Determination of glucose in a synthetic biological matrix with decimated time-domain filtered near-infrared interferogram data

Fourier transform near-infrared interferogram data are used to determine the physiological levels of glucose in a synthetic biological matrix consisting of varying levels of bovine serum albumin and triacetin in a pH 7.4 phosphate buffer. Finite impulse response (FIR) digital filters are applied to short segments of the interferogram data, and the resulting filtered interferogram intensities are used as independent variables in constructing multivariate calibration models based on quadratic partial least-squares regression analysis. To increase the performance characteristics of the filters, the collected interferograms are decimated to reduce their frequency bandwidth. By use of a combination of optical and digital filtering, the spectral bands in the region of 4000 to 5000 cm(-1) are aliased to the range of 0 to 1975 cm(-1). This allows the attenuation in the stopbands of the filters to be increased from 7 to 50 dB, thereby increasing their selectivity. By use of FIR filters centered on the glucose C-H combination band at 4397 cm(-1) and the triacetin C-H combination band at 4446 cm(-1) calibration model for glucose is constructed that achieves a standard error of calibration of 0.484 mM and a standard error of prediction of 0.618 mM over the 1-20 mM concentration range. This prediction performance represents a 20% improvement over the performance of a model built with the full-bandwidth data. The improved model is also observed to outperform a similar calibration model based on the conventional analysis of absorbance spectra. (C) 2000 Elsevier Science B.V. All rights reserved.