Application of automated electron microscopy imaging and machine learning to characterise and quantify nanoparticle dispersion in aqueous media

Abstract

Summary For many nanoparticle applications it is important to understand dispersion in liquids. For nanomedicinal and nanotoxicological research this is complicated by the often complex nature of the biological dispersant and ultimately this leads to severe limitations in the analysis of the nanoparticle dispersion by light scattering techniques. Here we present an alternative analysis and associated workflow which utilises electron microscopy. The need to collect large, statistically relevant datasets by imaging vacuum dried, plunge frozen aliquots of suspension was accomplished by developing an automated STEM imaging protocol implemented in an SEM fitted with a transmission detector. Automated analysis of images of agglomerates was achieved by machine learning using two free open‐source software tools: CellProfiler and ilastik. The specific results and overall workflow described enable accurate nanoparticle agglomerate analysis of particles suspended in aqueous media containing other potential confounding components such as salts, vitamins and proteins. Lay Description In order to further advance studies in both nanomedicine and nanotoxicology, we need to continue to understand the dispersion of nanoparticles in biological fluids. These biological environments often contain a number of components such as salts, vitamins and proteins which can lead to difficulties when using traditional techniques to monitor dispersion. Here we present an alternative analysis which utilises electron microscopy. In order to use this approach statistically relevant large image datasets were collected from appropriately prepared samples of nanoparticle suspensions by implementing an automated imaging protocol. Automated analysis of these images was achieved through machine learning using two readily accessible freeware; CellProfiler and ilastik. The workflow presented enables accurate nanoparticle dispersion analysis of particles suspended in more complex biological media.

Quantifying Nanoparticle Assembly States in a Polymer Matrix through Deep Learning

“…The current model, which excels at many NP detection tasks compared with traditional computer tools, is deployed on GitHub. Different from a shallow pixel classifier, 42 the performance of the deep neural network predictor can still be further improved by enriching the dataset to include more diverse cases, such as nonspherical particles, extremely large or small particles, and particles at image boundaries. Detection of particles of other shapes, such as nanotubes and nanolayers, can also be made possible, but the model has to be carefully retrained.…”

Role of Artificial Intelligence and Machine Learning in Nanosafety

“…[62] Image profiling using open source CellProfiler and a CNN-based algorithm ilastik, [64] were employed by Ilet et al to study nanoparticle distributions. [65] A particularly impressive use of ML in nanosafety was published recently by Lazerovits et al [66] They conducted experiments aimed at understanding the adsorption of blood proteins on nanoparticles immediately after intravenous injection, how this interface changes during circulation, and how it alters to distribution of nanoparticles in vivo. They showed that the evolution of proteins on nanoparticle surfaces predicts the biological fate of nanoparticles in vivo.…”

Automated and manual classification of metallic nanoparticles with respect to size and shape by analysis of scanning electron micrographs

“…This can lead to particle size distributions after automated image analysis, and also to the quantification of agglomerates (number of primary particles, size, etc.) [13,17]. The classification of nanoparticles with respect to their shape is more difficult.…”