Wavelength Selection And Stability Analysis Of NIR Spectroscopy For Human Blood Glucose

Diabetes is a kind of high incidence, harm of serious diseases. There is no effective radical measures at present. Patients must frequently detect glucose in order to control the illness. Existing blood glucose detection method used vein exsanguination and through biochemical reactions to get the glucose concentration.This method likely to cause infection and environmental pollution.Near-infrared spectroscopy technology is a new research branch of non-invasive (or no reagent) glucose testing and high signal-to-noise ratio waveband selection of blood glucose is an important basic research work. This dissertation through the near-infrared experimental research analysis of the glucose aqueous solution sample and the serum sample, optimize the near-infrared blood grucose testing waveband range and providing detection methods for non-invasive(no-reagent) blood glucose detection.This paper adopts Savitzky-Golay smoothing, MWPLS and ECMWMLR method to establish analysis models for the glucose aqueous solution sample and the serum sample. Using root mean square error of prediction (RMSEP) and correlation coefficient of prediction (RP) as model evaluation indicators to select high signal to noise ratio wavebands of the glucose aqueous solution sample and the serum sample.Considering the fluctuation of model parameters due to different divisions of calibration set and the prediction set, to get the stable prediction results, model parameters were optimized according to the average effect of 50 different divisions of calibration set and the prediction set. The results as follows:(1) The results of the glucose aqueous solution sample and the serum sample indicate that glucose information interval located in long wavelength near-infrared region (eliminating the absorbance higher than 3) 1100-1880 and 2082-2350nm.(2) Using MWPLS method to futher optimize wavebands in the above Spectral region. The results of glucose aqueous solution sample indicated that 1510-1848nm was optimal wavebands with high signal to noise ratio, and the corresponding PLS factor, RMSEP and RP were 8,0.290 mmol/L and 0.9994 respectively. The results of the serum sample indicated that 1556～1846nm was optimal wavebands with high signal to noise ratio, and the corresponding PLS factor, RMSEP and RP were 15,0.899 mmol/L and 0.8995 respectively.(3) Using ECMWMLR method to futher optimize wavelength combination in the above Spectral region. The results of the glucose aqueous solution sample indicated that the optimal wavelength combination:beginning wavelength, number of adopted wavelength and gaps of adopted wavelength were 1504 nm,6 and 42, and the corresponding RMSEPAve and RO,Ave were 0.249mmol/L and 0.9997. The results of the serum sample indicated that the optimal wavelength combination:beginning wavelength, number of adopted wavelength and gaps of adopted wavelength were 1558 nm,29 and 10, and the corresponding RMSEPAve and RP,Ave were 0.815 mmol/L and 0.9172.The above results provides valuable reference for establishing a high precision analysis model in human blood glucose and desining a special analysis instrument.