The complex dance of human evolution has spanned an immense stretch of time, yet our understanding of fetal development is still a rapidly evolving field. This intricate process can be marred by various disturbances, from genetic quirks to environmental stressors and metabolic changes. These interruptions can significantly impact the development of a fetus — particularly the brain, one of the earliest organs to form.
Typically, potential aberrations are flagged during routine prenatal ultrasounds. Sonographers, the technicians who carry out these procedures, often spot signs of possible brain abnormalities, leading to more thorough investigations with MRI scans. Regardless, the nuances of current imaging techniques can sometimes result in overlooked or incorrect diagnoses. This situation poses a distinct challenge for expectant parents, health care providers, and the wider medical community.
In a commendable joint endeavour by MIT, Boston Children’s Hospital, and Inria, a pioneering deep learning model has been developed. Known as E(3)-Pose, this model is primed to bolster early detection and treatment of fetal brain abnormalities. It’s designed to capture high-resolution images of a moving fetus’s head, giving radiologists a more detailed view and thereby aiding in precise examinations. This ground-breaking research was unveiled at the 2026 European Conference on Computer Vision.
The key obstacle in using MRI is the potential for misalignment of ‘slices’ due to fetal movement, leading to gaps and odd orientations in the final scan. Ramya Muthukrishnan, an MIT PhD student and the first author of the E(3)-Pose project, explains that such disruptions often hamper the detection of vital brain areas and complicate the identification of developmental anomalies.
E(3)-Pose stands out in its exceptional capacity to track fetal motion, thanks to the incorporation of basic physics into its neural network. One such principle is the left-right symmetry of the fetal head, which empowers the model to accurately determine brain orientation. The initial journey wasn’t a smooth one though; Muthukrishnan had to overcome numerous challenges, including a shortage of background knowledge in MRI physics. Despite multiple mishaps and crashes with the scanner, her determination to overcome the difficulties observed during fetal MRI procedures at Boston Children’s Hospital never waned.
The venture was a collective effort, with valuable support from a team of high-ranking experts like Borjan Gagoski from Harvard Medical School and Boston Children’s Hospital, and Aryn Lee, who helped recruit MRI scan volunteers. Senior authors like Elfar Adalsteinsson, Ellen Grant, and Polina Golland provided their insights and guidance throughout the undertaking. Ellen Grant, the leader of the Fetal Neonatal Neuroimaging and Developmental Science Center, highlighted how transformative this project is. It not only speeds up and simplifies fetal MRI processes but also offers higher-quality images for analysis.
As for Muthukrishnan’s ambitions, they reach beyond just refining fetal MRIs. She sees potential for E(3)-Pose in other imaging challenges and even unrelated fields like autonomous driving. She’s particularly keen on developing the model further for imaging fetuses in their second trimester, which tend to be smaller and more active. Muthukrishnan’s focus on early detection is an integral aspect of this venture, as timely intervention can significantly alter outcomes.
The project is supported by the MIT Jameel Clinic, the MIT Takeda Program, the Chou Family Transformative Research Fund, and several national institutes. Further information on this research can be found in the original news article on the MIT news site. Interested in how AI automation can revolutionize your business processes? Find out more at implementi.ai.
This website uses cookies.