Why teaching AI about symmetry could make prenatal diagnosis more reliable
Despite millions of years of evolution, the development of the human fetus remains a complex process that is vulnerable to genetic, environmental and metabolic disruptions. These disruptions could result in developmental abnormalities, starting with one of the first organs to begin developing in the fetus that continues to develop after birth: the brain.
The development of a human fetus is a complex process vulnerable to genetic, environmental, and metabolic disruptions during the first organs' formation. A prenatal ultrasound may flag potential brain abnormalities for further magnetic resonance imaging (MRI). However, imperfect imaging can lead to missed or incorrect diagnoses.
Researchers at MIT, Boston Children's Hospital, and Inria have developed a deep learning model called E(3)-Pose to enhance fetal head imaging during motion, resulting in high-resolution images for radiologists to examine. This model, presented at the 2026 European Conference on Computer Vision, addresses the issue of fetal movement causing inconsistent 2D images during MRI scans.
The model incorporates physics principles, particularly left-right symmetry, to improve the accuracy of brain orientation identification, reducing gaps and orientation issues in the final 3D scan. The development of E(3)-Pose was driven by the need to improve fetal MRI scans, which often face challenges due to fetal movement. Co-authors, including researchers from MIT, Boston Children's Hospital, and Inria, collaborated to overcome technical difficulties and develop a robust training sequence.
The model's potential applications extend beyond fetal MRI, including other hard-to-imaging anatomy and autonomous driving, where limited sensors need to capture noisy 3D data for object tracking.
Written by urgent.news from Medical Xpress's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
