Discrete Tomography

February 1, 2009 – An international team with Ute Kaiser from Ulm University has shown that discrete tomography improves electron tomograms of nanomaterials. The DART algorithm, which assumes a few materials with one gray level each, reduced missing-wedge artifacts, delivered segmented volumes of erbium silicide nanocrystals in silicon carbide and of a catalyst in a carbon nanotube, and rebuilt a gold nanoparticle from only 15 images.

A tomogram should show a gold particle as gold and empty space as empty space, nothing in between. In practice, electron tomography delivers a haze of gray values: streaks where tilt angles are missing, blurred edges, and particles that look brighter or dimmer depending on their size. Before anything can be measured, someone has to decide by hand where a particle ends. An international team from the University of Antwerp, the Fraunhofer Institute in Bremen, the University of Cambridge, Ulm University and the National University of Córdoba in Argentina, with Ute Kaiser from Ulm among the authors, turned that weakness around. Their discrete algebraic reconstruction technique (DART) assumes from the start that a sample consists of only a few materials, each with its own gray level, and uses this prior knowledge to rebuild the volume. Tested on three very different samples, among them erbium silicide nanocrystals in silicon carbide from Ulm, DART gave sharper reconstructions with far fewer missing-wedge artifacts, already segmented into their materials, and a gold nanoparticle in full 3D from only 15 images.

Why tomograms need cleaning up
Electron tomography reconstructs the inside of a nanomaterial from a series of projection images taken while the sample is tilted in the beam[1]–[3]. Images can be recorded in bright-field TEM, in high-angle annular dark-field (HAADF) STEM, where heavier atoms scatter more and appear brighter, or in annular dark-field TEM[4]. Most reconstructions then use weighted back-projection, long the workhorse of biological tomography[5], or iterative algebraic methods such as the simultaneous iterative reconstruction technique (SIRT)[6],[7], which have become popular as computing power grew.

All these methods need many images, preferably more than 100 taken in steps of 1 to 2° over at least ±60°[8],[9]. Even then the sample holder blocks the highest tilts, and the missing angles leave a “missing wedge” of information that stretches and smears every feature along the beam direction. Recording a second series around a perpendicular axis shrinks the wedge to a “missing pyramid”[10],[11], at the price of a much more complex experiment. And after reconstruction, the volume still has to be segmented, each voxel assigned to a material. Because of the artifacts, that step is often done by hand, takes time and is never free of bias.

A few gray levels instead of many
Discrete tomography starts from the opposite end. If a sample consists of only a few materials, an ideal reconstruction contains only a few gray levels, one per material[12]. One variant of the idea works at atomic resolution, assuming in addition that the atoms sit on a known lattice; in simulations it rebuilt a gold nanocrystal from very few projections along zone axes[13]. The variant used here makes no assumption about a lattice and works at the lower resolution where atoms are not resolved, only materials. The mathematics had been studied since the 1990s, but no algorithm was fast and robust enough for real data until Joost Batenburg and Jan Sijbers proposed DART, which reconstructs images of 1024 × 1024 pixels in a few minutes on a standard PC[14]. The team set out to show that it works as a general tool for materials science.

DART alternates between ordinary SIRT steps and steps that force the gray levels into their discrete values; how one cycle runs is explained in the box below, with an example in Fig. 1.

▸ How DART works

The threshold matters little for the final result. What DART does need is the set of gray levels, one per material, before it starts. For all three samples the team first computed 25 SIRT iterations, used them as the starting point and ran DART for 50 iterations, with the same SIRT code in both, so that the comparison is fair.

Erbium silicide crystals in silicon carbide
The first sample came from Ulm research on nanocrystals buried in silicon carbide. Erbium ions were implanted into SiC at a high dose of 1016 cm−2 and 700 °C, then the crystal was annealed rapidly at 1600 °C[17]. This forms ErSi2 nanoparticles from a few atoms up to about 25 nm across in a narrow band under the surface, along with small voids near the surface. In two dimensions, Ute Kaiser and colleagues had studied their structure in detail and described them as hill-like crystals[18]–[20] (Fig. 2). Tomography was to reveal their full 3D shape[21].

