Effects of temperature on near-infrared spectroscopic measurement of glucose

Abstract

The noninvasive monitoring of sugars and, in particular, glucose using near-infrared (NIR) spectroscopy would be useful for a number of applications including regulating the nutrients in cell culture media, monitoring on-line processes in the food industry, and in vivo monitoring for control of glucose in diabetic patients. The focus of this research was the investigation of the temperature effects across a 10.6 to 40.4 'C range on Fourier filtered and unfiltered single-beam as well as absorbance NIR spectra (2.0-2.5 Jim) of glucose solution. It is known that the positions of water absorption bands centered at 1.923 and 2.623 um depend heavily on temperature effects while the glucose bands (-2.105, 2.237, and 2.326 um) are temperature insensitive.'O," The water absorption bands were shown to shift to longer wavelengths while the bandwidth (the distance between two bands at 1.5 absorbance units, including three glucose absorption peaks) decreased with decreasing temperature. A simple linear regression model for absorbance at 2.273 um as a function of glucose concentration showed high linearity across the temperature range. Partial least squares (PLS) calibration models with glucose single-beam and absorbance spectra were constructed at five different temperatures. The SEP values in the calibration model built with unfiltered glucose single-beam spectra decreased, providing a trend across the temperature range, as the sample temperature approaches the temperature used for the calibration set. The use of Fourier filtered single-beam spectra reduced the SEP but still showed an increase in SEP as large temperature differences were produced. When either filtered or unfiltered glucose absorbance spectra, with the reference taken at the same temperature, were used with PLS regression, the SEP was generally lower than the filtered single-beam spectra in each case. When absorbance spectra were used with the reference scans taken at the same temperature as the sample scans, there was no significant difference and no pattern in the standard errors of prediction (SEP) for the validation data with temperature. The filtering technique had a minimal impact on the absorbance spectra, providing no significant difference between the filtered spectra and unfiltered spectra and in some cases, an increase in the SEP.