Chemometric determination of blood parameters using visible-near-infrared spectra

Visible and near-infrared (NIR) integrating sphere spectroscopy and chemometric multivariate linear regression were applied to determine hematocrit (HCT) and oxygen saturation (SatO(2)) of circulating human blood. Diffuse transmission, total transmission, and diffuse reflectance were measured and the partial least squares method (PLS) was used for calibration considering different wavelength ranges and selected optical measurement parameters. HCT and SatO(2) were changed independently. Each parameter was adjusted to different levels and four designs with blood from different donors were carried out for the calibration with PLS. The calibration included the changes in hemolysis as well as inter-individual differences in cell dimensions and hemoglobin content. At a sample thickness of 0.1 min the HCT and SatO(2) were predicted with a root mean square error (PRMSE) of 1.4% and 2.5%, respectively, using transmission and reflectance spectra and the full Vis-NIR range. Using only diffuse NIR reflectance spectroscopy and a sample thickness of 1 mm, HCT and SatO(2) could be predicted with a PRMSE of 1.9% and 2.8%, respectively. Prediction of hemolysis was also possible for one blood sample with a PRMSE of 0.8% and keeping HCT and SatO(2) stable with a PRMSE of 0.03%.