The tilt series was recorded in HAADF-STEM mode on an FEI Tecnai F20 with a probe of about 200 pm: 177 images over ±77°, aligned by cross-correlation and marker tracking. Because the tilt range is wide, SIRT already shows the shape of most particles, but with streaks along the beam. In the DART reconstruction, computed with just two gray levels, one for the particles and one for the SiC, the particles appear as clean, compact objects (Fig. 3), even those at the far left and right of the field of view that left the image in many projections and would normally be discarded.

The real test is segmentation. Larger particles come out brighter in SIRT than smaller ones, so no single threshold captures them all: a low one swells the large particles, a high one loses the small ones. DART’s result already contains nearly all particles, except possibly some extremely small ones and some non-stoichiometric features with less erbium. For quantitative work on particle size and shape, the authors conclude, DART can significantly improve reconstructions of binary samples.

A catalyst inside a carbon nanotube
The second sample was more complex: the catalyst particle at the tip of a bamboo-like carbon nanotube grown by chemical vapor deposition from an alkali-modified Cu/MgO catalyst[22]. The particle contains both Cu and Cu2O[23], has cavities, and fills some of the hollow bamboo compartments of the tube. A series of 66 HAADF-STEM images was recorded on an FEI Titan 80-300 with a Fischione ultra-high-tilt holder, over the smaller range of −54° to +74° in 2° steps (Fig. 4a).

With three gray levels, for background, Cu and Cu2O, DART returned a volume in which every voxel belongs to one of the three. Orthogonal slices show the difference: SIRT smears the particle vertically in one slice and horizontally in another, and its boundaries spread over a wide range of gray values, so it is hard to tell which voxel belongs to which phase; DART shows sharp boundaries between the copper, the oxide and the cavities, without the typical missing-wedge elongation (Fig. 4b–g). The volume is ready for automatic analysis of the phase ratio or the size of the cavities. DART, the team concluded, also works with several compositions, as long as they are spatially separated.

A gold particle from 15 images
The third case pushed the method furthest. Anisotropic gold nanoparticles were grown from small seeds in two growth steps, and one pentagonal particle on an amorphous carbon film was imaged in HAADF-STEM on a Tecnai F20, only 15 times, at angles between −48° and +74° (Fig. 5a). The weak contrast of the carbon support was subtracted first. Continuous reconstruction fails here: SIRT shows a blurred particle with streaks. DART, with gold and background as the only two levels, gave a sharp particle without missing-wedge artifacts (Fig. 5b, c). In a 3D rendering, the SIRT particle has no clear shape, while the DART particle shows the regular pentagonal form with all its facets clearly visible (Fig. 6).

This case also revealed the method’s weak point. Finding the gold gray level was trial and error: set too high, the particle got holes and an irregular boundary, so the team lowered it until the surface was smooth and the interior closed. “We admit that this manual procedure is unsatisfactory,” the authors write, and they announced work on estimating the gray levels within the reconstruction itself.

How close DART comes to the truth cannot be checked on a real sample, since no ground truth exists. The team therefore built a test object for each case and compared both methods with it, as described in the box “Validation: DART vs. SIRT” below, which also holds the test reconstructions (Fig. 7).

▸ Validation: DART vs. SIRT

Missing information along the beam is not a problem of tilt-series tomography alone, as a study from Cornell University published in the same weeks showed for aberration-corrected STEM; the box “Comparison: missing wedge and missing cone” below sets the two approaches side by side.

▸ Comparison: missing wedge and missing cone

Microscopes, samples and software are listed in the box below.

▸ Instruments and samples

Challenges named by the authors
DART needs to know the gray levels before it starts. For the ErSi2 crystals and the catalyst they could be read from the SIRT volume, but where SIRT is very poor, as for the gold particle, there is no clear procedure, and the manual choice took trial and error. Features that do not fit the assumed levels, such as very small particles or regions with a different erbium content, may be lost. Compositions must be separated in space; the method suits samples made of a few homogeneous materials, not gradual changes in composition.

