Chemical imaging is currently utilized as a powerful tool for the determination of the chemical composition and heterogeneity with a sample. It is a widely-used process in analytical chemistry, but one method alone cannot be used to fully characterize complex samples. The combination of imaging and multivariance statistics, whilst a promising approach for this, is still in its infancy. A large team of Researchers from Vienna and Budapest have come up with a multivariant statistical approach using various imaging datasets to generate a comprehensive analytical representation of a samples image.
Chemical imaging techniques have become a widespread tool through analytical chemistry, with recent developments surrounding elemental, vibrational and mass-spectrometric chemical imaging techniques producing images with high spatial resolutions (50–200 nm) and within a timescale of a few hours.
Improved imaging speeds, lower detection limits and increased computational power have all contributed to growth of imaging techniques within the chemical sciences, however, the inability of one technique being insufficient in providing a comprehensive structural determination of a complex molecule has been a challenging process to overcome.
The combination of bulk materials and the extrapolation of data through statistical means to provide a comprehensive analytical understanding has been well established for different techniques. However, an approach using multivariance statistics to combine the data sets of different imaging techniques is an area still in its infancy, and is mainly due to the lack of analytical methodologies available for data fusion and analysis approaches.
The approach known as chemical structural determination (CSD) uses a combination of complimentary analytical techniques to collectively solve scientific problems. Whilst this is a fundamental principle across analytical chemistry in general, it had yet to be applied to imaging techniques.
The research team have now demonstrated image-based chemical structural determination approach by applying multivariance statistics to energy dispersive X-Ray (EDX, FEI Quanta 200 scanning electron microscope with an EDAX EDX detector), Raman micro-spectroscopic (RMS, WITec alpha 300RSA+ Raman microscope with Andor iDus Deep Depletion CCD imaging camers) and time-of-flight secondary ion mass spectrometry (TOF-SIMS, ION-TOF GmbH) imaging datasets into a combined multisensor hyperspectral imaging (MSHSI) datacube.
Within these datasets, the Researchers employed multiple statistical analysis approaches, including spectral descriptor (SPDC) based principal component analyzes (PCA), k-means cluster analyzes, hierarchical cluster analyses (HCA) and vertex component analyses (VCA).
The multivariate approach using a fused HSI dataset was found to be a unique method for chemical structure determination, as it allowed the Researchers to link analytical information across various individual techniques using a combined statistical approach.
The Researchers imaged different metal-based complexes on substrates. They deduced the advantages of the image-based CSD technique whilst analyzing copper sulphate particles which were deposited on a purified aluminum substrate.
The combination of imaging techniques outlined above identified the composition of the particles by integrating the CuS signals from each method within the same sub-cluster of a PCA-HCA dendrogram and through k-means clustering analyzes.
The Researchers have also shown the substantial benefits of this method through using practical samples. The Researchers tested samples from a wide range of environments, including from the life sciences, materials science and geoscience, and more specifically, tumour cells, technical ceramics and an environmental aerosol.
Firstly, the Researchers used a cell sample with a spatially allocated bromine-containing drug. This sample could only utilize EDX and SIMS methods, but the process allowed for SIMS fragment of lipids from the –CH2 band of the RMS (the method not used) to be extrapolated from the dataset.
Another example, is that of the 3Y-ZrO2 technical ceramic sample. This sample could be identified using all three methods, but information surrounding its mullite (3Al2O3.SiO2) matrix could only be identified through EDX and SIMS methods. ZrSiO4 was found to only be extracted using the RMS dataset by VCA.
Because of the many constituents, the aerosol sample required all three interpretations to be utilized in a combined approach. Using this, EDX alone was required for the silicate particles, EDX and RMS for the black and organic carbon particles, EDX and RMS for the ash particles, RMS and SIMS for the sodium hydrogen carbonate particles, and all three were required to determine the sodium nitrate particles.
The application and combination of the complimentary techniques, with comparable spatial resolutions, has provided a method with a low risk towards overlooking important information. The cross-correlation of different methods has also helped to prevent the over-interpretation of insufficient data.
This approach using the MSHSI datacube has provided a focused method of image-based chemical structure determination. By applying the 3D and/or temporal-correlated imaging techniques, multivariance statistics and dimensional/dynamical HSI datasets, the Researchers have deepened the understanding of complex materials and their processes, alongside producing a new chemical structure determination approach.
“Image-Based Chemical Structure Determination”- Ofner J., et al, Scientific Reports, 2017, DOI:10.1038/s41598-017-07041-x