| Summary: | Near-infrared measurements of glucose in human plasma are performed using a custom, rapid, high-throughput filter-based spectrometer covering a spectral range between 2080 and 2315 nm. Quality of the measured glucose signals is quantified through the use of two figures of merit: selectivity and limit of detection. Selectivity measures the uniqueness of the glucose spectrum from among the interfering spectra. Limit of detection measures the smallest glucose concentration change detectable. The proposed system, which includes the spectroscopic hardware and a spectral preprocessing algorithm, is shown to produce a selectivity value of 0.57, with zero being nonselective and one being fully selective, and a limit of detection value of 2.2 mM. Prediction of an independent dataset is also performed using net analyte signal-based and partial least-squares multivariate calibration techniques, which produce standard error of prediction values of 1.14 and 1.45 mM, respectively.
|
|---|