Ultrasounds have become somewhat of a staple in the process of prenatal care, offering a fascinating glimpse into the unborn world. For expectant parents, it’s a cherished first “meeting” with their child. For clinicians, these monochrome, two-dimensional images give invaluable insight into fetal development, such as identifying the baby’s sex or detecting potential abnormalities like heart defects or a cleft lip.
Sometimes, if doctors need a more in-depth look, they resort to the magnetic resonance imaging (MRI). MRIs use magnetic fields to produce detailed, layered images that can be merged to form a 3D view of the fetus. Although interpreting these in-depth 3D MRI scans can be complicated due to our natural visual system not being adept at processing complex volumetric data, the field of machine learning now steps in to assist.
Introducing “Fetal SMPL,” a new machine learning model designed by a cooperative team from the Computer Science and Artificial Intelligence Laboratory at MIT, Boston Children’s Hospital, and Harvard Medical School. This innovative model brings a new level precision to the process by creating more accurate 3-dimensional representations of fetal health through the modeling of their shape and movements.
Fetal SMPL is a derivative of SMPL, otherwise known as the Skinned Multi-Person Linear model – a 3D modeling framework initially developed for adult body shapes and poses. The researchers trained their fetal edition on over 20,000 MRI volumes; it learned to anticipate the size and position of fetuses, creating nearly sculpture-like 3D representations. Each model includes a sophisticated system of 23 interconnected joints, accurately mirroring the motion of the fetus.
Let’s talk precision. The predictions made by Fetal SMPL were off by an average of an almost unbelievable 3.1 millimeters, a size smaller than a grain of rice. This extraordinary level of detail empowers clinicians to take measurements of vital anatomical features such as the size of the head or abdomen, and subsequently compare these results to the standardizations for healthy development at particular gestational ages.
To put the system to test, the research team stacked Fetal SMPL against SMIL, another model developed to document the growth of infants. Even after making necessary adjustments to make a fair comparison—scaling down the SMIL model by 75% to match the size of a fetus—Fetal SMPL triumphed over SMIL.
The accuracy of Fetal SMPL isn’t its only strength—it proved to be efficient too. The model reached a reliable alignment with the MRI data in a mere three iterations, demonstrating its strong performance.
Right now, Fetal SMPL is concentrating on the fetus’s exterior shape and skeletal structure, which is a substantial step forward in itself. However, this is just the beginning. The team aims to improve the model further by including the fetus’s internal structure — organs and muscles — that can contribute to monitoring critical entities such as lung and liver development. Should these plans materialize, it would revolutionize the model into a comprehensive volumetric representation, providing an even more profound insight into fetal health.
Fetal SMPL not only promises to enhance prenatal diagnostics but also potentially deepen our understanding of fetal evolution. It’s compatible with existing models for adults and infants, laying a robust foundation for extensive studies on human development. The implications of these advancements, needless to say, are significant and lead the way to potentially life-changing discoveries and innovations.
While still in the early stages, with further testing and refinement, Fetal SMPL could become an integral part of prenatal care, benefiting both clinicians and expectant parents by presenting a clearer and better-rounded picture of life, quite literally, in the making.
For more details on the project and its potential impact, you can visit the original article from MIT News here.
This website uses cookies.