Interpretation of TOF-SIMS images: multivariate and univariate approaches to image de-noising, image segmentation and compound identification

Interpretation of SIMS images combines all of the challenges of SIMS spectral interpretation with the challenges of image processing. This goal is further complicated by the extremely low signal to noise ratio typical in images, difficulties in isolating pure component spectra and interference from topographic and matrix effects. Identifying compounds and distinguishing between chemical and topographical features requires simultaneous analysis of multiple ion images. As a result multivariate statistical techniques, including principal components analysis, factor analysis, neural networks and mixture models, can be powerful tools for exploring the spectral images. Segmentation of SIMS images, that is, identification of chemically distinct regions within an image, can often be complicated by the very low signal to noise ratios which degrade image contrast and make identification of region boundaries problematic. Examples where these statistical techniques have been used to successfully separate chemical and topographical features and to isolate pure component spectra from the images are presented. PCA and latent profile analysis have enormous utility for exploring and quantifying TOF-SIMS images. Wider application of these statistical image analysis techniques is needed to fully exploit the potential of TOF-SIMS imaging.