Quantitative Analysis Research Of Non-Invasive Blood Glucose Measurement Based On Mid-Infrared Absorption Spectroscopy

The number of diabetics has increased in terms of the research of the international diabetes federation,increasing the demand for blood glucose concentration measurement equipment in many countries.At present,medical institutions have developed minimally invasive or non-invasive blood glucose concentration measurement equipment.However,these types of equipment have deficiencies in usability and measurement accuracy.Mid-infrared absorption spectroscopy,where it detects the fundamental frequency vibration of the molecular structure,is advantageous for improving the sensitivity of the measurement.However,the development of non-invasive equipment based on mid-infrared has been hindered in terms of three factors:the serious overlapping of peaks,the vulnerability to environmental or human factors,and uninformative wavenumbers.Therefore,this thesis deeply studies the blood glucose measurement method based on mid-infrared absorption spectroscopy.The main works are summarized as follows:（1）Study on outlier detection methods in non-invasive mid-infrared blood glucose concentration measurements.In order to solve the problem that outlier determination methods can be affected by the threshold value,which leads to a misjudgment of outliers.This thesis proposes dual check outlier determination method（DC）.Firstly,DC utilizes the sensitivity of prediction residual to outlier to screen outlier preliminarily,improving the stability of the data set.Secondly,verify the optimal threshold value of outliers based on the square of the mahalanobis distance approximately obeys the chi-square distribution to determine the outlier adaptively.This method is compared with a variety of outlier determination methods to validate the effectiveness and accuracy.（2）Study on wavenumber selection methods in mid-infrared noninvasive blood glucose concentration measurements.In order to further improve the accuracy of wavenumber selection methods based on model population analysis,L₁ regularization and tournament selection methods are introduced to improve the quality of optimal models,avoiding the overfitting and the reduction of exploration ability.Meanwhile,in order to solve the problem of the low efficiency of model population analysis,the absorbance concentration ratio is proposed to modify the weight of wavenumbers,guiding the wavenumber selection method to the optimal solution and reducing its cost.This method is compared with a variety of wavenumber selection methods to validate the accuracy and robustness.（3）Study on calibration model methods in mid-infrared noninvasive blood glucose concentration measurements.The maximum information coefficient is introduced to measure the nonlinear correlation of spectra in terms of the characteristics of mid-infrared spectra.In order to solve the problem of cross-distribution of informative wavenumber,uninformative wavenumber and redundant wavenumber in spectra,flexible interval division method is proposed to concentrate the informative wavenumber,reduce the multicollinearity,and retain the synergy.Therefore,calibration sub-models accurately estimate the distribution of glucose concentration information in spectra,which improves the accuracy of the final fusion model.This method is compared with a variety of calibration model methods to validate the accuracy and robustness.