MIT embeds symmetry into neural network to fix fetal brain MRI, E(3)-Pose presented at ECCV 2026
MIT News AI · rss · 2026-10-20
Researchers at MIT, Boston Children's Hospital and Inria built E(3)-Pose, a deep learning model that corrects fetal motion in brain MRI, presented at ECCV 2026.
The problem: Fetal MRI stacks 2D slices into a 3D image, but fetal movement misaligns slices, creating coverage gaps that can cause radiologists to miss or misdiagnose brain abnormalities.
The approach: Rather than learning physics from data, physical priors are baked into the architecture — left-right symmetry of the fetal head (equivariance) is embedded in the network, making orientation estimation far more accurate.
Clinical impact: First author Ramya Muthukrishnan was motivated by watching pregnant patients abandon hour-long scans halfway. Co-authors at Harvard Medical School developed extra MRI sequences for training data. The team says it could automate fetal brain MRI, replacing slow manual workflows with faster, more comfortable scans and higher-quality images.
Next steps: The model is trained mostly on third-trimester fetuses; extending to more mobile second-trimester fetuses is a priority, since earlier detection enables easier intervention. The method could extend to other hard-to-image anatomy and even autonomous driving's 3D tracking.
More from Embodied
- Robotics practitioner: LLM-generated 'robotics startups' all chase the same fake problems — kscottz · 2026-10-20
- Maker Arm starter robot arm now runs Dream-1, bringing zero-shot robotics to more builders — gharik · 2026-10-20
- GPT-6 Astra in spatial XR: Jarvis-style live Earth demo on a PICO device — Scobleizer · 2026-10-20
- Long queues at RTX Spark SF launch for local Qwen demos built with ODS software — TheZachMueller · 2026-10-20
- "There are no robot foundation models, just foundation models" — tkipf · 2026-10-20
- Dream-1 Beta runs robot tasks at $1.72 per success, up to 41x cheaper than LLM agents — gharik · 2026-10-20