Automated Facet and Volume Segmentation of Nanoparticles in Simulated CTEM Images Using Deep Learning

Area of Science:

• Materials Science and Engineering
• Nanotechnology
• Computational Imaging

Background:

• Nanoparticles are crucial in materials science, with applications in biomedicine and electronics.
• Electron microscopy is key for understanding nanoparticle properties, but manual analysis is time-consuming and labor-intensive.
• Artificial intelligence, especially deep learning, offers automated solutions for image analysis, but specialized models for nanoparticle segmentation are scarce.

Purpose of the Study:

• To evaluate deep learning models for segmenting facets and volumes of gold nanoparticles in simulated electron microscopy images.
• To assess the effectiveness of YOLOv11n-seg, a lightweight instance-segmentation model, for this task.
• To explore the potential of transfer learning for efficient nanoparticle analysis.

Main Methods:

• Simulated transmission electron microscopy (CTEM) images of gold nanoparticles (decahedral and icosahedral) were generated.
• The YOLOv11n-seg deep learning model was employed for instance segmentation.
• Transfer learning techniques were utilized to adapt the model for nanoparticle segmentation.

Main Results:

• The YOLOv11n-seg model demonstrated capability in segmenting facets and volumes in simulated nanoparticle images.
• The lightweight architecture of YOLOv11n-seg suggests potential for efficient computational performance.
• Performance was evaluated within the simulation domain, highlighting the need for experimental validation.

Conclusions:

• Deep learning, specifically lightweight models like YOLOv11n-seg, can be a viable approach for automated nanoparticle segmentation.
• The study provides a foundation for developing efficient AI tools for analyzing electron microscopy data.
• Further validation using experimental micrographs is essential for real-world application of the developed model.