Multivariate statistics applications in scanning transmission electron microscopy X-ray spectrum imaging

A modern STEM-EDS spectrum imaging system can quickly and easily collect huge quantities of data. Then the challenge to the analyst is how to turn a large, often noisy, dataset into insight to the materials science and engineering problem at hand. This review discussed the application of MVSA methods, primarily PCA and MCR, to address this question. MVSA methods are generally used to either (1) perform noise filtering on the raw data or (2) provide a reduced-rank, and more interpretable, description of the raw data. However, neither application is straightforward in the general case. PCA and PCA followed by further manipulations can provide both noise-filtering and qualitative interpretations of the data, but effects such as a matrix phase fully surrounding precipitates can lead to significant ambiguities in the quantitative application of score images or loading spectra. MCR-based techniques have the potential to address these ambiguities, but much more work is needed. MCR methods are also subject to subtle artifacts that must be carefully considered. Experimental parameters such as signal level, detector resolution, and phases present in the SI data can significantly affect both the quantitative and qualitative results of MVSA computations, and precautions such as ensuring that the rank of the sample and the pseudo-rank of the MVSA results match, are important to ensure useful MVSA applications. This is also an area of ongoing research.