigated. The effect of correlations between the analyte and a known absorbing species upon PLS prediction errors is found. Finally, the phantom data set is used to compare PLS validation procedures.; A spectrometer is modified to yield high quality spectra of human tissues. The tongue is found to be superior to the webbing for a human noninvasive measurement. For the first time, glucose is measured noninvasively in the human body using a valid calibration model.