AI system guides ultrasound technicians to correct fetal brain imaging planes in real time

A team of researchers has built an artificial-intelligence system designed to help sonographers consistently capture the correct anatomical views during fetal brain ultrasound exams—a task that currently relies heavily on operator skill and experience.

The Challenge in Clinical Practice#

Identifying and holding the correct imaging planes during fetal ultrasound is notoriously difficult. Technicians must navigate through three-dimensional ultrasound data while maintaining precise probe positioning, and even small deviations result in images that are unsuitable for clinical assessment. There is significant variability in performance depending on sonographer experience.

How the System Works#

The researchers created a two-part approach combining segmentation and pose estimation. Their semi-supervised model learns from labeled standard-plane images alongside unlabeled scans from 3D ultrasound volumes, achieving strong accuracy on both types of data. Critically, the system includes a classification step that filters out frames lacking fetal brain tissue before attempting to measure plane proximity—a pragmatic design choice that increases clinical relevance.

Rather than simply declaring "yes, this is the correct plane" or "no, it isn't," the pipeline provides continuous distance metrics, showing how far the current probe position is from the ideal orientation at any moment.

Real-World Testing and Deployment#

The team validated their approach on actual fetal scan videos captured by 17 sonographers with varying levels of expertise. The results revealed that experienced operators tend to freeze the image near—but not exactly at—the point of maximum proximity to the standard plane, mimicking natural clinical behavior. Interestingly, proximity measurements alone did not predict how an expert would rate the final image quality.

The system also demonstrated clinical viability by running on an edge device (NVIDIA Clara AGX) at 39 frames per second, well above the speed required for real-time clinical scanning.

Implications for Prenatal Care#

This research represents a step toward image-guided navigation in ultrasound scanning. While the system shows promise as a complementary tool to existing fetal ultrasound technologies, the findings also hint at the complexity of clinical expertise—proximity to an ideal plane is necessary but not sufficient to guarantee diagnostic quality. The work opens a path toward more standardized, accessible prenatal imaging across diverse clinical settings.