Study On Theories And Experiments Of Signal Structure And Acquisition For Non-Invasive Blood Glucose Sensing With Near-Infrared Spectroscopy

As a global disease, diabetes and its complications are major threats to the health and life of human beings. Since the non-invasive blood glucose sensing technique has many advantages such as free from pain and no risk of infection, it's very important to relieve the pain of diabetes and realize continuous and real-time monitoring of blood glucose concentration. This paper focuses on non-invasive blood glucose concentration sensing with near-infrared (NIR) diffuse reflectance spectroscopy.From the point of view of signal source, the structure of near-infrared diffuse reflectance spectroscopy, which is used for blood glucose concentration sensing, is analyzed for the first time. And the effect of the noise from the measurement process on the prediction error of multivariate model is quantified based on the custom-built near infrared instrument. The results show that, the error induced by the noise from measurement conditions is most significant and the effect of baseline drift and Johnson noise can be almost ignored.The chance correlation in NIR spectroscopy and its effect on multivariate model are conducted. The probability of chance correlation in PLS model is much lower than that in MLR model, and it can be ignored when the criterion for acceptable model is RC 2V >0.8. For the in vitro experiments, the chance correlation induced by instrument-based factors can be avoided or reduced by sampling randomly and background corrected, and correct experiment design is helpful for getting rid of the chance correlation between different chemical components. Furthermore, a method based on the subspace spanned by background spectra is proposed to calculate the net analyte signal of glucose, which is free from reference glucose concentration error and chance correlation and can be used for identify the validity of complicated multivariate model.The optimal pathlength in NIR spectroscopy quantitative analysis is investigated. The theory deduction and experiments prove that the combined optimal pathlength (COP) method can greatly improve the prediction accuracy. And the mean pathlengths of photons in different layers of skin are estimated by Monte Carlo simulation. The result show that, the mean pathlength in the dermis is close to the...