Pablo Villanueva Perez: X-ray imaging opportunities at MAX IV: novel methods + end-to-end AI

Abstract:
MAX IV, the world’s first operational diffraction-limited storage ring, enables X-ray imaging with spatiotemporal resolutions several orders of magnitude beyond those achievable at previous-generation synchrotron facilities. Realizing this potential, however, requires new imaging methods and reconstruction algorithms tailored to the unique properties of diffraction-limited storage rings.

This talk will present (i) novel methods that push the limits of spatiotemporal resolution, and (ii) end-to-end unsupervised AI frameworks that make these methods feasible. We will first introduce X-ray Multi-Projection Imaging, a technique capable of recording 4D (3D+time) datasets at least two orders of magnitude faster than current state-of-the-art approaches. Because this technique relies on inherently sparse measurements, it demands reconstruction schemes that effectively incorporate prior knowledge. AI-driven approaches have proven particularly powerful for this purpose. In the second part, we will discuss end-to-end AI+physics frameworks that integrate modern machine learning with fundamental models of X-ray interaction, propagation, image formation, and sample dynamics. We will highlight how these AI+physics approaches can drive advances in (i) phase retrieval under coherent illumination, (ii) high-quality 3D and 4D reconstructions from sparse spatiotemporal data, and (iii) efficient analysis pipelines for large-scale experiments. Together, these AI+physics approaches and novel methods enable high-speed, high-fidelity X-ray imaging at previously unattainable spatiotemporal resolutions, opening new possibilities for materials science, biology, and industrial applications.