Outlook
The three datasets came from different microscopes, different operators and very different materials, and none was recorded with discrete tomography in mind. “In fact, the decision to attempt a reconstruction using discrete tomography was made after acquisition in all three cases,” the authors write. DART reduced missing-wedge artifacts and noise without any change to the experiment and delivered volumes that are already segmented, ready for measuring volumes, surface areas or angles between facets, which for gold particles, whose optical properties depend on small changes of shape, is exactly what is needed. “We believe that DART is a very powerful approach to reconstructing homogeneous nanoparticles in general,” the authors conclude. Fewer images also mean less dose, which matters for beam-sensitive materials such as those studied in Ulm.

Resource:
Batenburg, K. J., Bals, S., Sijbers, J., Kübel, C., Midgley, P. A., Hernandez, J. C., Kaiser, U., Encina, E. R., Coronado, E. A., & Van Tendeloo, G. (2009).
3D imaging of nanomaterials by discrete tomography.
Ultramicroscopy 109(6), 730–740.
https://doi.org/10.1016/j.ultramic.2009.01.009

Update – June 7, 2011

The news article of February 1, 2009 ends above. Later work that builds on it is listed below.

Part of the materials characterization toolbox
In a review of characterization methods for materials science, Ian M. Robertson from the University of Illinois, Christopher A. Schuh from MIT and some twenty co-authors listed the DART study among the tomography work that reveals the positions, orientations and morphologies of nanoparticles, and among the examples of HAADF-STEM tomography of nanostructures[26]. In the same year, Lothar Houben and Maya Bar Sadan from Jülich published a marker-free refinement of the image alignment for high-resolution tomography that works with any reconstruction algorithm, DART included[27].

Update – March 21, 2012

Toward atomic resolution in 3D
Discrete tomography with a lattice, the atomic variant mentioned in the paper, had meanwhile rebuilt a silver nanoparticle a few nanometers across atom by atom from HAADF-STEM images along a few zone axes[28]. Mary C. Scott, Jianwei Miao and colleagues from the University of California, Los Angeles, and Lawrence Berkeley National Laboratory went a different way: with a new alignment and equal-slope tomography they reconstructed a gold nanoparticle of about 10 nm from 69 HAADF-STEM images over ±72.6° at 240 pm resolution, without assuming a lattice[29]. They noted that discrete tomography needs prior knowledge of the lattice and atoms that sit rigidly on it.

Update – June 13, 2012

Depth from a single view
Dirk Van Dyck from Antwerp, Joerg Jinschek from FEI and Fu-Rong Chen from Taiwan proposed “Big Bang” tomography, which finds the height of each atom from the phase of a reconstructed exit wave seen from a single direction[30]. As a test they used single- and double-layer graphene imaged with an aberration-corrected microscope at 80 kV and found a layer separation of 328 pm, close to the ideal 335 pm. In a review of electron tomography from the nanoscale to atoms, Zineb Saghi and Paul Midgley from Cambridge described DART as a method that retrieves artifact-free volumes from very few projections or a limited tilt range[31].

Update – August 17, 2012

Fifteen images are enough
Rowan Leary, Paul Midgley and John Meurig Thomas from Cambridge reviewed electron tomography in materials chemistry and reprinted the gold particle of this study, SIRT against DART[32]. “With this prior knowledge, a high-fidelity reconstruction from only 15 images is possible, alleviating in particular in this example the problem of the missing wedge,” they write. Days later, Gustaaf Van Tendeloo and colleagues from Antwerp cited the work in a review of advanced electron microscopy for discrete tomography as implemented in DART, which significantly reduces missing-wedge artifacts and carries out the segmentation during the reconstruction[33].

Update – March 27, 2013

Dislocations and elements atom by atom
Chien-Chun Chen, Jianwei Miao and colleagues from UCLA imaged dislocations inside a multiply twinned platinum nanoparticle in 3D at atomic resolution, with equal-slope tomography and 3D Fourier filtering of an HAADF-STEM tilt series over ±72°[34]. Two days later Sara Bals, Sandra Van Aert and Gustaaf Van Tendeloo from Antwerp reviewed high-resolution electron tomography and named DART, citing this study, as an important algorithm that exploits regions of homogeneous density[35]. Later that year Bart Goris and the Antwerp team mapped the positions and the element of each atom in core–shell Au@Ag nanorods from four HAADF-STEM images along different zone axes, combining statistical parameter estimation with compressive sensing[36].

