Multimodal X-ray nano-spectromicroscopy analysis of chemically heterogeneous systems

Abstract

Understanding the nanoscale chemical speciation of heterogeneous systems in their native environment is critical for several disciplines such as life and environmental sciences, biogeochemistry, and materials science. Synchrotron-based x-ray spectromicroscopy tools are widely used to understand the chemistry and morphology of complex material systems owing to their high penetration depth and sensitivity. The multi-dimensional (4D+) structure of spectromicroscopy data poses visualization and data reduction challenges. This paper reports the strategies for the visualization and analysis of spectromicroscopy data. We created a new graphical user interface and data analysis platform named XMIDAS to visualize spectromicroscopy data from both image and spectrum representations. The interactive data analysis toolkit combined conventional analysis methods with well-established machine learning classification algorithms (e.g. non-negative matrix factorization) for data reduction. The data visualization and analysis methodologies were then defined and optimized using a model particle aggregate with known chemical composition. Nanoprobe-based x-ray fluorescence (nano-XRF) and x-ray absorption near edge structure (nano-XANES) spectromicroscopy techniques were used to probe elemental and chemical state information of the aggregate sample. We illustrated the complete chemical speciation methodology of the model particle by using XMIDAS. Next, we demonstrated the application of this approach in detecting and characterizing nanoparticles associated with alveolar macrophages. Our multimodal approach combining nano-XRF, nano-XANES, and differential phase-contrast (DPC) imaging efficiently visualizes the chemistry of localized nanostructure with the morphology. We believe that the optimized data reduction strategies and tool development will facilitate the analysis of complex biological and environmental samples using x-ray spectromicroscopy techniques.

Manifold Projection Image Segmentation for Nano-XANES Imaging

“…1, where background spectra are filtered out such that only the sample region is examined. This sampleand thus dataset is the same as the one found in Pattammattel, et al 17 Each "pixel" (150 nm wide) represents a processed XANES spectrum. Each pixel is 150 nm.…”

“…12 We measured Fe K-edge XANES using nano-XANES imaging, where our data consist of a 3D image with 155 x 160 spatial pixels and just over 70 photon energies between 7.08 to 7.20 keV, or approximately 25,000 spectra. Further processing of the stack was performed via the XMIDAS program 17 . The energy stack was first spatially aligned using the image registration tool in XMIDAS that uses the PyStackReg package 18 .…”

Accelerating nano-XANES imaging via feature selection

“…The sample and experimental data is the same as it appeared in A. Pattammattel, et al 25 and S. Tetef, et al 26 See those works for the experimental details. Briey, the sample was composed of stainless steel (SS), lithium iron phosphate (LFP), pyrite (Pyr), and hematite (Hem) nanoparticles.…”

Accurate, Uncertainty-Aware Classification of Molecular Chemical Motifs from Multimodal X-ray Absorption Spectroscopy

“…Inspired by the work in the literature, we show how uncertainty-aware neural network classifiers can be used to predict 41 different chemical motifs at very high accuracy using multiple simulated XANES signals as input. Multimodal data should, in principle, be of more value than the sum of its parts to ML models and are already being employed to great effect in many domains, − including the study of XANES spectra. − The ML classifiers we develop can have important applications in environmental chemistry, paleontology, organic geochemistry, and space science . We also evaluate the effectiveness of the uncertainty quantification (UQ) algorithms and demonstrate a “multimodal advantage”: the accuracy of the classifiers is significantly higher when using multiple XANES signals from different absorbing elements when compared to that when using only single modalities independently.…”