Non-invasive assays for blood glucose can be based on near infrared spectrometry of skin tissue using the diffuse reflectance technique. Using a straightforward spectral variable selection based on choices from the optimum partial least-squares (PLS) regression vector yields better results than using PLS calibration models with full spectrum evaluation previously reported. The pairs of variables are selected from the maxima and minima of the regression weights, respectively, in decreasing order. Substantial improvements in the prediction performance of such calibration models, compared to previous calibrations based on full spectrum evaluation, are obtained. Another aspect is the reduced number of spectral variables needed for robust calibration modeling. In addition, evidence is provided for the physical effect, as manifested by the spectral glucose absorptivities, underlying the individual single-person calibration models. Their regression vector structure shows very similar features as calculated for a glucose calibration experiment based on random human plasma samples. Novel techniques are presented for probing the intravascular fluid space using time-resolved near infrared spectroscopy of oral mucosa. The pulsatile blood spectrum can be derived from these diffuse reflectance lip spectra by Fourier analysis. Future applications and prospects for non-invasive blood analysis are discussed.
Oscillator Noise Analysis