Comparison Of Multivariate Calibration With PCR And PLS For Glucose Using Infrared Spectra Of Human Whole Blood

Publisher Website
Google Scholar
Abstract
The calibration of glucose in whole EDTA blood was carried out using FT-IR spectra obtained with a micro-CIRCLE cell in the range of 1500 - 750 cm-1. The reference concentrations had been measured by the enzymatic glucose dehydrogenase method. The P-matrix approach was chosen applying the Partial-least-squares (PLS) and Principal component regression (PCR) algorithms. The optimum model for predictipns was obtained by PLS achieving an average prediction error of 19.8 mg/dL based on PRESS1 from crossvalidation. The calibration results for both algorithms are discussed. PLS is favoured on the grounds of numerical and statistical advantages.

This publication has 0 references indexed in Scilit: