AI Data Segmentation & Quantitative Electron Microscopy
SEM Segmentation · Human GT Drag to compare · Use arrow keys when focused · Reference annotation shown
A single microscopy image can contain thousands of particles and interfaces, while 4D-STEM and EELS generate vast multidimensional datasets. Turning these observations into meaningful measurements requires reliable identification of the regions we want to analyze. AEML develops AI-assisted segmentation workflows that separate particles, coatings, pores, and other regions of interest, then quantify their size, shape, and spatial distribution. By connecting these regions with diffraction and spectroscopy, we extend local observations into statistically meaningful materials analysis.
From images to measurements
Electron microscopy reveals fine detail, but an image alone does not tell us how many particles are present, how their sizes are distributed, or how uniformly a coating covers a surface. Segmentation turns the regions of interest in an image into objects we can measure. We develop AI-assisted workflows that identify particles, coatings, pores, and interfaces and connect their geometry with quantitative materials analysis.
The aim is to compare entire populations across fields of view, not just a few selected features. This provides a basis for detecting changes after processing, describing sample-to-sample variation, and testing which microstructural features are associated with performance.
Why particle boundaries are a difficult problem
A bimodal microstructure can look easy to interpret: large rounded particles coexist with smaller, irregular particles. Yet similar contrast, surface texture, overlapping particles, and agglomeration obscure their boundaries. A model may split one particle into several objects or merge neighboring particles into one. Both errors can distort the resulting size distribution even when the overlay looks plausible.
Our approach combines microscopy-informed annotation, image processing, and AI segmentation. The definition of an object follows the scientific question: separating individual particles, identifying a coating, and measuring pore connectivity require different criteria. We examine the resulting masks against reference annotations and assess how errors affect the measurements used to compare materials.
Beyond morphology: structure and chemistry
The value of a segmented region grows when it can be connected to other microscopy signals. EDS / EELS can reveal composition and bonding, while 4D-STEM provides information on phase, orientation, and strain. Linking these signals to the same region would allow us to ask whether a particular particle population has different chemistry, whether a coating remains continuous, or whether structural changes concentrate near an interface.
We are developing analysis strategies that connect spatial alignment, region selection, and signal integration. Averaging within physically meaningful regions can improve interpretability while retaining differences that disappear in a whole-image average. The goal is a coherent description of each feature—its shape, its structure, and its chemical environment—rather than separate maps that are difficult to compare.
Measurements that guide materials design
Useful quantitative analysis must remain consistent across fields of view and samples. Imaging conditions, spatial resolution, boundary particles, and overlapping features all influence the statistics. We therefore consider segmentation quality together with sampling and measurement definitions, so that changes in an analysis pipeline are not mistaken for changes in the material.
Our workflows connect particle size and shape distributions, area fractions, coating coverage, and spatial relationships with processing conditions and material response. The outcome is a set of measurements that helps choose the next experiment. By identifying which populations change, where heterogeneity develops, and which features correlate with degradation, quantitative microscopy can support process optimization and more focused materials design.