Abstract

A digital Fourier filter is combined with partial least-squares (PLS) regression to generate a calibration model for glucose that is insensitive to sample temperature. This model is initially created by using spectra collected over the 5000 to 4000 cm<sup>-1</sup> spectral range with samples maintained at 37°C. The analytical utility of the model is evaluated by judging the ability to determine glucose concentrations from a set of prediction spectra. Absorption spectra in this prediction set are obtained by ratioing single-beam spectra collected from solutions at temperatures ranging from 32 to 41°C to reference spectra collected at 37°C. The temperature sensitivity of the underlying water absorption bands creates large baseline variations in prediction spectra that are effectively eliminated by the Fourier filtering step. The best model provides a mean standard error of prediction across temperatures of 0.14 mM (2.52 mg/dL). The benefits of the Fourier filtering step are established, and critical experimental parameters, such as number of PLS factors, mean and standard deviation for the Gaussian shaped Fourier filter, and spectral range, are considered.

PDF Article

Cited By

OSA participates in CrossRef's Cited-By Linking service. Citing articles from OSA journals and other participating publishers are listed here.