• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Tuesday, November 17, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Technology

Neural Networks Advance Atomic Electron Tomography for Three-Dimensional Nanomaterial Characterization

by
November 12, 2026
in Technology
Reading Time: 5 mins read
0
Neural Networks Advance Atomic Electron Tomography for Three-Dimensional Nanomaterial Characterization

Neural Networks Advance Atomic Electron Tomography for Three-Dimensional Nanomaterial Characterization

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

Accurate determination of three-dimensional atomic arrangements remains essential for understanding and tailoring the functional properties of innovative nanomaterials across catalysis, electronics, energy storage, quantum technologies, and structural engineering. While transmission electron microscopy has long served as a primary instrument for structural evaluation at the atomic scale, conventional transmission imaging only generates two-dimensional projections that cannot capture complete three-dimensional atomic configurations. Scanning probe microscopy resolves surface topographies but cannot inspect subsurface volumes, whereas crystallographic techniques requiring periodic symmetry fail to characterize non-periodic features such as grain boundaries, dislocations, point defects, and interfaces. Atom probe tomography provides three-dimensional coordinates destructively, rendering dynamic in situ studies difficult, and coherent diffractive imaging has not yet attained true atomic resolution. In contrast, atomic electron tomography has emerged as a non-destructive characterization methodology that achieved an atomic coordinate precision of nineteen picometers in 2015 by pairing aberration-corrected transmission electron microscopy with advanced iterative reconstruction algorithms.

Following its initial experimental validation, atomic electron tomography successfully investigated an extensive array of complex atomic features, including point defects, stacking faults, local lattice distortions, heterointerfaces, amorphous structures, and three-dimensional strain tensors. Because atomic electron tomography does not destroy examined specimens, it enabled four-dimensional studies combining spatial coordinates with temporal evolution to track phenomena such as nucleation and crystal growth. Moreover, experimentally measured atomic coordinates have been directly coupled with ab initio theoretical calculations to correlate atomic configurations with chemical, physical, and electronic performance. However, atomic electron tomography faces an intrinsic geometric challenge known as the missing wedge problem, where specimen holders or transmission grids obstruct the imaging electron beam at high tilt angles. This restricted angular sampling introduces Fourier ringing artifacts and elongation along the optical axis of the beam, which heavily distort reconstructions and present significant obstacles to determining three-dimensional surface atomic structures accurately.

Resolving surface configurations is critically important because surface atoms primarily govern catalytic reactivity, adhesion, corrosion resistance, electronic transport, and interfacial phenomena across diverse nanomaterials. To mitigate distortions caused by missing angular projections, researchers have integrated deep learning models, particularly convolutional neural networks, into atomic electron tomography workflows. These computational frameworks exploit the atomicity principle, a physical prior assuming that the imaged material consists entirely of discrete atomic potentials, to overcome conventional reconstruction constraints. A comprehensive review published by Juhyeok Lee and Yongsoo Yang in Applied Microscopy examines recent computational and experimental developments in neural network-assisted atomic electron tomography. Their survey details how machine learning architectures recover unmeasured data, eliminate severe reconstruction artifacts, and significantly improve three-dimensional atomic coordinate precision for both internal and surface atomic sites across diverse nanoscale systems.

One prominent computational strategy involves deploying neural networks to recover missing angular data prior to or during volumetric reconstruction. Ding and colleagues established a two-step deep learning pipeline that initially infills uncollected tilt angles in the sinogram domain using a generative adversarial network built with residual-in-residual dense blocks. The resulting sinogram is then reconstructed into a three-dimensional volume, after which an encoder-decoder network featuring skip connections suppresses remaining artifacts. This combined workflow generated reconstructions exhibiting substantially higher structural fidelity than classical algorithms such as weighted back-projection and the simultaneous algebraic reconstruction technique, even when over eighty percent of the tilt angular range was omitted. This framework was subsequently expanded into the information recovery and de-artifact model, which successfully achieved a sub-angstrom resolution of point seven angstroms during the electron tomographic analysis of nanoporous gold structures.

