Researchers at the University of Tokyo and the Innovation Center of NanoMedicine (iCONM) have developed an artificial intelligence (AI) approach that identifies the morphology of nanoparticles in liquid using data from standard nanoparticle tracking analysis (NTA), a widely used technique for particle sizing. The method achieved classification accuracies exceeding 80% for non-spherical nanoparticles without requiring modification of existing instruments.
The study, published in ACS Applied Nano Materials , demonstrates that information previously unused in standard NTA measurements can be extracted through deep learning to improve nanoparticle characterization.
Why nanoparticle morphology matters
The morphology of a nanoparticle strongly influences its behavior, shaping diffusion, optical characteristics, interactions with biological systems, and performance in fields ranging from drug delivery and extracellular vesicle research to environmental monitoring and advanced materials development.
However, characterizing nanoparticle morphology in liquid remains challenging. Techniques such as transmission electron microscopy can resolve fine structural details, but because they require drying or immobilizing samples, they are not suited for rapid, high‑throughput evaluation of particles in their native liquid environment.
Nanoparticle tracking analysis is widely used to measure particle size by recording the Brownian motion of individual nanoparticles suspended in liquid. Standard NTA systems record both the trajectories of these scattered‑light spots and their intensity signals. In practice, however, most analyses rely almost exclusively on trajectory information to estimate particle size, leaving other information contained in the measurements largely unexamined.
Extracting previously unused information from NTA data
The research team developed a deep-learning framework that integrates two types of information contained in NTA measurements: Brownian motion trajectories of individual nanoparticles and temporal fluctuations in scattered-light intensity.
The framework combines a one-dimensional convolutional neural network with a bidirectional long short-term memory network. This architecture enables simultaneous learning of motion-related and optical time-series patterns from individual particles. Rather than relying solely on standard statistical descriptors, the AI model analyzes the temporal behavior of both signals and identifies patterns associated with nanoparticle morphology.
The researchers evaluated the method using gold nanoparticles with three distinct morphologies: spheres, rods, and plates. In binary classification tasks involving pairs of particle types, the integrated approach consistently outperformed models based on a single information source. Using approximately one second of measurement data, corresponding to 100 frames, the binary classifiers achieved accuracies exceeding 0.82, indicating a high overall proportion of correctly identified particles.
Furthermore, the method also performed multi‑class classification of all three particle types, reaching an average class correctness of approximately 80%, indicating that each particle type was correctly identified at roughly this rate. Importantly, the model maintained stable performance even when the amount of available data was greatly reduced. Reliable classification was achieved using observation windows as short as about 0.2 seconds, corresponding to 20 frames, and under conditions with limited particle counts.
These results suggest that standard NTA measurements contain richer information than previously recognized, and that AI can effectively use this information for morphology‑related nanoparticle classification.
Expanding the value of existing instruments
A key feature of the approach is that it does not require new hardware. Instead of developing a new measurement instrument, the researchers enhanced the analytical value of data already produced by widely used NTA systems. This creates opportunities for upgrading existing nanoparticle measurement workflows through software-based analysis.
“Our goal is to translate this technology into practical nanoparticle analysis tools by incorporating it into future NTA systems,” said Professor Takanori Ichiki of the University of Tokyo and iCONM. “We hope it will contribute to medicine and industry, where nanoparticles are playing an increasingly important role.”
Because the method works with very small sample volumes and short measurement times, it could be particularly useful in areas where sample availability is limited. Potential applications include quality control of nanomedicines and nanoparticle-based therapeutics, characterization of extracellular vesicles and other biological nanoparticles, monitoring of engineered and environmental nanoparticles, and evaluation of colloidal materials.
The researchers believe that integrating AI with established nanoparticle characterization tools could enable more comprehensive evaluation of nanoparticle properties while preserving the simplicity and accessibility of existing measurement platforms. The study demonstrates how machine learning can make use of previously unused information contained in familiar experimental techniques and potentially expand the analytical capabilities of nanoparticle research without the need for new instrumentation.
Publication
Hiromi Kuramochi, Keisuke Yamamoto, Kento Toyoda, Yasushi Shibuta, and Takanori Ichiki, “Shape-Resolved Nanoparticle Analysis from Standard Nanoparticle Tracking Analysis via Integrated Motion and Scattering Signatures,” ACS Applied Nano Materials (2026). DOI: 10.1021/acsanm.6c01701
Funding
This work was supported by the Japan Science and Technology Agency (JST) Program on Open Innovation Platform with Enterprises, Research Institutes and Academia (COI-NEXT), Grant Number JPMJPF2202.
ACS Applied Nano Materials
Experimental study
Not applicable
Shape-Resolved Nanoparticle Analysis from Standard Nanoparticle Tracking Analysis via Integrated Motion and Scattering Signatures
19-Jun-2026
The authors declare no competing financial interest.