Study of a noninvasive blood glucose detection model using the near-infrared light based on SA-NARX

Accumulated attempts have been made to develop models for near-infrared noninvasive measurement of human blood glucose concentration. Most of them focus on the relationship between near-infrared absorbance and blood glucose concentration, but do not consider the fluctuation regularity of blood glucose concentration and the influence of environmental factors and human physiological state on near-infrared absorption. In order to improve the performance of prediction model, a hybrid method is proposed in this paper. The nonlinear autoregressive model with exogenous input (NARX) was introduced as prediction model. 7 variables, including 1550 nm near-infrared absorbance, ambient temperature, ambient humidity, systolic pressure, diastolic pressure, pulse rate and body temperature were introduced as initial input variables. The sensitivity analysis (SA) method was employed to select the relative important input variables for NARX model. Based on the result of SA, a robust and accurate NARX model with 4 input variables (1550 nm near-infrared absorbance, systolic pressure, pulse rate and body temperature) was derived. Compared with the back propagation neural network (BPNN) with the same selected 4 input variables and the BPNN with initial 7 input variables, the NARX model developed there showed better prediction performance, of which the root mean square error and correlation coefficients were 0.72 mmol/L and 0.85 respectively for the 10-fold cross validation set. The percentages of the 10-fold cross validation set samples falling in region A and B of the Clarke error grid analysis were 90.27% and 9.73% respectively. These results demonstrate the potential of our model for noninvasive measurement of blood glucose concentration. (C) 2019 Elsevier Ltd. All rights reserved.