Quantifying Monomer-Dimer Distribution of Nanoparticles from Uncorrelated Optical Images Using Deep Learning

Area of Science:

• Materials Science
• Nanotechnology
• Biomedical Research

Background:

• Nanoparticles in polymer matrices enhance composite material properties.
• Accurate detection and quantification of nanoparticles are vital for understanding their behavior.
• Existing methods often rely on destructive imaging techniques like SEM/TEM.

Purpose of the Study:

• To develop an automated deep learning approach for nanoparticle detection and oligomerization quantification in optical images.
• To overcome limitations of optical imaging such as noise and low contrast.
• To provide a non-destructive alternative to SEM/TEM for nanoparticle analysis.

Main Methods:

• Fine-tuning the YOLOv8 deep neural network architecture.
• Utilizing a dataset of correlated optical and SEM images of gold nanospheres (AuNSs).
• Training on annotated images to identify AuNSs and their oligomeric states.

Main Results:

• Achieved a weighted average accuracy of 80.7% for AuNS quantification and oligomerization state determination.
• Demonstrated superior performance compared to manual image processing methods.
• Validated speed and effectiveness on high-density optical images.

Conclusions:

• The proposed optical image-based method offers accurate and efficient nanoparticle analysis.
• This technique is applicable to live samples for in vivo imaging, enabling studies on nanoparticle uptake and interactions.
• Potential applications include cell signaling, drug delivery, and understanding nanoparticle-cell interactions.