medicine
New AI captures fetal brain scans in one visit
A mother quit the scan halfway. Someone was watching.
Researchers at MIT, Boston Children's Hospital, and Inria have built a deep learning model that tracks a moving fetus during an MRI scan, aiming to capture clear brain images in one sitting and spare families repeat hospital visits. The tool, called E(3)-Pose, was presented at the 2026 European Conference on Computer Vision. During a fetal MRI, the head shifts constantly, scattering the 2D image slices and leaving gaps where critical brain detail should be. The model bakes the left-right symmetry of the fetal head into its architecture to figure out orientation. First author Ramya Muthukrishnan arrived with a computer science background and no MRI physics. She learned on the hardware, which fought back. "I was not prepared for how many times I would crash the scanner," she says. The project started when she shadowed a radiologist and watched a mother quit a scan halfway through, uncomfortable after an hour. Next, she wants it to work on second-trimester fetuses, which move more because they have the room.
