University of Oxford OMNI Oxford Machine
Learning in
NeuroImaging Lab

Shadow reduction in ultrasound using differentiable simulation and radiance field decomposition

Our new preprint introduces RFlash, which reduces acoustic shadows in ultrasound images using a differentiable ultrasound simulator.

Acoustic shadows from the skull hide deeper brain anatomy in fetal ultrasound. RFlash splits each ultrasound image into attenuation and scatter with a differentiable simulator, then reduces the shadows. The work was led by Valentin Bacher, with collaborators including Bernhard Kainz and Michael Gray.

You can compare results with the original images using interactive sliders, on fetal brain and abdominal scans, on the project page. The code is available on GitHub.

Read the preprint: arXiv:2609.29373.

View citation on our publications page

Previous post
Two new papers on fetal brain mapping and neurodevelopmental outcomes