Comparison between transmittance and reflectance measurements in glucose determination using near infrared spectroscopy

Glucose determination based on near-IR spectroscopy is investigated for reflectance and transmittance measurement. A wavelength range is 1100 to 2500 nm, which includes both the combination and overtone bands of glucose absorption. Intralipid solutions are used as samples, where glucose concentrations vary between 0 and 1000 mg/dl. Sample thickness for reflectance is 10 cm and 1- and 2-mm-thick samples are used for transmission. Partial least-squares regression (PLSR) analyses are performed to predict glucose concentrations. The standard errors of calibration are comparable between reflectance and 2-mm-thick transmittance. The reflectance method is inferior to the transmittance method in terms of the standard errors of prediction. Loading vector analysis for reflectance does not show glucose absorption features. Reflected light may not have enough information of glucose since a major portion of detected light has a short optical path length. In addition, prediction becomes more dependent on medium scattering rather than glucose, compared with transmission measurement. Loading vectors obtained from a PLSR transmittance analysis have glucose absorption profiles. The 1-mm-thick samples give better results than the 2-mm-thick samples for both calibration and prediction models. The transmittance setup is recommended for noninvasive glucose monitoring.