Update – August 11, 2014

Noble-metal particles and a single image
Reviewing 3D characterization of noble-metal nanoparticles, Sara Bals and colleagues cited the study for DART, which alternates SIRT steps with discretization steps and so reduces missing-wedge artifacts[37]. Six weeks later, Chun-Lin Jia, Knut Urban, Andreas Thust and colleagues from Xi’an and Jülich determined the 3D shape of a thin MgO crystal, including its surface steps and adsorbed atoms, from a single aberration-corrected HRTEM image taken at 300 kV, an approach they recommend for radiation-sensitive crystals[38].

Update – June 29, 2015

New ways to need fewer images
Two further algorithms aimed at limited data. Eva Bladt, Daniël Pelt, Sara Bals and Joost Batenburg trained a neural network to reconstruct gold nanoparticles from five times fewer projections, naming DART as the classic example of reconstruction with prior knowledge[39]. Singanallur Venkatakrishnan, Charles Bouman and colleagues from Purdue University and the Air Force Research Laboratory introduced a model-based iterative reconstruction for bright-field tomography that treats Bragg-scattered measurements as anomalies, so that crystalline samples can also be reconstructed from bright-field data[40].

Update – July 7, 2016

An excellent start
In MRS Bulletin, Sara Bals and colleagues from Antwerp reviewed how electron tomography reached atomic resolution. On the discrete reconstructions they write: “The assumptions for the discrete reconstruction provide an excellent start to investigate different nanomaterials,” citing this study[41].

Update – September 23, 2016

Atomic electron tomography
In Science, Jianwei Miao from UCLA, Peter Ercius from Lawrence Berkeley National Laboratory and Simon Billinge from Columbia University reviewed atomic electron tomography, which determines the 3D positions of atoms in crystal defects and potentially disordered systems without averaging over identical copies[42]. Discrete tomography, they note, had imaged a silver nanoparticle in an aluminum matrix from only two HAADF-STEM images by fitting the atoms rigidly onto a crystal lattice[28]; general methods without that constraint have since revealed grain boundaries, stacking faults and the cores of dislocations at atomic resolution. As next frontiers they name more complex samples with several elements and disorder, light elements, single atoms at surfaces and interfaces, and faster acquisition.

Update – October 1, 2016

Fewer artifacts in practice
In a practical tomography guide in Micron, Misa Hayashida and Marek Malac from the National Institute for Nanotechnology in Edmonton recommended discrete tomography as one of the advanced algorithms that minimize missing-wedge artifacts and listed DART for samples of a few discrete intensities[43]. Lionel Cervera Gontard, Jesús Cintas and Rafal Dunin-Borkowski from Sevilla and Jülich cited the study among the methods that discretize intensities in a tilt series before reconstruction, and the Ulm tomography of carbon soot among the algorithms with fewer missing-wedge artifacts[44],[45].

Update – July 8, 2024

Depth from ptychography
Colum O’Leary, Jianwei Miao and colleagues from UCLA, ShanghaiTech, Berkeley and Lawrence Berkeley National Laboratory located the buried interface between two twisted hexagonal boron nitride flakes from a single view by multislice ptychography, with 57 pm lateral and 2.5 nm depth resolution, about 2.7 times better in depth than annular dark-field STEM[46].

Download this news article as PDF – for reading on paper, taking notes or filing.

References
  1. Midgley, P. A., Weyland, M., Thomas, J. M., Johnson, B. F. G. (2001). Z-Contrast Tomography: A Technique in Three-Dimensional Nanostructural Analysis Based on Rutherford Scattering. Chemical Communications, , 907–908. https://doi.org/10.1039/b101819c

  2. Weyland, M., Midgley, P. A., Thomas, J. M. (2001). Electron Tomography of Nanoparticle Catalysts on Porous Supports:  A New Technique Based on Rutherford Scattering. The Journal of Physical Chemistry B, 105, 7882–7886. https://doi.org/10.1021/jp011566s

