Linear chemically sensitive electron tomography using DualEELS and dictionary-based compressed sensing

Machine learning in electron microscopy for advanced nanocharacterization: current developments, available tools and future outlook

In the last few years, electron microscopy has experienced a new methodological paradigm aimed to fix the bottlenecks and overcome the challenges of its analytical workflow. Machine learning and artificial intelligence are answering this call providing powerful resources towards automation, exploration, and development. In this review, we evaluate the state-of-the-art of machine learning applied to electron microscopy (and obliquely, to materials and nano-sciences). We start from the traditional imaging techniques to reach the newest higher-dimensionality ones, also covering the recent advances in spectroscopy and tomography. Additionally, the present review provides a practical guide for microscopists, and in general for material scientists, but not necessarily advanced machine learning practitioners, to straightforwardly apply the offered set of tools to their own research. To conclude, we explore the state-of-the-art of other disciplines with a broader experience in applying artificial intelligence methods to their research (e.g., high-energy physics, astronomy, Earth sciences, and even robotics, videogames, or marketing and finances), in order to narrow down the incoming future of electron microscopy, its challenges and outlook.

Compressed sensing for electron cryotomography and high-resolution subtomogram averaging of biological specimens

Cryoelectron tomography (cryo-ET) and subtomogram averaging (STA) allow direct visualization and structural studies of biological macromolecules in their native cellular environment, in situ. Often, low signal-to-noise ratios in tomograms, low particle abundance within the cell, and low throughput in typical cryo-ET workflows severely limit the obtainable structural information. To help mitigate these limitations, here we apply a compressed sensing approach using 3D second-order total variation (CS-TV²) to tomographic reconstruction. We show that CS-TV² increases the signal-to-noise ratio in tomograms, enhancing direct visualization of macromolecules, while preserving high-resolution information up to the secondary structure level. We show that, particularly with small datasets, CS-TV² allows improvement of the resolution of STA maps. We further demonstrate that the CS-TV² algorithm is applicable to cellular specimens, leading to increased visibility of molecular detail within tomograms. This work highlights the potential of compressed sensing-based reconstruction algorithms for cryo-ET and in situ structural biology.

•

Compressed sensing (CS-TV²) for cryo-ET using 3D second-order total variation

•

CS-TV² increases signal contrast while retaining high-resolution information

•

Improved subtomogram averaging from CS-TV² reconstructions of small datasets

•

Increased contrast and detail in CS-TV² reconstructions of cellular specimens

Spectrum imaging of complex nanostructures using DualEELS: II. Absolute quantification using standards

Nanometre-sized Ti_xV_(1-x)C_yN_z precipitates in an Fe20%Mn steel matrix with a thickness range from 14 to 40 nm are analysed using DualEELS. Their thicknesses, volumes and compositions are quantified using experimental binary standards and the process used to give robust results is described. Precisions of a few percent are achieved with accuracies that are estimated to be of a similar magnitude. Sensitivities are shown to be at 0.5-1 unit cells range in the thinnest matrix region, based on the assumption that a sub-lattice is fully populated by the element. It rises to the 1-2 unit cell range for the metals and 2-3 unit cells for the non-metal in the thickest matrix region. The sensitivities for Ti and N are greater than those for V and C respectively because the O K-edge from surface oxide needs to be separated from the V L_2,3-edge, and the C K-edges from C in the matrix and amorphous C on the surface have to be separated from the C in the precipitate itself. Separation of the contributions from the bulk and the surface is demonstrated, showing that there is significant and detectable C in the matrix but no O, while there is significant O but little C in the surface oxide. Whilst applied to precipitates in steel in this work, the approach can be adapted to many multi-phase systems.

Optimization of Three-Dimensional (3D) Chemical Imaging by Soft X-Ray Spectro-Tomography Using a Compressed Sensing Algorithm

Soft X-ray spectro-tomography provides three-dimensional (3D) chemical mapping based on natural X-ray absorption properties. Since radiation damage is intrinsic to X-ray absorption, it is important to find ways to maximize signal within a given dose. For tomography, using the smallest number of tilt series images that gives a faithful reconstruction is one such method. Compressed sensing (CS) methods have relatively recently been applied to tomographic reconstruction algorithms, providing faithful 3D reconstructions with a much smaller number of projection images than when conventional reconstruction methods are used. Here, CS is applied in the context of scanning transmission X-ray microscopy tomography. Reconstructions by weighted back-projection, the simultaneous iterative reconstruction technique, and CS are compared. The effects of varying tilt angle increment and angular range for the tomographic reconstructions are examined. Optimization of the regularization parameter in the CS reconstruction is explored and discussed. The comparisons show that CS can provide improved reconstruction fidelity relative to weighted back-projection and simultaneous iterative reconstruction techniques, with increasingly pronounced advantages as the angular sampling is reduced. In particular, missing wedge artifacts are significantly reduced and there is enhanced recovery of sharp edges. Examples of using CS for low-dose scanning transmission X-ray microscopy spectroscopic tomography are presented.