In the realm of fetal health, the team at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), together with collaborators from Boston Children's Hospital (BCH) and Harvard Medical School, has introduced a machine-learning tool that steps beyond traditional 2D ultrasounds and MRIs to provide doctors with a richer 3D portrait of fetal development. This tool, dubbed "Fetal SMPL", is adapted from a 3D model initially meant to capture adult body shapes and is able to create detailed 3D representations of fetuses, complete with a kinematic tree of 23 joints, enabling it to mimic fetal poses seen in MRI scans. The research, aimed at enhancing the diagnostic capabilities of prenatal monitoring, was reportedly demonstrated at the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in September, according to a statement obtained by MIT News.
The technology, which learned from an impressive dataset of 20,000 MRI volumes, boasts an average misalignment of just 3.1 millimeters, offering measurements for elements like a baby's head or abdomen that could be compared against those of a healthy fetus at the same developmental stage. Lead author and MIT PhD student Yingcheng Liu explained, "It can be challenging to estimate the shape and pose of a fetus because they’re crammed into the tight confines of the uterus," detailing the methodology that uses a coordinate descent algorithm to alternately guess pose and shape from complex data until a reliable estimate is achieved, as MIT News outlined.
When comparing the tool's capabilities against a system that models infant growth, the team reduced that model by 75 percent for a fair comparison and found that Fetal SMPL could recreate real scans with higher precision, aligning closely with real MRIs. The tool also demonstrated efficiency in matching 3D models to images, needing minimally three iterations for a reasonable alignment on initial tests. These promising results are, however, just a starting point; researchers highlighted the importance of applying the system to larger populations and a variety of disease cases to validate and understand the system's broad-spectrum clinical utility further.
While the model currently focuses on external fetal body representation due to the skeletal-like nature of the structure beneath the skin, the research team plans to enhance their model to include the internal anatomy of fetuses for a more comprehensive analysis of fetal health," Liu told MIT News, addressing the exploration of inner organ development such as the liver, lungs, and muscles. Experts unaffiliated with the study have acclaimed the tool; Kiho Im from Harvard Medical School highlighted its potential “to improve the diagnostic utility of fetal MRI, but also provide insights into the early functional development of the fetal brain in relation to body movements.”
The broader significance of the Fetal SMPL lies not only in its immediate clinical application but also in its compatibility with adult and infant body models, which opens doors to long-term studies of human shape and pose development over time, noted Sergi Pujades from the University Grenoble Alpes in remarks shared with MIT News. The cross-collaborative effort brought together minds from the BCH, Inria, and MIT, a testament to interdisciplinary innovation in pursuit of advancing prenatal care and understanding.
