Spectral deconvolution and concentration mapping in complex biochemical stains

Abstract

Spectral Imaging (SI) combines spectroscopy and imaging, enabling the acquisition of a stack of images at narrow spectral bands comprising the so-called spectral cube. A complete spectrum can be calculated for every image pixel from the multidimensional spatio-spectral space of the cube. This study aims at identifying the concentration of solvents in mixtures of multiple biochemical stains with overlapping spectral signatures. More specifically, a series of experiments has been undertaken via experimental design arrangements (full factorial, face-centered & half factorial) employing both spectrum acquisition by spectrophotometer and spectral imaging acquisition. Furthermore, an extensive number of algorithmic methods, based on Beer Lambert's law generalization, including Classical Least Squares (CLS), Inverse Least Squares (ILS) with forward or backward selection, Principal Components Regression (PCR) and Partial Least Squares (PLS) has been implemented and applied to both simulated and experimental data. It was found that PLS can predict the concentrations in mixtures of two and three solvents with high accuracy on the datasets of spectral images. The combination of SI with concentration prediction algorithms can provide a valuable tool for quantitative assessment of the uptake of biological stains in histochemistry applications.