Scanning While Imagining: A Scene-Graph World Model for Robotic Ultrasound Navigation
SonoGraph-WM plans robotic ultrasound by predicting future anatomical scene graphs instead of images.
SonoGraph-WM is an action- and goal-conditioned world model for robotic ultrasound that represents anatomy as scene graphs and jointly forecasts future graphs and probe poses with a Transformer. A receding-horizon planner imagines trajectories and follows a short execution horizon before replanning. On four held-out CT cases, spatial-relation F1 stayed above 93% over 20 steps, with closed-loop success of 77.50% for the gallbladder and 75.00% for the pancreas; robot-phantom trials reached the target view in 73.7% of runs.
- Predicts future anatomical scene graphs and probe poses, not images
- Training data generated from CT label maps along probe trajectories
- Spatial-relation F1 stayed above 93% over 20 prediction steps
- Closed-loop success 77.5% gallbladder and 75% pancreas
- Phantom robot trials reached the target view in 73.7% of runs
- AI models
- SonoGraph-WM
From the original article
1 min read at arxiv.orgUltrasound (US) acquisition depends on the operator's ability to interpret anatomy and anticipate how the view will change with probe motion. Many robotic US navigation methods select actions without explicitly predicting these anatomical changes. We propose SonoGraph-WM, an action- and goal-conditioned world model for anticipatory probe navigation.
By Xuesong Li · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.