X-ray Nano-Holotomography Reconstruction

X-ray nano-holotomography (XNH) is a powerful technique for imaging biological samples at nanometer resolution, but reconstruction from raw data is a hard inverse problem. Raw measurements encode phase and absorption information in a complex, entangled way that requires careful physical modeling to disentangle.

We are building physics-based reconstruction pipelines in JAX that combine phase retrieval, Paganin-based regularization, and large-scale optimization on multi-GPU infrastructure. The pipelines are designed to handle the full complexity of real XNH data — including angular discontinuities in sinograms, inter-window background offsets, and noise — while remaining accessible and reproducible for experimentalists.

This project is a collaboration with Jakob Reichmann at the Crick Institute, and builds directly on the differentiable simulation framework developed in Chromatix.

Key methods: JAX · Phase retrieval · Inverse problems · Large-scale optimization · Multi-GPU computing

Collaborators: Jakob Reichmann, The Francis Crick Institute