  3. Arslan, I., Yates, T. J. V., Browning, N. D., Midgley, P. A. (2005). Embedded Nanostructures Revealed in Three Dimensions. Science, 309, 2195–2198. https://doi.org/10.1126/science.1116745

  4. Bals, S., Van Tendeloo, G., Kisielowski, C. (2006). A New Approach for Electron Tomography: Annular Dark-Field Transmission Electron Microscopy. Advanced Materials, 18, 892–895. https://doi.org/10.1002/adma.200502201

  5. Baumeister, W. (2002). Electron Tomography: Towards Visualizing the Molecular Organization of the Cytoplasm. Current Opinion in Structural Biology, 12, 679–684. https://doi.org/10.1016/S0959-440X(02)00378-0

  6. Crowther, R. A., DeRosier, D. J., Klug, A. (1970). The Reconstruction of a Three-Dimensional Structure From Projections and Its Application to Electron Microscopy. Proceedings of the Royal Society of London. A. Mathematical and Physical Sciences, 317, 319–340. https://doi.org/10.1098/rspa.1970.0119

  7. Gordon, R., Bender, R., Herman, G. T. (1970). Algebraic Reconstruction Techniques (ART) for Three-Dimensional Electron Microscopy and X-Ray Photography. Journal of Theoretical Biology, 29, 471–481. https://doi.org/10.1016/0022-5193(70)90109-8

  8. Gilbert, P. (1972). Iterative Methods for the Three-Dimensional Reconstruction of an Object From Projections. Journal of Theoretical Biology, 36, 105–117. https://doi.org/10.1016/0022-5193(72)90180-4

  9. Tong, J., Arslan, I., Midgley, P. (2006). A Novel Dual-Axis Iterative Algorithm for Electron Tomography. Journal of Structural Biology, 153, 55–63. https://doi.org/10.1016/j.jsb.2005.10.005

  10. Penczek, P., Marko, M., Buttle, K., Frank, J. (1995). Double-Tilt Electron Tomography. Ultramicroscopy, 60, 393–410. https://doi.org/10.1016/0304-3991(95)00078-X

  11. Arslan, I., Tong, J. R., Midgley, P. A. (2006). Reducing the Missing Wedge: High-Resolution Dual Axis Tomography of Inorganic Materials. Ultramicroscopy, 106, 994–1000. https://doi.org/10.1016/j.ultramic.2006.05.010

  12. Herman, G. T., Kuba, A. (Eds.) (1999). Discrete Tomography: Foundations, Algorithms and Applications. Birkhäuser, Boston.

  13. Jinschek, J., Batenburg, K., Calderon, H., Kilaas, R., Radmilovic, V., Kisielowski, C. (2008). 3-D Reconstruction of the Atomic Positions in a Simulated Gold Nanocrystal Based on Discrete Tomography: Prospects of Atomic Resolution Electron Tomography. Ultramicroscopy, 108, 589–604. https://doi.org/10.1016/j.ultramic.2007.10.002

  14. Batenburg, K. J., Sijbers, J. (2007). DART: A Fast Heuristic Algebraic Reconstruction Algorithm for Discrete Tomography. Proc. IEEE Int. Conf. Image Process. (ICIP) 2007, 4, 133–136. https://doi.org/10.1109/ICIP.2007.4379972

  15. Kak, A. C., Slaney, M. (2001). Principles of Computerized Tomographic Imaging. SIAM, Philadelphia.

  16. van der Sluis, A., van der Vorst, H. A. (1990). SIRT- and CG-Type Methods for the Iterative Solution of Sparse Linear Least-Squares Problems. Linear Algebra and its Applications, 130, 257–303. https://doi.org/10.1016/0024-3795(90)90215-X

  17. Bechstedt, F., Fissel, A., Furthmüller, J., Kaiser, U., Weissker, H., Wesch, W. (2003). Quantum Structures in SiC. Applied Surface Science, 212-213, 820–825. https://doi.org/10.1016/S0169-4332(03)00068-0

  18. Kaiser, U., Muller, D., Chuvilin, A., Pasold, G., Witthuhn, W. (2004). The Formation of Clusters and Nanocrystals in Er-Implanted Hexagonal Silicon Carbide. Microscopy and Microanalysis, 10, 301–310. https://doi.org/10.1017/S1431927604040449

