An AI model from MIT tracks a moving fetus during MRI, aiming for clearer scans of the developing brain

Researchers at MIT, Boston Children's Hospital and France's Inria have developed an AI model that tracks a fetus's head as it moves during an MRI scan, so the scanner can adjust and capture a complete, high-resolution image of the developing brain, MIT News reported on 19 October. The work was presented at the 2026 European Conference on Computer Vision.

Movement leaves gaps in fetal brain scans

When a routine prenatal ultrasound flags a possible brain abnormality, doctors often order an MRI. The scan builds a 3D picture from a stack of 2D “slices,” but if the fetus moves, each slice can be taken from a different angle, leaving gaps and odd orientations that can lead to missed or wrong diagnoses.

“If you have coverage gaps, you can't see critical regions, and you need to be able to see well to detect abnormal development,” said first author Ramya Muthukrishnan, a PhD student in the lab of MIT professor Polina Golland.

Muthukrishnan said she chose the project after shadowing a radiologist at Boston Children's Hospital and seeing pregnant patients spend an hour in the scanner only to stop halfway, without the radiologist getting a usable image.

The model uses the symmetry of the head

The model, called E(3)-Pose, estimates the position and rotation of the fetal head from quick, low-resolution 3D scans, so the scanner can then set the angle of the detailed slices. Instead of learning the geometry from data alone, it has the left-right symmetry of the head built into its design, which helps it work out which way the head is facing. In the paper, the team reports state-of-the-art accuracy on clinical MRI volumes and better robustness than existing methods.

Getting it to run on a real scanner took persistence. Co-author Borjan Gagoski of Boston Children's Hospital and Harvard Medical School developed an MRI sequence that collected extra scans for training and helped troubleshoot when the scanner crashed, and research assistant Aryn Lee recruited the pregnant volunteers.

The work “will automate fetal brain MRI, replacing today's slow, manual process with faster and more comfortable scans for mothers.” — Ellen Grant, director of the Fetal Neonatal Neuroimaging and Developmental Science Center at Boston Children's

Still a research tool, trained mostly on late pregnancy

E(3)-Pose is not yet in routine clinical use, and MIT did not say when it might be. The model was trained mainly on third-trimester fetuses. Muthukrishnan wants it to work in the second trimester, when fetuses are smaller and move more, because “the earlier you detect brain abnormalities, the easier it is to intervene.” The code is publicly available. The work was funded by the MIT Jameel Clinic, the MIT Takeda Program, the Chou Family Transformative Research Fund and two US National Institutes of Health institutes.