Illuminating the Void: Data-Consistent Diffusion Reconstruction for Brain Imaging in X-Ray Laminography

by Wenxuan Fang

Thursday 17 Sept 2026, 11:30 → 13:00 Europe/Zurich

OHSA/B17

Abstract:

Mapping neural circuits requires three-dimensional imaging at nanometer resolution across large volumes of brain tissue. X-ray laminography is a promising approach for imaging extended biological specimens, but its acquisition geometry leaves part of Fourier space unmeasured, creating a severe missing-cone problem that degrades reconstruction quality.

In this talk, I will introduce LUCID, a data-consistent reconstruction framework that combines multi-view diffusion priors with the physical forward model of X-ray laminography. By alternating between generative inference and measurement-consistency updates, LUCID recovers missing structural information while remaining faithful to the acquired data.

I will provide an intuitive introduction to tomography, laminography, and diffusion models for inverse problems, and demonstrate how the integration of physics and AI enables high-fidelity reconstruction of brain tissue from incomplete measurements.

TEAMS link