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

Abstract

Automated image analysis has been applied to scanning electron micrographs (transmission mode; STEM) of metallic nanoparticles (silver and gold; about 10 nm to 20 nm). For a reliable particle identification, scanning electron microscopic images must be recorded with distinct contrast and resolution parameters. The particles were separated from the background and classified according to shape and size by machine learning (machine learning). Training images were created with model particles cut out of real electron microscopic images. The automated analysis of the particle size (expressed as area) was well possible, but overlapping particles could not be safely separated. The assignment of particle to six different shape classes (sphere, triangle, square, pentagon, hexagon, rod) by automated analysis was difficult. The fact that real particles never have an ideal geometrical shape but are always distorted or have rough edges or cropped tips is the fundamental reason of this problem. This effect also occurred with human image evaluators and poses a considerable obstacle in the training process for machine learning. Image analysis by machine learning techniques is difficult if different human evaluators disagree on the shape assignment of given particles because a proper training cannot be provided.

Deep learning for automated size and shape analysis of nanoparticles in scanning electron microscopy

“…For a precise size analysis, agglomerates must therefore be excluded. 3 We achieved this by explicitly training the classification network to identify agglomerates.…”

“…As confidence limit we used the typical human certainty when classifying a given particle of about 75%. 3 Note that the classes for particles identified as “covered” and “agglomerate” (STEM only) were defined as individual classes during training. This assignment was not perfect which is not surprising because there are many different shapes for partially covered or agglomerated particles.…”

“…This is a time-consuming process which may be biased by a human reviewer. 3 To accomplish a fast and objective analysis, there is a strong need for non-biased, rapid methods to analyse SEM images. 4 Machine learning has been applied to detect and identify objects of high variability in SEM images.…”

“…This is a time-consuming process which may be biased by a human reviewer. 3 To accomplish a fast and objective analysis, there is a strong need for non-biased, rapid methods to analyse SEM images. 4 …”

Rapid classification of micro-particles using multi-angle dynamic light scatting and machine learning approach

“…Lugnan et al (2020) developed a machine learning method for high-throughput single particle analysis using flow cytometry to achieve interference pattern classification of transparent PMMA microparticles with diameters of 15.2 and 18.6 μm. Bals et al (2022) used scanning electron microscopy images to record contrast ratio and resolution, then classified the acquired images by machine learning based on the shape and size of micro-particles. The above methods are limited to imaging analysis and require more space for expensive detection instruments (Klug et al, 2019;Di et al, 2022;Yue et al, 2022).…”