Image-Based Chemical Structure Determination

Abstract

Chemical imaging is a powerful tool for understanding the chemical composition and nature of heterogeneous samples. Recent developments in elemental, vibrational, and mass-spectrometric chemical imaging with high spatial resolution (50-200 nm) and reasonable timescale (a few hours) are capable of providing complementary chemical information about various samples. However, a single technique is insufficient to provide a comprehensive understanding of chemically complex materials. For bulk samples, the combination of different analytical methods and the application of statistical methods for extracting correlated information across different techniques is a well-established and powerful concept. However, combined multivariate analytics of chemical images obtained via different imaging techniques is still in its infancy, hampered by a lack of analytical methodologies for data fusion and analysis. This study demonstrates the application of multivariate statistics to chemical images taken from the same sample via various methods to assist in chemical structure determination.

Figure 1

Basic concept of multisensor hyperspectral imaging with subsequent multivariate statistics to generate an analytical representation of a sample for image-based chemical structure determination.

Figure 2

Superposition of the high-resolution SEM image with the CuS particle cluster of the HCA of the loadings of the PCA, dendrogram of the PCA-HCA indicating the background (black) and CuS particle cluster (green), and the extracted cluster spectra from the k-means clustering.

Figure 3

Superposition of a light-microscopic image of two tumour cells and results from the HCA of the PCA loadings and extracted cluster spectra from k-means clustering of the MSHSI dataset.

Figure 4

Superposition of the SEM image of the ceramic mullite/ZrO₂ composite with the image of the HCA of the PCA loadings (upper part) and with the extracted endmembers of the VCA (lower part), which represents the diamond Raman spectrum (red) and the ZrSiO₄ Raman spectrum (orange).

Figure 5

Superposition of the SEM image of the aerosol sample with the virtual image of the HCA of the MSHSI dataset and assignment of chemical species to the sub-clusters, based on extracted cluster spectra.

Figure 6

Examples of extracted sub-cluster spectra from the HCA of the environmental aerosol sample and assignment of spectral features.