  19. Kaiser, U., Biskupek, J., Muller, D., Gärtner, K., Schubert, C. (2002). Properties of GeSi Nanocrystals Embedded in Hexagonal SiC. Crystal Research and Technology, 37, 391–406. https://doi.org/10.1002/1521-4079(200204)37:4<391::AID-CRAT391>3.0.CO;2-Y

  20. Kaiser, U., Muller, D., Grazul, J., Chuvilin, A., Kawasaki, M. (2002). Direct Observation of Defect-Mediated Cluster Nucleation. Nature Materials, 1, 102–105. https://doi.org/10.1038/nmat729

  21. Kübel, C., Kaiser, U. (2006). Electron Tomographic Characterization of ErSi2 and GexSi1−x Nanoparticles Prepared by Doping of 4H-SiC. Microscopy and Microanalysis, 12, 1546–1547. https://doi.org/10.1017/S1431927606064932

  22. Tao, X., Zhang, X., Cheng, J., Wang, Y., Liu, F., Luo, Z. (2005). Synthesis of Novel Multi-Branched Carbon Nanotubes With Alkali-Element Modified Cu/MgO Catalyst. Chemical Physics Letters, 409, 89–92. https://doi.org/10.1016/j.cplett.2005.04.084

  23. Bals, S., Batenburg, K. J., Verbeeck, J., Sijbers, J., Van Tendeloo, G. (2007). Quantitative Three-Dimensional Reconstruction of Catalyst Particles for Bamboo-Like Carbon Nanotubes. Nano Letters, 7, 3669–3674. https://doi.org/10.1021/nl071899m

  24. Xin, H. L., Muller, D. A. (2009). Aberration-Corrected ADF-STEM Depth Sectioning and Prospects for Reliable 3D Imaging in S/TEM. J. Electron Microsc., 58, 157–165. https://doi.org/10.1093/jmicro/dfn029

  25. Midgley, P. A., Dunin-Borkowski, R. E. (2009). Electron Tomography and Holography in Materials Science. Nat. Mater., 8, 271–280. https://doi.org/10.1038/nmat2406

  26. Robertson, I. M., Schuh, C. A., Vetrano, J. S., Browning, N. D., Field, D. P., Jensen, D. J., Miller, M. K., Baker, I., Dunand, D. C., Dunin-Borkowski, R., et al. (2011). Towards an Integrated Materials Characterization Toolbox. J. Mater. Res., 26, 1341–1383. https://doi.org/10.1557/jmr.2011.41

  27. Houben, L., Bar Sadan, M. (2011). Refinement Procedure for the Image Alignment in High-Resolution Electron Tomography. Ultramicroscopy, 111, 1512–1520. https://doi.org/10.1016/j.ultramic.2011.06.001

  28. Van Aert, S., Batenburg, K. J., Rossell, M. D., Erni, R., Van Tendeloo, G. (2011). Three-Dimensional Atomic Imaging of Crystalline Nanoparticles. Nature, 470, 374–377. https://doi.org/10.1038/nature09741

  29. Scott, M. C., Chen, C.-C., Mecklenburg, M., Zhu, C., Xu, R., Ercius, P., Dahmen, U., Regan, B. C., Miao, J. (2012). Electron tomography at 2.4-ångström resolution. Nature, 483, 444–447. https://doi.org/10.1038/nature10934

  30. Van Dyck, D., Jinschek, J. R., Chen, F.-R. (2012). ‘Big Bang’ Tomography as a New Route to Atomic-Resolution Electron Tomography. Nature, 486, 243–246. https://doi.org/10.1038/nature11074

  31. Saghi, Z., Midgley, P. A. (2012). Electron Tomography in the (S)TEM: From Nanoscale Morphological Analysis to 3D Atomic Imaging. Annu. Rev. Mater. Res., 42, 59–79. https://doi.org/10.1146/annurev-matsci-070511-155019

  32. Leary, R., Midgley, P. A., Thomas, J. M. (2012). Recent Advances in the Application of Electron Tomography to Materials Chemistry. Acc. Chem. Res., 45, 1782–1791. https://doi.org/10.1021/ar3001102

