Researchers at MIT, Boston Children's Hospital and Inria have developed an artificial intelligence model called E(3)-Pose to help produce clearer brain images before birth. During routine ultrasound checks, possible brain problems can lead to an MRI scan for a closer look. However, poor-quality images can cause doctors to miss a problem or identify one incorrectly.
An MRI scan takes a series of flat images, known as slices, which are combined into a detailed three-dimensional picture. If the baby moves its head, the slices may come from different angles, leaving gaps in the final image. E(3)-Pose helps track this movement. Its design includes the left–right symmetry of the baby's head, rather than requiring it to learn this physical principle only from examples. This helps it identify the brain's position and direction more accurately.
The model was mainly trained using scans from the final three months of pregnancy. The researchers now want to improve it for younger babies, which have more space to move in the womb. Detecting brain problems earlier can make treatment easier.
