A new machine-learning tool developed by researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), Boston Children’s Hospital (BCH), and Harvard Medical School is transforming how doctors assess fetal health through MRI scans. The tool, called Fetal SMPL, provides a more accurate and detailed 3D representation of fetuses by leveraging advanced modeling techniques to interpret complex MRI data. Traditional ultrasounds offer 2D images that reveal basic information such as fetal size, position, and certain abnormalities. For more detailed views, doctors may turn to MRI, which generates 3D volumetric scans. However, interpreting these scans is challenging because the human brain is not naturally equipped to process depth and internal structures in 3D. This is where Fetal SMPL comes in. Adapted from the SMPL (Skinned Multi-Person Linear) model originally designed for adult body shapes and poses, Fetal SMPL is tailored specifically for fetuses. It uses a kinematic tree of 23 articulated joints to simulate realistic fetal movements and body configurations. Trained on 20,000 real MRI volumes, the model learns to predict the size, shape, and pose of fetuses with remarkable precision—misaligning by only about 3.1 millimeters on average, less than the width of a grain of rice. This level of accuracy allows clinicians to make highly reliable measurements, such as head or abdominal circumference, and compare them to established benchmarks for fetal development. In early testing, the system successfully aligned real-world MRI scans with high consistency, demonstrating its potential for clinical use. Lead researcher Yingcheng Liu, a PhD student at MIT and CSAIL researcher, explains that the system overcomes the difficulty of modeling fetuses in the confined space of the uterus by using an internal skeletal structure that guides pose estimation. The model uses a coordinate descent algorithm, iteratively refining its predictions of shape and pose until it arrives at a stable, accurate result. While the current version focuses on surface-level anatomy—specifically bone structures beneath the skin—the team aims to expand the tool to model internal organs like the liver, lungs, and muscles. This would allow for a more comprehensive assessment of fetal health, including early signs of developmental issues. Experts outside the research team have praised the innovation. Kiho Im, associate professor at Harvard Medical School and staff scientist at BCH, notes that the method significantly improves the diagnostic value of fetal MRI and could help uncover how fetal movements relate to brain development. Sergi Pujades, an associate professor at University Grenoble Alpes, highlights the broader implications: Fetal SMPL bridges the gap between adult and infant body models, enabling long-term studies of human growth and development. This opens new possibilities for understanding how health conditions affect shape and motion from the earliest stages of life.