Characterization and quantification of light scattering in protein and glucose sample solutions by near-infrared transmission spectroscopy

Light scattering is a common feature of many biological matrices. Development of near infrared spectroscopy for noninvasive measurements in biological matrices must consider the impact of sample scattering on measurement accuracy and reliability. To this end, the ability to generate functional multivariate calibration models from near infrared transmission spectra collected through highly scattering samples is investigated in a systematic manner.; Lysozyme, a common protein, is the analyte of interest in the first series of experiments and polystyrene latex microspheres are the light scattering agent. Near infrared spectra were collected over the 5000–4000 cm −1 (2.0–2.5 μm) spectral range for a series of sample solutions composed of lysozyme and microspheres with diameters of either 0.6 or 6.4 μm. Partial least squares calibration models were evaluated for both sizes of beads. In both cases, accurate models are possible in these highly scattering solutions. Prediction errors are approximately a factor of two larger, however, compared to analogous models generated from non-scattering solutions.; Calibration models are also evaluated for measuring glucose in solutions designed to mimic the basic scattering properties of human skin tissue. In this set of experiments, 1.0 μm polystyrene latex microspheres serve as the scattering bodies and calibration models are generated for measuring physiological levels of glucose. Again, accurate calibration models are demonstrated with prediction errors of 2.0 mM from spectra collected over the 5000–4000 cm−1 spectral range.; The ability to measure glucose directly in blood is examined by evaluating calibration model generated from near infrared transmission spectra collected through standard solutions of glucose with varying amounts of freshly isolated red blood cells. Solutions are prepared with hematocrit levels from 14 to 50% and glucose concentrations from 1 to 30 mM and spectra are collected over the 5000–4000 cm−1 spectral range. The resulting prediction errors are 1.7 mM for the best models. Furthermore, a detailed investigation determines that the principal impact of scattering is to reduce the overall radiant power at the detector, which reduces the spectral signal-to-noise ratio, thereby degrading model performance.