SCIENCE HIGHLIGHT 8
25 February 2025
A new AI-driven solution for automating TEM alignment
Using deep learning tools to automate alignment of electron microscopes can significantly enhance the efficiency of a time-consuming and expertise-intensive process
The integration of artificial intelligence (AI) into scientific instrumentation is more than just a trend, it is a transformative shift. A recent study published in the journal Ultramicroscopy explores the use of AI, particularly convolutional neural networks (CNNs), to automate the process of transmission electron microscope (TEM) alignment.
Developed by a research team from CEMES-CNRS, a partner in the IMPRESS project, this innovative method has the potential to significantly reduce the time and expertise required for TEM alignment, while improving efficiency.
The use of convolutional neural networks in electron microscopy
The study investigates how AI can radically improve TEM alignment. By leveraging deep learning, and more specifically CNNs, the research team has developed a method that predicts the necessary adjustments to realign the condenser aperture in a single step. This approach minimizes the traditionally time-consuming and expertise-intensive alignment process.
At the core of this innovation is the direct computer control of the microscope and simplified digital twins that allow the automatic acquisition of learning datasets. The CNNs are trained to predict the x- and y-shifts required for alignment, effectively replicating human-level estimation. This automation not only streamlines the process but also ensures consistency, a critical factor for long-duration experiments where drift can degrade data quality.
Implications for the IMPRESS Project and future prospects
The findings of this study align with the goals of the IMPRESS project, which aims to develop an interoperable platform for electron microscopy. By incorporating AI-driven automation, the project envisions a future where TEM components can be seamlessly integrated and customized across various scientific instruments, enhancing nanoscale research capabilities.
This research, led by experts from CEMES-CNRS and ENAC in Toulouse, France, exemplifies the spirit of scientific cross-disciplinary collaboration driving innovation. Their work provides a valuable contribution to the IMPRESS project’s mission and offers a glimpse into broader AI applications in microscopy and beyond.
The study highlights several key advancements and implications:
- Application of convolutional neural networks for TEM alignment
- Automation of the alignment process, reducing time and expertise required
- Capability for continuous correction of alignment drift
“The integration of AI into TEM alignment not only enhances precision but also aligns with the broader objectives of the IMPRESS project, paving the way for a new era of interoperability in scientific research”, concludes Martin Hytch, Research Director at CEMES-CNRS and IMPRESS Project Partner.
Publication details
Ultramicroscopy, 2024, 267, 114047
Principle of TEM alignment using convolutional neural networks: Case study on condenser aperture alignment
Loïc Grossetête, Cécile Marcelot, Christophe Gatel, Sylvain Pauchet and Martin Hytch