Blood Glucose Detection Experiments And Data Processing Based On Mid-infrared ATR Spectroscopy

The determination of the blood glucose level is a necessary procedure in diabetes therapy. Mid-IR ATR spectroscopy not only provides an economic way to avoid the shortcomings of biochemical test methods, reduce material wastage and speed up measurement, but also it is a potential method of non-invasive blood glucose measurement.In this paper, basic glucose absorption characteristics in the mid-infrared spectrum were studied systematically based on ATR technique and the major glucose absorption characteristics in the infrared band (900~1200 cm-1) was identified, including the peak position and the corresponding absorbance. Based on this study, method of continuous and discrete mid-infrared ATR spectroscopy monitoring blood glucose was test, and a variety methods of data pre-processing and data modeling were studied and compared.The paper consists of three aspects as follows:Firstly, a series of spectral experiments have been carried out to investigate this method. Initially, the experimental conditions, such as finger pressure, location and skin were studied, and proved that the mid-infrared ATR spectroscopy glucose detection method is feasible, and obtained preliminary experimental results. Meanwhile, we use squalane oil and study the relevance of ATR spectra and blood glucose over oil diffusion time for squalane oil can strength finger skin moisture. In addition, the glucose absorption wavelength characteristic in the mid-infrared spectrum was studied specially with the use of CO2 laser which light wavelength source is tunable.Secondly, by comparing a variety method of data preprocessing, we use singular value removing, internal standard correction, normalization, denoising methods such as wavelet smoothing to preprocess the original spectrum, and to enhance correlation of finger mid-infrared ATR absorption spectrum and glucose concentration.Thirdly, data modeling methods were studied. In PLS model, the data correlation value increased from 0.67 to 0.95, and the RMSEP value decreased from 17.6894mg/dl to 11.2287mg/dl, while using ANN model, the RMSEP value decreased from 3.1339mg/dl to 1.8961mg/dl. The predictability of models is better when only use glucose peaks absorbance data to analysis. And when the data is strong nonlinear, the predictability of ANN algorithm is better than PLS.As the prediction accuracy was improved by using pretreatment methods and two kinds of prediction model as PLS and ANN that our work is valuable for the achievement of non-invasive blood glucose measurement.