A robust and accurate analytical methodology for low-content (<0.1%) quantification in the solid-state using
Raman spectroscopy, subsampling, and chemometrics was demonstrated using a piracetam−proline model. The method
involved a 5-step process: collection of a relatively large number of spectra (8410) from each sample by Raman mapping,
meticulous data pretreatment to remove spectral artifacts, use of a 0−100% concentration range partial least-squares (PLS) regression model to estimate concentration at each pixel, use
of a more accurate, reduced concentration range PLS model to calculate analyte concentration at each pixel, and finally statistical analysis of all 8000+ concentration predictions to produce an
accurate overall sample concentration. The relative prediction accuracy was ∼2.4% for a 0.05−1.0% concentration range, and the limit of detection was comparable to high performance liquid chromatography (0.03% versus 0.041%). For data pretreatment, we developed a unique cosmic ray removal method and used an automated baseline correction method, neither of which required subjective user intervention and thus were fully automatable. The method is applicable to systems which cannot be easily analyzed chromatographically, such as hydrate, polymorph, or solvate contamination.