Multivariate Statistical Analysis of Spectrum Images

Digital Micrograph Plugin for Multivariate Statistical Analysis

This plugin for Digital Micrograph allows to perform Principal Component Analysis (PCA) on hyperspectral data, such as EELS, EFTEM or EDX. Noise reduction can be easily performed by selecting only components explaining most of the variance, thus improving signal-to-noise ratio. Factor rotations can be applied to selected components in order to get a more realistic representation of the data. Segmentation of the data is possible on the basis of the matrix factorization by mean of scatter diagrams.

 

 

 Installer for DM 2.x 64 bit: MSA-DM2-64bit-4.3.0 .msi

 

Please cite the following article if you have used this plugin for your work:

G. Lucas et al., Micron 52-53 (2013) p.49-56

 

Release Notes:

V4.3 (1 october 2014) – Independent Component Analysis (ICA) added. Minor bug fixes.

V4.2 (7 october 2013) – First official release.

Example on the segmentation of 3D hyperspectral data

Acquisition of three-dimensional (3D) spectral data is nowadays common using many different micronalytical techniques. In order to proceed to the 3D reconstruction, data processing is necessary not only to deal with noisy acquisitions but also to segment the data in term of chemical composition. Multivariate statistical analysis (MSA) methodscan  applied to reach this goal, allowing fast and reliable results. Using scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDX) coupled with a focused ion beam (FIB), a stack of spectrum images have been acquired on a sample produced by laser welding of a nickel-titanium wire and a stainless steel wire presenting a complex microstructure. These data have been analyzed using principal component analysis (PCA) and factor rotations. PCA allows to significantly improve the overall quality of the data, but produces abstract components. Rotated components can be used without prior knowledge of the sample to help the interpretation of the data, obtaining quickly qualitative mappings representative of elements or compounds found in the material.

Fe mapping resulting from factor rotations in spectral domain.

Ti mapping resulting from factor rotations in spectral domain.


Ni mapping resulting from factor rotations in spectral domain.

Such abundance maps can then be used to plot scatter diagrams and interactively identify the different domains in presence by defining clusters of voxels having similar compositions. Identified voxels are advantageously overlaid on secondary electron (SE) images  with higher resolution in order  to refine the segmentation.

The 3D reconstruction can then be performed using available commercial softwares on the basis of the provided segmentation.

 3D reconstruction resulting from the segmentation using rotated PCA.

 

MSA and 3D reconstruction: Guillaume Lucas (EPFL-LSME)

FIB/EDX acquisition: Pierre Burdet (EPFL-CIME)

Sample preparation. Jonas Vannod (EPFL-LSMX)