Backpropagation Artificial Neural Network for Determination of Glucose Concentration from Near-infrared Spectra

This paper proposes Backpropagation Artificial Neural Network algorithm as a calibration technique for prediction of glucose concentration. For the experimental work, simulated blood was formed by mixing only three constituents glucose, triacetin, and Urea in a phosphate buffer solution. Spectra of this solution were obtained from a near-infrared spectrophotometer in the region of 2100nm to 2400 nm and the spectral resolution of 1nm was used to collect the spectra. The improvement in the standard error of prediction and correlation coefficients as compared to Principal Component Regression and Partial Least Square regression techniques demonstrate that Backpropagation Artificial Neural Network algorithm could be used as an alternate calibration technique for non-invasive glucose measurement and thereby could resolve one of the impediments for the translation of non-invasive glucose measurement into clinical practice.