AI-Driven In Situ Electron Microscopy: From Image Acquisition to Autonomous Analysis
In situ electron microscopy (EM) enables direct observation of dynamic material processes, but experiments are often limited by noise, beam sensitivity, and the challenge of analysing large, cumbersome datasets. Artificial intelligence (AI) is increasingly addressing these constraints by improving both data quality and analysis within the experimental workflow.
Recent work in Ultramicroscopy (2025) demonstrates an approach applied to liquid cell TEM data acquired with the Poseidon AX system. Using deep learning, high-quality images are reconstructed from low-dose, noisy inputs while preserving structural detail (Figure 1). This is particularly valuable for beam-sensitive systems, where increasing electron dose is not feasible due to the high chance of radiolysis side reactions. In this case, AI enables meaningful interpretation of data that would be more difficult to interpret.
Figure 1. Using different ‘aquadenoising’ techniques on gold nanoparticles in a liquid cell results in clear images. Moncomble, A. et al. (2025) Ultramicroscopy, 271, 114121
AI is also advancing the analysis of dynamic processes. A study published in Microscopy and Microanalysis (2025) combined Fusion AX with our AXON Synchronicity machine-vision software to show how machine learning can automate the tracking of grain boundary evolutions under high temperature and at the nano-scale (Figure 2). By identifying and following interfaces across frames, these approaches enable quantitative measurement of structural evolution with improved consistency and throughput compared to manual methods.
Figure 2. Grain boundaries in a 100 nm thin Al film tracked by hand or by the U-net model. Patrick, M.J. et al. (2023) Microscopy and Microanalysis, 0, 1–12
Additionally, machine learning is being applied to electron diffraction to analyse the diffraction patterns automatically (Figure 3). A recent study in Current Applied Physics (2024) highlights how AI can assist with diffraction pattern interpretation at high temperature using the Fusion AX system to heat the sample. The written software supports tasks such as phase identification and structural classification. This is particularly relevant for in situ experiments, where diffraction data evolves rapidly.
Figure 3. Temperature induced phase transformation in FeOOH nanoparticles, analysed using electron diffraction, and automatically analysed with the Random Sample Consensus algorithm. Lim, S. et al. (2024) Current Applied Physics, S156717392400110X
The developments in artificial intelligence for in situ microscopy reflect a broader shift toward integrated workflows in electron microscopy. Platforms such as AXON Synchronicity play a key role by enabling image enhancement, feature tracking, and analysis within a unified environment, supporting more efficient and quantitative in situ experimentation.
Learn more about AXON Synchronicity and AI in in situ microscopy below.
