Researchers at MIT, Boston Children’s Hospital, and the French research institute Inria have built an AI model to make fetal brain scans more reliable. The model is called E(3)-Pose.
It tracks the movement of a fetus’s head during an MRI scan. The goal is to capture clear, high-resolution images that radiologists can review.
A paper on the work was presented at the 2026 European Conference on Computer Vision. MIT shared details of the project on October 19, 2026.
Why Fetal Movement Causes Problems in MRI Scans
During a routine prenatal ultrasound, a sonographer may flag possible brain problems. In those cases, doctors often order an MRI for a closer look.
A fetal MRI takes a series of flat 2D images, called slices. When stacked, these slices form one 3D image of the fetal head.
If the fetus moves, each slice may be taken from a different angle. This can leave gaps in the final scan and lead to missed or incorrect 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, an MIT PhD student.
How E(3)-Pose Works
The model is built around physics. Instead of learning physics only from data, the left-right symmetry of the fetal head is built directly into the neural network.
This design makes it easier for the model to find the brain’s orientation as the fetus moves. The scan can then be adjusted on the fly.
Muthukrishnan works in the lab of Polina Golland, an MIT professor of electrical engineering and computer science. She said moving the model from a computer to a real scanner was hard, and she crashed the scanner many times.
She chose the project after watching fetal MRI scans at Boston Children’s Hospital. Some pregnant patients spent an hour in the scanner and then stopped halfway, leaving radiologists without a full, clear image.
Borjan Gagoski, an assistant professor of radiology at Harvard Medical School, developed a scan sequence to help train the model. Research assistant Aryn Lee recruited pregnant volunteers and helped collect the scans.
Ellen Grant directs the Fetal Neonatal Neuroimaging and Developmental Science Center at Boston Children’s. She said the work “will automate fetal brain MRI,” leading to faster, more comfortable scans and better images.
The researchers said the model could also help image other hard-to-see body parts. It may even have uses in self-driving cars, which track 3D objects with limited sensors.
The model is mostly trained on scans of third-trimester fetuses. Muthukrishnan now hopes to adapt it for second-trimester fetuses, which are smaller and move around more.
“The earlier you detect brain abnormalities, the easier it is to intervene,” she said.
