SMAXI: a machine-learning-powered open-source software for multidimensional full-field X-ray image analysis

Full-field X-ray imaging techniques using laboratory-based and synchrotron sources are powerful and versatile tools used by academia and various industries to non-destructively characterize the internal structures of samples. In particular, synchrotron high-speed radiography and computed tomography (CT) have helped researchers address many critical problems by providing multiscale and multidimensional structural information. Owing to increased source brightness and improved spatial and temporal resolutions of modern detection systems, the generation rate of X-ray imaging data has accelerated dramatically in recent years. Consequently, manually analyzing the massive volume of X-ray image data has become a persistent bottleneck that hinders fast scientific discoveries and data-driven insights. To address this challenge, we developed SMAXI (Software for Machine-Learning-assisted Analysis of X-ray Images), an open-source software designed for analyzing multidimensional full-field X-ray image data. SMAXI distinguishes itself from other conventional closed-source software by being fully customizable and open-source. It is GPU-accelerated and uses state-of-the art machine learning (ML) models to analyze time-resolved multidimensional X-ray imaging data. SMAXI's capabilities for image processing and analysis are demonstrated here using 2D static X-ray images, dynamic in situ X-ray videos, and complex volumetric CT data. By integrating advanced ML algorithms into a single customizable workflow, SMAXI can pre-process X-ray images using computer vision algorithms and automate object segmentation and tracking tasks, while utilizing a large language model-based chatbot for interactive geometry feature analysis. Ultimately, SMAXI will empower X-ray community members to overcome big-data limitations, accelerating the pace of scientific discovery that incorporates full-field X-ray imaging techniques.