Near-infrared Noninvasive Blood Glucose Detection Based On Isoluminous-energy Wavelength And Random Forest

Diabetes is the most common chronic disease of the 21st century and it seriously threatens human health.Self-monitoring and precise management of blood glucose levels is very important to diabetics.Compared with invasive blood glucose detection methods,near-infrared non-invasive blood glucose detection technology could not only eliminate pain of patients and reduce infection,but also achieve continuous real-time glucose monitoring.The main technical difficulty of glucose detection is that near-infrared spectrum is very sensitive to the change of samples and external interference,which leads to lower prediction accuracy.In order to eliminate the influence of changes in samples and external interference and establish a more accurate correlation model,a near-infrared non-invasive blood glucose detection scheme is proposed.Instead of background-elimination method and partial least squares regression algorithm,isoluminous-energy wavelength method and random forest regression algorithm are used to correct spectrum and establish correlation model.The dissertation includes the following five parts:First,the significance,present situation and technical difficulties of near-infrared noninvasive blood glucose detection are introduced.Secondly,based on the interaction between light and biological tissue,Boltzmann radiative transfer equation,Monte Carlo simulation and near-infrared spectroscopy are introduced as the theoretical basis of the dissertation.Thirdly,the existence of the isoluminous-energy wavelength and the spectral correction method for scattering medium are deduced.Meanwhile,the existence and effect factors of the isoluminous-energy wavelength are simulated through Monte Carlo simulation.Fourthly,the traditional chemometrics methods,random forest regression method and methods for model evaluation are introduced.Meanwhile,the partial least-squares regression model and the random forest regression model are built and evaluated by Monte Carlo simulation.Finally,based on the near-infrared noninvasive blood glucose detection scheme,an experimental system for acquisiting diffuse reflectance spectrum is set up.The validity of the isoluminous-energy wavelength in spectral correction and the performance of random forest regression model are verified through imitative experiment based on the system.The results show that the performance of the near-infrared non-invasive blood glucose detection scheme has been successfully improved.Thus,all the experimental results are corresponded with the theory.