From reconstruction to interpretation: zero-setup deployable model for multi-phase segmentation of X-ray tomography data

X-ray tomography enables the nondestructive characterization of materials, and recent advances in micro-CT imaging technology have accelerated the acquisition and reconstruction of volumetric data. However, rapid scientific interpretation remains limited by image segmentation, which typically requires manual thresholding, user prompting, or material-specific model training before meaningful analysis can begin. We present a zero-setup deployment framework for multi-phase segmentation of synchrotron X-ray tomography data that produces interpretable segmentations of previously unseen tomography datasets without user prompting, manual parameter tuning, or additional retraining during deployment. The framework combines a material-agnostic deployment strategy based on reusable attenuation- and morphology-based mask categories, including background, sample, bright regions, dark-gray regions, light-gray regions, and porosity, with a pretrained semantic segmentation network to generate rapid first-pass segmentations within minutes of reconstruction. In inference-time benchmarks, the framework segmented a 500-slice \:2560×\:2560 volume in 67.1s and a 2000-slice volume in 253.5s using four A100 GPUs. Unlike conventional deep learning pipelines that require dataset-specific annotations and retraining, the proposed approach can be applied immediately to new scans, enabling rapid assessment of experimental quality and preliminary microstructural analysis during ongoing beamline experiments. Users can subsequently refine or fine-tune the generated masks when application-specific segmentation is required. Experiments on held-out and qualitatively evaluated synchrotron micro-CT datasets suggest that the framework produces consistent and physically meaningful first-pass segmentations on new tomography datasets while substantially outperforming conventional intensity-based thresholding. By bridging the gap between high-speed data acquisition and immediate interpretation, the proposed framework enables near real-time beamline feedback and supports scalable AI-assisted scientific imaging workflows.