Related Experiment Video
Updated: Jul 12, 2025
09:11
Automation of the Micronucleus Assay Using Imaging Flow Cytometry and Artificial Intelligence
Published on: January 27, 2023
2.2K
UTILE-Gen: Automated Image Analysis in Nanoscience Using Synthetic Dataset Generator and Deep Learning
André Colliard-Granero1,2,3, Jenia Jitsev2,4, Michael H Eikerling1,2,3
1Theory and Computation of Energy Materials (IEK-13), Institute of Energy and Climate Research, Forschungszentrum Jülich GmbH, 52425 Jülich, Germany.
ACS Nanoscience Au
|October 23, 2023
Summary
This study introduces an AI workflow for autonomous nanoparticle image analysis, generating synthetic data to train models. This approach enhances microscopy image analysis accuracy for diverse nanoparticle shapes.
Area of Science:
- Nanoscience
- Materials Science
- Computer Vision
Background:
- Manual annotation of microscopy images for nanoparticle analysis is time-consuming and labor-intensive.
- Developing high-performance models for accurate nanoparticle characterization requires large, well-annotated datasets.
- Existing methods struggle with the diversity of nanoparticle shapes and sizes.
Purpose of the Study:
- To develop and implement a deep learning-based workflow for autonomous image analysis in nanoscience.
- To create a versatile, agnostic, and configurable tool for generating synthetic instance-segmented imaging datasets of nanoparticles.
- To demonstrate the effectiveness of synthetic data generation and deep learning for improving nanoparticle classification and segmentation.
Main Methods:
- A synthetic data generator employing domain randomization was developed to create image/mask pairs.
- Supervised deep learning models, specifically convolutional neural networks, were trained on the expanded synthetic dataset.
- The trained models were evaluated for classification and instance segmentation performance on various nanoparticle shapes (spherical, cubic, rod-shaped).
Main Results:
- The synthetic generator successfully produced diverse, instance-segmented imaging datasets.
- Training deep learning models on the expanded dataset significantly improved classification and instance segmentation performance.
- The autonomous workflow demonstrated high accuracy in analyzing microscopy images of nanoparticles with different morphologies.
Conclusions:
- Deep learning workflows utilizing synthetic data generation offer an efficient and accurate solution for autonomous nanoparticle image analysis.
- The developed tool and methodology overcome the limitations of manual annotation, enabling high-performance microscopy image analysis.
- This approach has broad applicability for accelerating research and development in nanoscience and materials characterization.
