Researchers create a unified microstructure segmentation method

WASHINGTON DC: Researchers have developed a technology that can automatically identify and quantify materials microstructure from microscopic images using human-in-the-loop machine learning.

Microscopic imaging systems see information on material structure at numerous scales, from the nanoscale to the mesoscale.

The process of obtaining structural data from microscopic pictures is known as quantitative analysis of microstructure. However, given to the complexity and diversity of microstructure, humans and AI have had several limits in performing this alone.

The research team created an integrated framework for quantitative microstructure analysis by efficiently merging human and AI capabilities. This approach allows AI to execute microstructure segmentation with a single microstructure image and its related scribbling annotation by domain experts.

Furthermore, the AI interacts with humans by actively asking for scribbling annotations from professionals in order to improve the model's performance and reliability.

The research team confirmed that the framework of human-AI collaboration is ubiquitous and can be applied to a wide range of materials, microstructures, and microscopic imaging systems through comprehensive trials.

Previously, enormous volumes of dense annotation were necessary for past research; however, this study significantly lowered annotation expenses by substituting dense annotation with scribbling annotation, which can be simply produced using a pen or mouse.

This technology will be implemented into KIMS's Automated Microstructure Quantitative Analysis System (TIM). This will make it simple to utilise for general researchers. Dr. Juwon Na, a senior researcher at KIMS, said, “This study is the result of improving the existing subjective and time-consuming quantitative analysis of microstructure into an objective and automated process.”

Professor Seungchul Lee of POSTECH, added, “Our framework that interacts with experts is expected to be widely used as a core analysis technology in industry and research, and through this, we expect to dramatically reduce the cost and time of new materials research and development and further significantly improve reliability.”