An alternative deep learning methodology applies three-dimensional neural networks as an augmentation step directly following initial volumetric reconstruction. In 2021, Lee and collaborators introduced a post-reconstruction workflow utilizing a three-dimensional U-Net architecture to refine raw tomograms generated by iterative reconstruction algorithms such as GENFIRE. Because initial reconstructions suffer from missing wedge elongation and spatial blurring, the three-dimensional U-Net was trained with the atomicity constraint to transform distorted density distributions into isolated atomic peaks characterized by expected Gaussian intensity profiles. By learning to enforce discrete atomic potentials, the model generalized effectively to structures distinct from its training data. This augmentation approach improved atomic coordinate precision to fifteen point one picometers, demonstrating that machine learning can resolve blurred density profiles into distinct atomic coordinates even under demanding experimental angular limits.

Building upon U-Net concepts, Yu and coworkers developed the ensemble cross U-Net transformer model, designated EC-UNETR, to further optimize three-dimensional tomogram reconstruction. The EC-UNETR architecture incorporates hierarchical subnetworks that process structural feature maps along orthogonal axes, integrating a skip fusing unit that normalizes and concatenates intermediate outputs to enhance operational stability and computational flexibility. In addition, transformer-based self-attention mechanisms are embedded within both encoder and decoder stages to capture long-range spatial dependencies across the volumetric data. This hybrid transformer design enhanced the recovery of fine structural details, suppressed residual tomographic artifacts, and demonstrated a five point five percent reduction in root-mean-square error relative to previous neural network models, confirming the benefits of attention mechanisms for atomic-resolution image reconstruction.

The quantitative benefits of neural network augmentation were demonstrated through the experimental characterization of individual platinum nanoparticles. In an investigation analyzing an isolated platinum nanoparticle containing nearly fifteen hundred atoms, applying a three-dimensional U-Net significantly improved atom detectability and positional accuracy. The fraction of correctly traced atoms rose from ninety-six point five percent in the raw tomogram to ninety-eight point eight percent after neural network processing, while the root-mean-square deviation of atomic coordinates dropped from twenty-six point one picometers to fifteen point one picometers. Furthermore, the crystallographic R-factor comparing experimental tilt images with forward projections of the reconstructed atomic coordinates decreased from nineteen point two percent to seventeen point four percent. The enhanced reconstruction revealed low-coordination surface atoms, enabled identification of facet-dependent strain between the {100} and {111} surfaces, and detected localized compressive strain at the particle-support interface.

Beyond single particles, deep learning-assisted atomic electron tomography enabled detailed structural investigation of particle coalescence interfaces. Researchers examined a dumbbell-shaped platinum nanoparticle formed by the fusion of two clusters, reconstructing its complete three-dimensional atomic lattice. The analysis identified a double twin boundary at the contact junction alongside notable atomic disorder throughout the adjoining lattice. By mapping the full three-dimensional strain tensor from the determined atomic coordinates, investigators detected pronounced tensile strain concentrated along protruding {100} facets. Coupling these experimental strain values into density functional theory calculations established a quantitative correlation between the localized lattice strain and enhanced oxygen reduction reaction catalytic activity. This demonstration highlighted how deep learning reconstruction provides reliable atomic coordinates that serve as direct empirical inputs for quantum mechanical modeling of functional catalytic properties.

In addition to addressing missing tilt angles, neural networks have been developed to eliminate substrate background interference in supported catalyst studies. Iwai and colleagues implemented a convolutional neural network image inpainting methodology to isolate metal nanoparticle signals from substrate scattering in experimental tilt series. By predicting and subtracting background contributions from an alumina support, the model allowed unobstructed reconstruction of an eleven-nanometer palladium nanoparticle. The resulting three-dimensional tomogram unveiled a deformed cuboctahedral morphology with high-index facets linked to elevated catalytic methane combustion. Measuring local palladium-palladium interatomic bond distances revealed interfacial strain and structural disorder at the palladium-alumina boundary. This preprocessing inpainting approach broadens atomic electron tomography to supported catalytic systems where substrate overlap previously hindered quantitative atomic tracing.

