Abstract

Determination of glucose and other clinically important blood constituents based on IR spectrometry and multivariate calibration techniques, such as partial least-squares (PLS) and principal components regression (PCR), has been an active research area. In our recent investigations of glucose determination in undiluted human whole blood samples, we noticed that the application of multivariate calibration based on PLS in combination with adaptive neural networks (PLS-ANN) resulted in significant improvement in glucose prediction compared with results from either the PLS or PCR technique. In the study reported here, we have applied this technique for the determination of different constituents in human blood serum. The specific objective of this study was to compare the capabilities of the PLS, PCR, and PLS-ANN techniques for the prediction of cholesterol, total proteins, glucose, and urea in human blood serum samples.