Why teaching AI about symmetry could make prenatal diagnosis more reliable.
Researchers at MIT, Boston Children's Hospital and Inria built E(3)-Pose, a deep learning model that uses built-in symmetry to track a moving fetal head and help capture complete MRI scans of the brain.
Researchers at MIT, Boston Children's Hospital and Inria have developed a deep learning model called E(3)-Pose. It is meant to help image a fetus's head while the fetus is moving, so radiologists can examine a high-resolution scan. The team presented a paper on the work at the 2026 European Conference on Computer Vision.
The problem starts with how fetal MRI works. A routine prenatal ultrasound can lead a sonographer to flag possible brain abnormalities for follow-up MRI. That scan stacks a series of 2D image slices into one 3D picture of the head. If the head shifts during the scan, slices can be captured from different angles, leaving gaps and unusual orientations that may cause a physician to miss or misdiagnose a problem. First author Ramya Muthukrishnan, an MIT PhD student, said: "If you have coverage gaps, you can’t see critical regions."
According to the researchers, E(3)-Pose tracks motion well because of how physics is built into it. Instead of expecting the model to learn physical principles from data alone, the team embedded the left-right symmetry of the fetal head directly in the network's architecture. That makes it easier for the model to identify the orientation of the brain.
Muthukrishnan came to the project with a computer science background and little knowledge of MRI physics. She said that running the model on the scanner crashed the machine many times. Early in her PhD she shadowed a radiologist at Boston Children's and saw a pregnant patient grow uncomfortable during an hour-long scan and stop partway through, leaving the radiologist without a full, high-quality image. Co-author Borjan Gagoski developed a sequence that acquired additional scans for training the model, and co-author Aryn Lee recruited the pregnant volunteers.
Ellen Grant, who directs the Fetal Neonatal Neuroimaging and Developmental Science Center at Boston Children's, said the work will automate fetal brain MRI and replace a slow, manual process with faster, more comfortable scans and better images. The model was trained mainly on third-trimester fetuses, so Muthukrishnan hopes to extend it to the second trimester, when fetuses are smaller and move more. The researchers say it could also apply to other hard-to-image anatomy and to autonomous driving.
Note: The headline refers to more reliable diagnosis, but the page describes a tool for capturing better fetal MRI scans and reports no diagnostic accuracy results.
At a Glance
- The model, E(3)-Pose, was developed by researchers at MIT, Boston Children's Hospital and Inria.
- The paper was presented at the 2026 European Conference on Computer Vision.
- Left-right symmetry of the fetal head is built into the model's architecture.
- The model was trained mainly on third-trimester fetuses; extending it to the second trimester is the next goal.
- Funders include the MIT Jameel Clinic, the MIT Takeda Program, the National Institute of Biomedical Imaging and Bioengineering and the Eunice Kennedy Shriver NICHD.
Sources
This brief was written by Brightside Report from the reporting linked above. Spot an error? Request a correction.