Despite these successes, authors Lee and Yang emphasize several ongoing physical and methodological constraints in atomic electron tomography. At room temperature, surface atoms undergo spontaneous thermal diffusion on timescales between 10^-4 and 10^-7 seconds, while high imaging electron doses on the order of 10^5 electrons per square angstrom can induce structural perturbations. Furthermore, the absence of universally validated experimental ground truth benchmarks complicates performance evaluations. Future progress will require physics-informed neural network architectures alongside emerging experimental techniques, such as four-dimensional scanning transmission electron microscopy and multislice ptychography. High-speed pixelated detectors can resolve both light elements like oxygen and heavy metallic atoms while overcoming depth-of-focus limitations. Coupled with unsupervised denoising networks for low-dose and liquid-phase imaging, neural network-assisted tomography continues to advance toward robust, high-throughput atomic characterization.

Subject of Research: Neural network integration in atomic electron tomography for three-dimensional atomic structural determination

Article Title: Advancing atomic electron tomography with neural networks

Article References: Lee, J., & Yang, Y. (2025). Advancing atomic electron tomography with neural networks. Applied Microscopy, 55(1), Article 7. https://doi.org/10.1186/s42649-025-00113-7

Image Credits: AI Generated

DOI: 10.1186/s42649-025-00113-7

Keywords: atomic electron tomography, transmission electron microscopy, neural networks, deep learning, nanomaterials, missing wedge problem, convolutional neural networks, 3D reconstruction, Advancing, atomic, electron, tomography

News Source: Cassandra Pierce. (November 12, 2026). Neural Networks Advance Atomic Electron Tomography for Three-Dimensional Nanomaterial Characterization. Scienmag.

Tags: 3D reconstructionAdvancingatomicatomic electron tomographyconvolutional neural networksdeep learningelectronmissing wedge problemnanomaterialsNeural Networkstomographytransmission electron microscopy
Share12Tweet7Share2ShareShareShare1

Related Posts

PUCC Framework Cuts Cloud GPU Training Time by 75 Percent

PUCC Framework Cuts Cloud GPU Training Time by 75 Percent

November 17, 2026
Digital Health Literacy Alone Does Not Ensure Active Rheumatic Disease Decisions

Digital Health Literacy Alone Does Not Ensure Active Rheumatic Disease Decisions

November 17, 2026

Plant-Synthesized Silver Nanoparticles Enhance Black Gram Seedling Growth in Laboratory Tests

November 17, 2026

Offshore Wind Wake Predictions Vary Up to Four Percent Across Software Implementations

November 17, 2026

POPULAR NEWS

  • Aprepitant Increases Tacrolimus Clearance for Three Weeks After Stem Cell Transplantation

    Aprepitant Increases Tacrolimus Clearance for Three Weeks After Stem Cell Transplantation

    29 shares
    Share 12 Tweet 7
  • Refractory Generalized Myasthenia Gravis Incurs Substantially Higher Economic Burden in China

    29 shares
    Share 12 Tweet 7
  • Cerebral Autoregulation Index Fluctuates Widely During One-Lung Ventilation in Two Cases

    29 shares
    Share 12 Tweet 7
  • Bacteriophages Suppress Resistant Pseudomonas aeruginosa Isolates in Laboratory Assessment

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Aprepitant Increases Tacrolimus Clearance for Three Weeks After Stem Cell Transplantation

Refractory Generalized Myasthenia Gravis Incurs Substantially Higher Economic Burden in China

Cerebral Autoregulation Index Fluctuates Widely During One-Lung Ventilation in Two Cases

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 88 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.