HKUST's GrainBot: AI Toolkit That Quantifies Microstructure from Microscopy Images
Researchers at The Hong Kong University of Science and Technology (HKUST) built GrainBot to do one thing very well: turn microscopy images into reliable numbers. It automates grain segmentation, feature measurement, and correlation analysis-closing the gap between image-heavy workflows and data-driven materials research.
The system uses a convolutional neural network for precise grain segmentation, then applies custom algorithms to measure grain surface area, grain-boundary groove geometry, and related surface metrics. "This work highlights the broader relevance for emerging AI-driven scientific infrastructures," said GUO Yike, Provost and Chair Professor at HKUST.
Why this matters
Manual microstructure analysis is slow and inconsistent, which limits statistical power and cross-study comparability. GrainBot converts visual information into standardized descriptors, enabling large, shareable databases and faster structure-property exploration.
How GrainBot works
The pipeline integrates segmentation → measurement → correlation analysis. Beyond counting grains, it uses interpretable machine-learning models to test how features interact-for example, how grain surface area and groove angle jointly affect surface concavity depth. The output is a set of quantitative features ready for modeling, ranking, and design-of-experiments loops.
Validated on perovskite thin films
The team validated GrainBot on metal halide perovskite thin films-key materials for high-efficiency solar cells-using AFM images to build a database of thousands of grains. The analysis surfaced relationships among grain size, groove geometry, and surface roughness that were previously hard to quantify. The study was published in Matter on February 26, 2026 (Matter, Cell Press).
"GrainBot illustrates how AI can transform complex microscopy images into structured, reproducible datasets that can be readily shared, re-analyzed and integrated into larger research platforms," added Prof. GUO. Prof. ZHOU Yuanyuan emphasized accessibility: "Our goal is to lower the barrier for integrating microscopy characterization into data-driven studies and autonomous laboratory platforms."
What this enables for your lab
- Standardized microstructure metrics for structure-property studies and process optimization
- Faster hypothesis testing using interpretable models and correlation analysis
- Improved reproducibility across projects, instruments, and teams
- Direct fit with autonomous workflows-continuous, machine-readable outputs for decision systems
- Shareable datasets that support meta-analyses and cross-lab benchmarking
Technical highlights
- CNN-based grain segmentation tuned for precision on complex micrographs
- Feature extraction: grain surface area, grain-boundary groove angle and geometry, surface concavity depth, and related roughness metrics
- Correlation engine with interpretable models to probe feature interactions
- Outputs ready for downstream statistical analysis, modeling, and lab automation pipelines
Outlook
The framework is applicable to other polycrystalline thin films and will be used to study links between microstructure and device stability. As scientific workflows become more automated, toolkits like GrainBot will keep labs supplied with consistent, analysis-ready data. For broader context on integrating AI into research pipelines, explore AI for Science & Research.
Your membership also unlocks:
