Acoustic window planning for ultrasound acquisition

Rüdiger Göbl, Salvatore Virga, Julia Rackerseder, Benjamin Frisch, Nassir Navab, Christoph HennerspergerInt. J. Comput. Assist. Radiol. Surg 2017 · DOI

Introduction

Robotic ultrasound (US) systems hold the potential to standardize medical imaging, reducing the variability associated with manual sonography. However, a significant hurdle remains: most current robotic planners are "blind" to internal anatomy. They typically rely on simple geometric heuristics, such as positioning the probe perpendicular to the skin at the point nearest the target.

In practice, ultrasound physics is highly directional. Obstacles like ribs or air-filled lungs create "shadow zones" that obscure deep organs. This paper introduces a fully automated framework for planning robotic US trajectories that specifically target "acoustic windows"—paths that minimize attenuation and shadowing by leveraging pre-operative CT data.

Methodology: A Two-Stage Optimization Framework

The proposed system integrates geometric constraints with a physics-based transmission model to identify optimal probe placements. The process is divided into two primary stages:

  1. Geometric and Hard Constraints: The system first defines the search space based on the probe's physical properties, such as its field of view and maximum penetration depth. It also ensures proper acoustic coupling by requiring the probe to maintain contact with the patient's skin.
  2. Physics-Based Transmission Optimization: Using pre-operative CT scans, the system maps Hounsfield Units (HU) to acoustic impedance and attenuation coefficients. It then calculates a "transmission value" for thousands of potential probe poses. This estimate simulates the signal processing of a real US machine to predict image quality.
Overview of the two-stage optimization framework for acoustic window planning, moving from target identification to final pose selection.

The planning algorithm first optimizes the probe's rotation to maximize transmission through soft tissue gaps (like intercostal spaces). It then refines the surface position to ensure the target remains within the field of view while avoiding adverse objects like bone. This approach generalizes from single points to 3D "sweeps," identifying a single skin contact point that provides the best overall window for an entire organ.

Experimental Evaluation

The framework was validated using both synthetic data and physical experiments:

  • Synthetic Planning: The researchers utilized 20 annotated upper-torso CT scans from the SLIVER07 challenge dataset to test the algorithm across diverse human anatomies.
  • Robotic Phantom Trials: A custom tissue-mimicking gelatin-agar phantom was constructed, featuring a molded "liver" and chalk sticks to simulate the acoustic shadowing of ribs.

Key Results and Insights

The study demonstrates that the shortest geometric path to an organ is rarely the best path for ultrasound. By shifting the probe just a few centimeters to find an intercostal space, image quality improves drastically.

  • Shadow Reduction: In SLIVER07 tests, the proposed method significantly reduced the presence of non-soft-tissue (bone/air) artifacts compared to naive geometric planning.
  • Consistency in Imaging: Robotic trials showed that the planned trajectories maintained high visibility with minimal signal drops, whereas naive approaches suffered from massive shadowing when passing over simulated ribs.
  • Computational Feasibility: Optimizing for 10 target points took approximately 356 seconds. While not yet real-time, this performance is suitable for pre-operative planning workflows.

Limitations and Future Directions

Despite its success, the system faces several challenges. Its performance is highly dependent on the accuracy of the registration between pre-operative CT data and the patient's real-time position. Furthermore, the current model assumes a static body, ignoring tissue deformation caused by breathing or probe pressure.

Future work will focus on incorporating real-time image-based registration and respiratory motion compensation. Additionally, leveraging GPU acceleration or atlas-based priors could reduce computational costs, moving the system closer to interactive clinical use.

Summary updated

BibTeX
@article{gobl2017acoustic,
  author = {Rüdiger Göbl and Salvatore Virga and Julia Rackerseder and Benjamin Frisch and Nassir Navab and Christoph Hennersperger},
  title = {{Acoustic window planning for ultrasound acquisition}},
  journal = {Int. J. Comput. Assist. Radiol. Surg.},
  year = {2017},
  doi = {10.1007/s11548-017-1551-3},
}
Figure