Research theme
Structure: brain atlases and segmentation
Spatiotemporal atlases of the developing brain, and the tools to label every scan against them, including in the presence of acoustic shadows and incomplete views.
Research / Brain atlases and segmentation
The problem
Fetal brain anatomy changes week by week, and ultrasound images are noisy and partly hidden by acoustic shadows from the skull. Labelling structures reliably needs a reference for what a typical brain looks like at each gestational age.
Our approach
- Build the atlasA normative atlas of fetal brain maturation from large international cohorts (INTERGROWTH-21st).
- RegisterAlign each new scan to the atlas at the right gestational age.
- LabelSegment brain structures, with the atlas guiding the labels where shadows block the view.
- Model shapeDescribe how each structure’s shape and size change across gestation.
Key papers
All papers in this theme (8) →- Semi-supervised 3D Medical Segmentation from 2D Natural Images Pretrained Model
- Anatomically plausible segmentations: Explicitly preserving topology through prior deformations
- STAMP: Simultaneous Training and Model Pruning for low data regimes in medical image segmentation
- TEDS-Net: Enforcing Diffeomorphisms in Spatial Transformers to Guarantee Topology Preservation in Segmentations
- Low-Memory CNNs Enabling Real-Time Ultrasound Segmentation Towards Mobile Deployment
- Improving U-Net Segmentation with Active Contour Based Label Correction
- Multi-task CNN for Structural Semantic Segmentation in 3D Fetal Brain Ultrasound
- Cortical Plate Segmentation Using CNNs in 3D Fetal Ultrasound