  33. Van Tendeloo, G., Bals, S., Van Aert, S., Verbeeck, J., Van Dyck, D. (2012). Advanced Electron Microscopy for Advanced Materials. Adv. Mater., 24, 5655–5675. https://doi.org/10.1002/adma.201202107

  34. Chen, C.-C., Zhu, C., White, E. R., Chiu, C.-Y., Scott, M. C., Regan, B. C., Marks, L. D., Huang, Y., Miao, J. (2013). Three-Dimensional Imaging of Dislocations in a Nanoparticle at Atomic Resolution. Nature, 496, 74–77. https://doi.org/10.1038/nature12009

  35. Bals, S., Van Aert, S., Van Tendeloo, G. (2013). High Resolution Electron Tomography. Curr. Opin. Solid State Mater. Sci., 17, 107–114. https://doi.org/10.1016/j.cossms.2013.03.001

  36. Goris, B., De Backer, A., Van Aert, S., Gómez-Graña, S., Liz-Marzán, L. M., Van Tendeloo, G., Bals, S. (2013). Three-Dimensional Elemental Mapping at the Atomic Scale in Bimetallic Nanocrystals. Nano Lett., 13, 4236–4241. https://doi.org/10.1021/nl401945b

  37. Bals, S., Goris, B., Liz-Marzán, L. M., Van Tendeloo, G. (2014). Three-Dimensional Characterization of Noble-Metal Nanoparticles and Their Assemblies by Electron Tomography. Angew. Chem. Int. Ed., 53, 10600–10610. https://doi.org/10.1002/anie.201401059

  38. Jia, C. L., Mi, S. B., Barthel, J., Wang, D. W., Dunin-Borkowski, R. E., Urban, K. W., Thust, A. (2014). Determination of the 3D Shape of a Nanoscale Crystal With Atomic Resolution From a Single Image. Nat. Mater., 13, 1044–1049. https://doi.org/10.1038/NMAT4087

  39. Bladt, E., Pelt, D. M., Bals, S., Batenburg, K. J. (2015). Electron Tomography Based on Highly Limited Data Using a Neural Network Reconstruction Technique. Ultramicroscopy, 158, 81–88. https://doi.org/10.1016/j.ultramic.2015.07.001

  40. Venkatakrishnan, S. V., Drummy, L. F., Jackson, M., De Graef, M., Simmons, J., Bouman, C. A. (2015). Model-Based Iterative Reconstruction for Bright-Field Electron Tomography. IEEE Trans. Comput. Imaging, 1, 1–15. https://doi.org/10.1109/TCI.2014.2371751

  41. Bals, S., Goris, B., De Backer, A., Van Aert, S., Van Tendeloo, G. (2016). Atomic Resolution Electron Tomography. MRS Bull., 41, 525–530. https://doi.org/10.1557/mrs.2016.138

  42. Miao, J., Ercius, P., Billinge, S. J. L. (2016). Atomic Electron Tomography: 3D Structures Without Crystals. Science, 353, aaf2157. https://doi.org/10.1126/science.aaf2157

  43. Hayashida, M., Malac, M. (2016). Practical Electron Tomography Guide: Recent Progress and Future Opportunities. Micron, 91, 49–74. https://doi.org/10.1016/j.micron.2016.09.010

  44. Cervera Gontard, L., Cintas, J., Dunin-Borkowski, R. E. (2017). The Benefit of Thresholding Carbon Layers in Electron Tomographic Tilt Series by Intensity Downshifting. J. Microsc., 265, 298–306. https://doi.org/10.1111/jmi.12498

  45. SALVE news (2010). 3D Structure of Carbon Soot. salve-project.de, August 11, 2010. https://www.salve-project.de/newspress/3d-structure-of-carbon-soot.html

  46. O’Leary, C. M., Sha, H., Zhang, J., Su, C., Kahn, S., Jiang, H., Zettl, A., Ciston, J., Miao, J. (2024). Three-Dimensional Structure of Buried Heterointerfaces Revealed by Multislice Ptychography. Phys. Rev. Applied, 22, 014016. https://doi.org/10.1103/PhysRevApplied.22.014016