Error removal by orthogonal subtraction (EROS): A customised pre-treatment for spectroscopic data

Abstract

In some applications of diffuse reflectance spectroscopy there may be substantial variability between the spectra from replicate measurements of what is nominally the same sample. A method called error reduction by orthogonal subtraction (EROS) is proposed to ameliorate the effects of this. The first step is to use principal component analysis (PCA) to identify the structure in the variability of replicate measurements. This is followed by subtraction of the modelled effects from the original spectral data matrix X by projection onto the subspace orthogonal to factors derived from the PCA. An application to the clinical diagnosis of colon lesions is presented, in which pre-treatment of spectra using the proposed method is successful in reducing the complexity and increasing both the accuracy and interpretability of the subsequent classification model. Copyright © 2008 John Wiley amp; Sons, Ltd.