Noninvasive monitoring of blood/tissue glucose concentrations has been successfully accomplished in individual diabetic subjects by using near-infrared (NIR) spectroscopy coupled with chemometric methods. Three different spectrometer configurations were tested: a) a Fourier-transform infrared spectrometer with an indium antimonide detector; b) a grating monochromator equipped with a silicon (Si) array detector, without fiber optics; and c) a grating monochromator equipped with an Si detector, with fiber-optic sampling. NIR spectra were obtained from diabetic subjects by transmission through the finger during a meal-tolerance test. The maximum range of observed plasma glucose concentrations obtained from the blood samples was 2.5-27 mmol/L. The NIR spectra were processed by using the chemometric multivariate calibration methods of partial least squares and principal component regression. The best calibration yielded a cross-validated average absolute error in glucose concentration of 1.1 mmol/L. This predictive ability suggests that noninvasive glucose determinations by NIR/chemometrics is a viable analytical method.
Noninvasive monitoring of glucose in diabetic patients is feasible with the use of near-infrared spectroscopic measurements. As a step toward the final goal of the development of a noninvasive monitor, the near-infrared spectra (4250 to 6600 cm−1) of glucose-doped whole blood samples were obtained along with reference glucose values. Glucose concentrations and spectra of blood samples obtained from four subjects were subjected to multivariate calibration with the use of partial least-squares (PLS) methods. The cross-validated PLS standard errors of prediction for glucose concentration based on data obtained from each individual subject's blood samples averaged 33 mg/dL over the range from 3 to 743 mg/dL. Cross-validated standard errors for glucose concentration from PLS calibrations based on data from all four subjects were 39 mg/dL. However, when PLS models based upon three subjects' data were used for prediction on the fourth, glucose prediction abilities were poor. It is suggested that blood chemistry differences were sufficiently different for the four subjects to require that a larger number of subjects be included in the calibration for adequate prediction abilities to be obtained from near-infrared spectra of blood from subjects not included in the calibration.
The multivariate calibration method of partial least-squares (PLS) was applied to the mid-infrared spectra of whole blood for quantitatively determining blood glucose concentrations. Separate calibration models were developed on the basis of spectra of whole blood obtained from six diabetic subjects from either in vitro glucose-supplemented blood or blood obtained from the same subjects in the post-prandial state during meal tolerance tests. The cross-validated PLS calibrations yielded average errors in glucose concentration of 11 and 13 mg/dL, respectively. It is desirable to use the calibration models based on the in vitro glucose-supplemented blood for determining glucose concentrations in unknown blood samples. However, when these multivariate calibration models based upon in vitro blood spectra were applied to the spectra of the postprandial blood samples, a subject-dependent concentration bias was observed. The source of this bias was not identified, but when the glucose determinations were corrected for the bias, average concentration errors were found to be 14 mg/dL. Changes in spectrometer design or calibrations based on large numbers of subjects are expected to eliminate the presence of this bias. If these measures do not succeed in eliminating the bias, then methods are demonstrated that significantly reduce the bias while retaining the sensitive outlier detection capabilities of the PLS methods. These latter methods require that the infrared spectrum and reference glucose levels be obtained from a single blood sample from each subject.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.