Study Of Blood Glucose Measurement Model Based On Non-invasive Measurement Using Near-infrared Light

The number of diabetes patients in the world is increasing with each passing year and the patients showed a tendency of younger age.At present,no radical treatment of diabetes has been found clinically,and only conservative treatment can be adopted to control diabetes.Blood glucose detection plays a vital role in the prevention,diagnosis,and treatment of diabetes.At present,blood glucose concentration is mainly measured by photochemical or electrochemical analysis,which requires fingertip blood sampling.However,frequent invasive measurements can cause pain to patients,and they may be infected if they are not handled properly.Therefore,a non-invasive blood glucose concentration detection method is highly desired.Near-infrared spectroscopy has become one of the most promising methods in the field of non-invasive measurement of human blood glucose concentration due to its advantages such as no harm to the human body,strong ability to penetrate the skin and relatively low cost of equipment.The accuracy and stability of the detection model are the key factors for the clinical application of this method.Therefore,this paper mainly studies the detection model of blood glucose based on near-infrared non-invasive detection.(1)Considering the strong scattering of human tissues,the complexity of blood components,the interaction between blood components,and the regularity of fluctuations in human blood glucose concentrations,the nonlinear auto regressive model with exogenous input(NARX)was introduced in this paper.Considering that the near-infrared light signals detected in vitro are easily affected by environmental and physiological parameters.In addition to near-infrared absorbance,ambient temperature,ambient humidity,blood pressure(systolic blood pressure,diastolic blood pressure),pulse rate and body temperature were also introduced as initial input variables.The sensitivity analysis(SA)method was employed to select the input variables.Finally,four variables(near-infrared absorbance,systolic blood pressure,pulse rate,and body temperature)were selected as input variables of the NARX model,that is,the 4Vs-NARX model.The results show that the model has good prediction performance,of which the root mean square error(RMSE)of the prediction is 0.72 mmol / L and the correlation coefficient(CORR)is0.85.The proportions of the prediction results falling in the regions A and B of the Clark error grid analysis are 90.27 % and 9.73%,both meet the clinical requirements.(2)The NARX model requires two invasive blood glucose measurements before the non-invasive blood glucose concentration measurement is started,which is not convenient for the application scenarios of multiple people’s single or multiple blood glucose measurement,such as hospital clinic and health examination.Therefore,the back propagation neural network(BPNN)was introduced as detection model.Aiming at the disadvantage that the traditional gradient descent training algorithm tends to fall into the local optimal,the particle swarm optimization(PSO)method was introduced to train the BP model,that is,4Vs-PSO-BP model.The results showed that the prediction performance of this model was better than that of the traditional BP model,in which the root mean square error of prediction was 0.95mmol/L and the correlation coefficient was 0.74.The Clark error grid analysis results showed that the proportion of the predicted results falling into region A and region B was 84.39% and 15.61%,both meeting the clinical requirements.The model can quickly measure the blood glucose concentration of the subject,and has relatively high accuracy.