AUTOMATED CHEMOMETRIC PROCEDURE FOR QUANTITATIVE AND QUALITATIVE ANALYSIS OF CONFOCAL RAMAN IMAGES

A chemometric procedure for the analysis of confocal Raman images of biological samples is presented. First, the data matrix undergoes a pre-processing consisting of: removal of cosmic ray peaks, baseline correction [1], wavenumber restriction, noise smoothing, normalization, Principal Component (PC) compression and cluster analysis. The pre-processed data is then unmixed using Band-Target Entropy Minimization (BTEM) [2], a multivariate algorithm for extracting the spectra of pure components within a mixture. It works by seeking the linear combination of PCs that satisfies the following criteria: i) non-negativity of spectral intensities; ii) minimum entropy (which implies maximum simplicity); iii) presence of a certain peak (band-target). BTEM is applied to the whole spectral matrix, as well as to cluster-averaged spectra: the clustering fuzzes some details on major components, but highlights information on other minor ones. All the obtained solutions (one for each band-target) are post-screened to discard replicates and incorrect outcomes. The final set of selected spectra is used to produce images of the relative concentration for each component by a least-squares fit. Several steps of this procedure are automated by in-house written algorithms to minimize user intervention and subjective determination of parameters. This methodology enables the non-invasive mapping of various substances on the same image. In the shown examples taken from various plant tissues, up to seven components were identified, including some with overlapping peaks or small concentration. No a priori chemical knowledge on the sample is needed, although it can be useful for the final post-screening. Because this procedure allows for finding out which compounds are present in a sample (provided that they give a Raman signal), it is also a powerful tool for qualitative analysis. Moreover, it can provide information on certain molecular bonds from relevant peaks in the extracted spectra and on the orientation of ordered structures from their anisotropic response [3] [1] C. G. Bertinetto, and T. Vuorinen, Applied spectroscopy, 68(2):155–164, 2014. [2] E. Widjaja, C. Li, W. Chew, and M. Garland, Analytical chemistry, 75(17):4499–4507, 2003. [3] L. Galvis, C. Bertinetto, U. Holopainen, T. Tamminen, and T. Vuorinen, Journal of Cereal Science. Submitted.Non peer reviewe