Computational ultrasound: geometry and physics
Methods that combine geometric reasoning with physics-informed models of how ultrasound images are formed: from recovering the 3D anatomy of the fetal brain from freehand 2D scans, to reducing artefacts such as acoustic shadows.
The problem
One 2D slice shows only part of the brain. A 3D volume can be cut in any plane and measured as a whole, but 3D probes are expensive and rare where most pregnancies are scanned. Freehand 2D ultrasound is widely available; the challenge is recovering 3D anatomy from it.
Our approach
Two ideas run through this work.
Geometric reasoning
Every freehand frame is a slice through one underlying 3D anatomy. We estimate where each slice sits in space and enforce consistency between overlapping views, drawing on multi-view 3D vision.
Physics-informed models
Ultrasound is not a camera: images form as sound travels through tissue in the probe plane, attenuating and scattering. We build this image formation into reconstruction, which is also what allows acoustic shadows to be reduced rather than simply filled in.
- ScanA routine freehand 2D sweep, on any scanner.
- LocateEstimate each frame’s position and orientation (its 6-DoF pose) in a shared anatomical coordinate frame.
- ReconstructFuse the frames into a 3D volume with rendering models built around ultrasound’s slice-based image formation, rather than camera perspective.
- CorrectSeparate attenuation from scatter with a differentiable ultrasound simulator, reducing acoustic shadows from the skull.
Key papers
All publications →- Shadow Reduction in Ultrasound Imaging Using Differentiable Simulation and Radiance Field Decomposition
- UltraGauss: Ultrafast Gaussian Reconstruction of 3D Ultrasound Volumes
- RapidVol: Rapid Reconstruction of 3D Ultrasound Volumes from Sensorless 2D Scans
- Sensorless volumetric reconstruction of fetal brain freehand ultrasound scans with deep implicit representation
- Geometric Transformation Uncertainty for Improving 3D Fetal Brain Pose Prediction from Freehand 2D Ultrasound Videos