Speaker
Description
In my presentation, I will shortly summarize the main outputs of HE EIC project MHz tomoscopy [1] yielding not only the working prototype of apparatus at EuXFEL, but also already interesting examples of applications attracting new user community.
One of the bottlenecks on road to full use of MHz tomoscopy is proper reconstruction of rapid dynamics having only limited number of projections due to hardware limitations. In order to recover 4D (3D plus time) experimental data, apriori information about the physical processes and materials need to be supplied, complementing the insufficient information from projections only. In our case, we are dealing with haemodynamics and interactions of therapeutic ultrasound and vascular system at microscopic level. Using MHz tomoscopy, ultrasound processes such as fractionation of arterial calcifications can be watched at (sub)microscopic scale in near native, optically non-transparent samples, opening opportunity to optimize the intravascular litotrypsy (IVL) protocol and hopefully even more. In our search for proper data representation and processing, we have adapted 3D Gaussian splatting to X-ray phase contrast by giving each splatting primitive absorption and phase-shift values and embedding differentiable Fresnel propagation for scene projection into such compressed representation. Trained on twelve simulated red-blood-cell holograms, it recovers interpretable phase-dominant volumes in 15 minutes on one GPU. The learned fields are physically interpretable: the phase-shift map delineates RBC boundaries and biconcave morphology while the absorption map stays near zero, confirming that the optimisation recovers the correct phase-dominant material distribution without explicit phase retrieval [2].
References
[1] https://tomoscopy.eu
[2] Moško, D.; Szeles, P.; Draci, I.; Vagovič, P.; Uličný, J. Gaussian Splatting for Sparse-View X-Ray Phase-Contrast Reconstruction in MHz Tomoscopy. In Review July 1, 2026. https://doi.org/10.21203/rs.3.rs-10201845/v1
