Model-Fitting Weighted Least Squares as an Alternative to Principal Component Analysis for Analyzing...

Area of Science:

• Materials Science
• Analytical Chemistry
• Spectroscopy

Background:

• Energy-dispersive X-ray spectroscopy (EDS) is a key technique for material characterization.
• Multivariate statistical methods like principal component analysis (PCA) are commonly used for EDS data analysis.
• PCA can introduce artifacts, especially at low signal-to-noise ratios, and lacks fair comparisons with alternative methods.

Purpose of the Study:

• To directly compare PCA with a weighted least squares (WLS) based strategy for EDS spectrum image analysis.
• To evaluate the performance of model fitting followed by WLS (mfWLS) against PCA.
• To demonstrate the advantages of mfWLS in handling compositional gradients, inclusions, and artifacts.

Main Methods:

• Direct comparison of PCA with a weighted least squares (WLS) approach.
• Application of model fitting of the sum spectrum followed by WLS (mfWLS).
• Evaluation across four representative case studies of EDS spectrum images.

Main Results:

• mfWLS consistently outperforms PCA in identifying and describing compositional gradients and inclusions.
• mfWLS proves superior as a preprocessing step for clustering compared to PCA.
• Common artifacts and biases associated with PCA are avoided using the mfWLS approach.

Conclusions:

• mfWLS offers a superior alternative to PCA for analyzing EDS spectrum images.
• The mfWLS method provides simple and accurate signal modeling for EDS data.
• This approach enhances the reliability and accuracy of material characterization using electron microscopy.