Many solid-state materials such as active pharmaceutical ingredients (APIs) contain impurities that are typically present at concentrations of <0.5% w/w. These impurities need to be identified and quantified, and high performance
liquid chromatography (HPLC) is typically the method of choice in pharmaceutical applications. Unfortunately, HPLC
is not capable of analysing for some solid-state contaminants such as polymorphs and some solvates. Raman spectroscopy on the other hand is very well suited to the identification and quantification of most impurities including polymorphs and solvates. However, in the solid-state, the application of Raman spectroscopy for routine analyte quantification below ~0.5% w/w is not routine.Raman imaging (or more correctly, in this context, high-number, point-sampling, HNPS) has been used for low
level (0.025 to 0.1% w/w) impurity detection in tablets . To extend this to achieve robust and accurate, low content
(<0.1% w/w) quantification, we have been investigating the use of the signal heterogeneity contained in Raman imaging
data. Pixel-to-pixel Raman signal variation is caused by local fluctuations in sample consistency, and this sample
heterogeneity can be correlated with, for example, the concentration of impurities in the bulk sample. To extract the
relevant composition information to make these correlations we have to combine HNPS, with standard chemometric
methods and statistical analyses to provide robust and accurate quantification in the 0.01 to 0.1% w/w concentration
The first model system we studied comprised of solid mixtures of piracetam and proline (61 samples prepared in
triplicate), which was selected because both components have nearly equal Raman scattering co-efficiencies. For each
sample, ~8400 Raman spectra were collected from a 29×29 pixel map using a PhAT Imaging workstation (Kaiser). The
quantification method involved first building several discrete piracetam concentration models using partial least-squares
(PLS) regression . These PLS models were then used to predict the local concentration of piracetam at each pixel.
The combined local concentration predictions were finally statistically analysed to generate the true sample
concentration. The piracetam contaminant concentration was quantified with a relative accuracy of ~2.4% over the 0.05−1.0% w/w concentration range  and the limit of detection was 0.03% (comparable with HPLC). For this approach to be feasible in an industrial setting, all steps in the data analysis had to be automatable including baseline and cosmic ray artefact correction , and informative variable selection. A similar HNPS approach is now also being
applied to more complex four component powders and tablets, with a goal of quantifying API concentrations in the
0.001 to 0.1% w/w range.
Acknowledgement: This work was undertaken as part of the Synthesis and Solid State Pharmaceutical Centre funded
by Science Foundation Ireland and industry partners, and Enterprise Ireland (Grant No: TC-2012-5106). We thank Kaiser Optical Systems, Inc. and Mr. H. Owen for the loan of the Raman instrumentation.