Multivariate calibration of NIR spectroscopic sensors for continuous glucose monitoring

Abstract

Diabetes, a disorder in the control of blood-glucose levels, is one of the most serious metabolic diseases worldwide. Among the investigated technologies for continuous glucose monitoring (CGM), near-infrared spectroscopy (NIR) has received the most attention. There have been many attempts to develop NIR-based CGM systems with promising in-vitro results, but they lacked robustness for in-vivo use. We critically review the application of chemometrics for CGM and the research needed. Pre-processing and multivariate-calibration techniques, which allow exploiting expert knowledge on the potential interferences, are possible solutions. The combination and first overtone bands in the ranges 2050-2300 nm and 1500-1800 nm, respectively, are the most informative regions. We therefore recommended selecting the most informative variables and exploiting the available expert knowledge on known interferences in pre-processing or multivariate calibration to develop an NIR-based CGM sensor for in-vivo use.

Logistic regression analysis for identifying the factors affecting development of non-invasive blood glucose calibration model by near-infrared spectroscopy

“…Although several companies to date have introduced various products, they lack the precision and specificity of a blood glucose meter [4]. The near-infrared (NIR) spectroscopy has received the most attention, however many attempts to develop NIR-based techniques lacked robustness for in vivo use [5]. The fundamental problem is that the glucose specific signals are too small to accurately distinguish from the background absorption of the tissue components such as water, fat or protein.…”

Modeling research on wheat grain in different quality categories using multilayer support vector machine

“…Partial least squares based dimension reduction (PLSDR) is a high performance method for reducing dimensionality of complex data [9][10]. It uses a supervised mode to extract the potential characteristic variables from the original data space instead of original spectral data, which not only reduces the modeling time but also improves the accuracy of identification.…”

Partial Least Square with Savitzky Golay Derivative in Predicting Blood Hemoglobin Using Near Infrared Spectrum

“…Moreover, PLS has shown much better performance compared to artificial neural network (ANN) in term of RMSEP [24]. However, conventional PLS model need to use prior preprocessing step result to confront with the change in interferent structure in the test set and reducing the prediction error [8].…”