Robotic Ultrasound Makes CBCT Alive

A team of researchers has published a paper titled "Robotic Ultrasound Makes CBCT Alive," introducing a novel AI-driven framework to solve a critical problem in image-guided surgery. Intraoperative Cone Beam Computed Tomography (CBCT) provides essential 3D anatomical snapshots but is static. During procedures, soft tissues deform due to respiration, probe pressure, and surgical manipulation, causing the pre-operative scan to become misaligned with reality. The team's solution leverages a robotic ultrasound probe as a real-time sensor. By establishing an accurate multimodal correspondence between the ultrasound stream and the CBCT volume, their system can infer how tissues are moving and deforming.

The core of their method is a lightweight neural network called the ultrasound correlation UNet (USCorUNet). Trained with optical flow-guided supervision, this network learns deformation-aware correlation representations from the ultrasound video feed. It outputs a dense deformation field, which is then spatially regularized and applied to the original CBCT scan. This process creates updated, deformation-consistent visualizations without requiring new CBCT scans, thereby eliminating repeated radiation exposure for the patient. The researchers validated their approach through experiments, demonstrating real-time, end-to-end CBCT slice updating and physically plausible deformation estimation. The source code has been made publicly available, paving the way for integration into robotic-assisted surgical systems.