Reliable AI: The system helps you perform nanoparticle measurements and speeds up your research
Comparison of different segmentation algorithms applied to the same image of the TRIPS sample. Credit: Science Report (2025). doi:10.1038/s41598-025-86327-x
Nanoparticle researchers spend most of their time doing one thing. Count and measure nanoparticles. Each step in the road requires you to see the results. They usually do this. Although counting and measuring them takes a long time, this task is essential to complete the statistical analysis required to carry out the next properly optimized nanoparticle synthesis.
Alexander Witman is a professor of colloid chemistry at the University of Constance. He and his team repeat this process every day. “When I worked on my doctoral dissertation, we used a large particle counting machine for these measurements. It was like a cash register. And then, back then, when I could measure 300 nanoparticles a day, I was like, I was really happy,” Wittemann remembers.
However, reliable statistics require thousands of measurements for each sample. Today, the increasing use of computer technology means processes can move much faster. At the same time, automated methods are very error prone to researchers themselves having to perform many measurements, or at least recheck.
Even complex particles can count correctly
During the coronavirus pandemic, Good Luck contacted Whittemann with doctoral student Gabriel Monteiro. Wittemann and Monteiro have developed a program based on Meta’s open source AI technology, “Segment Anything Model.” This program allows for AI-supported counting of nanoparticles in microscopic images and subsequent automatic measurements of each individual particle.
Micrographs of three-leaf trimeric nanoparticles (TRIP) with (a) nanospheres (SPHPS), (b) two-leaf dumbbell nanoparticles (DIP), and (c) scale bars representing 250 nm. Throughout this study, these three particle sets were used as proof of concept for the automated SAM-based analysis presented here. Credit: Science Report (2025). doi:10.1038/s41598-025-86327-x
“The ‘basin method’ has worked very well so far, due to clearly definable particles. But our new method can also automatically count particles with dumbbells or caterpillar shapes made up of two or three overlapping sphere strings,” explains Wittemann. . “This saves a lot of time.
“The time usually takes to complete the particle synthesis and create the corresponding time-consuming measurements, allowing you to focus on particle synthesis and look under a microscope, but AI systems handle most of the rest. This was previously possible in a fraction of the time needed.
In addition to this, AI measurements are not only more efficient, but also more reliable. AI methods recognize individual fragments more accurately and measure them more accurately than others. As a result, subsequent experiments can be adapted and executed more accurately, leading to faster success in the test series.
This study has been published in the journal Science Reports.
Details: Gabriel AA Monteiro et al, Pre-trained Artificial Intelligence Supported Analysis of Nanoparticles Using All Models of Segment, Scientific Report (2025). doi:10.1038/s41598-025-86327-x
The research team has published new AI routines, the necessary code and data from Git-Hub and Kondata research, and data for other researchers to use and discuss.
Provided by the University of Constance
Citation: Reliable AI: The system helped to make nanoparticle measurements, and research obtained on February 12, 2025 from https://phys.org/2025-02 (2025, February 12)
This document is subject to copyright. Apart from fair transactions for private research or research purposes, there is no part that is reproduced without written permission. Content is provided with information only.
