Neural network based spectral analysis of multicomponent mixtures for glucose determination

The application of artificial neural network based analysis of infrared spectral data for accurate prediction of physiological glucose concentrations in multicomponent aqueous mixtures was assessed. Theoretical near-infrared spectra of aqueous solutions containing glucose, urea, albumin, and histidine were constructed by adding the digitized and physiologically scaled spectra of these components. An artificial feedforward neural network was first trained with a backpropagation learning scheme using these spectra. The network was then used to predict unknown glucose concentrations in new spectra. Random noise was added to the composite spectra to determine its effect on the performance of the network. The effect of different network parameters was also evaluated in terms of the prediction ability of the network. Preliminary results indicate that measurement of glucose concentrations in multicomponent aqueous solutions is feasible using a neural network approach