Prediction of glucose in whole blood by near-infrared spectroscopy: influence of wavelength region, preprocessing, and hemoglobin concentration.

Measurement accuracy for predicting glucose in whole blood was studied based on near-infrared spectroscopy. Optimal wavelength regions, preprocessing, and the influence of hemoglobin were examined using partial least-squares regression. Spectra between 1100 and 2400 nm were measured from 98 whole blood samples. In order to study the influence of hemoglobin, which is the most dominant component in blood, 98 samples were arranged such that glucose and hemoglobin concentrations were distributed in their physiological ranges. Samples were grouped into three depending on hemoglobin level. The results showed that glucose prediction was influenced by hemoglobin concentrations in the calibration model. It was necessary for samples used in the calibration model to represent the entire range of hemoglobin level. The cross-validation errors were the smallest when the wavelength regions of 1390 to 1888 nm and 2044 to 2393 nm were used. However, prediction accuracy was not very dependent on preprocessing methods in this optimal region. The standard error of glucose prediction was 25.5 mgdL and the coefficient of variation in prediction was 11.2%.