Ultrasound world model enables interactive probe pose control
SonoWorld: Towards Interactive Pose-Controllable Ultrasound World Models
Computer Vision and Pattern Recognition
Summary
Getting steady ultrasound images is hard because tiny movements of the probe cause big changes in what the ultrasound shows. The authors created SonoWorld, a computer model that learns from lots of ultrasound videos to predict how an ultrasound image changes as the probe moves in different directions. Their model can be controlled by realistic probe positions, letting robots or humans simulate and plan ultrasound scanning better. This helps make ultrasound scanning more consistent and less reliant on operator skill.
What this means in practice
- •For robotic system builders: Create simulation environments that predict ultrasound images responsive to probe movements for training and developing robotic ultrasound scanning policies.
- •For medical device developers: Develop ultrasound machines with built-in interactive simulators to support operator training and improve image acquisition consistency.$Commercial implications: The model enables selling ultrasound devices or software enhanced with realistic simulation and pose control for training purposes.
Authors
Zefan Yang, Zheyuan Zhang, Boris Maihe, Yikang Liu, Hang Yu, Ankush Mukherjee, Xiao Chen, Pingkun Yan, Terrence Chen, Shanhui Sun
Abstract
Robotic ultrasound can standardize image acquisition and reduce operator workload and variability, but policy learning remains limited by the lack of realistic, controllable simulation environments. While controllable world models have advanced rapidly in natural-image domains, ultrasound world modeling remains challenging because each frame captures only a cross-section of unobserved 3D anatomy, causing small probe motions to induce large appearance changes. Existing approaches rely on robot-tracked poses and latent-space prediction, but do not directly generate the observations required for interactive control. We propose SonoWorld, a causal ultrasound world model that learns video dynamics from unlabeled clinical videos and grounds them to 6-DoF control using a small set of freehand clinical scans tracked with an EM sensor. SonoWorld is trained in three stages: pretraining a latent-action-conditioned video diffusion model on around 6.9 million unlabeled multi-organ ultrasound frames; grounding the learned video dynamics to metric probe poses with a small EM-tracked thyroid dataset; and applying distribution matching distillation to enable causal, long-horizon autoregressive generation with selectively caching eviction tokens at inference time. We evaluate controllability through pose-following fidelity, round-trip consistency, and axis-specific control under synthetic actions on held-out tracked scans, and further validate it with a user study. In model-based action discrimination, SonoWorld correctly ranks the target-directed probe motion over an equal-magnitude opposite motion in up to 74.8% of trials. To our knowledge, SonoWorld is the first causal, pose-controllable ultrasound world model learned primarily from freehand clinical videos, providing a data-driven interactive environment for robotic ultrasound data acquisition and policy evaluation.