Unsupervised Analysis of Big ToF-SIMS Data Sets: a Statistical Pattern Recognition Approach.

We present a new method, fast and low demanding in terms of CPU performances, which is able to extract latent chemical information from ToF-SIMS big data sets, such as those arising from chemical imaging, by working on the unbinned raw data files. The method is able to evaluate the similarity/dissimilarity of very low intensity spectra, such as those arising from a single pixel, in terms of symmetry and asymmetry relationships of the count distribution in the Fourier transform domain. The tests performed so far on model samples show that the method supplies results that, without sacrificing mass or spatial resolution, are equivalent, at least, to those achievable by an experienced ToF-SIMS user by applying PCA techniques.