Towards Automated Ultrasound Scanning Using Vision-Based Navigation: From Physician Skill Learning to Robotic Reproduction

Ultrasound scanning is widely employed for disease diagnosis and procedure guidance. However, conventional ultrasound depends on the sonographer’s manual operation, resulting in poor repeatability. Although robot-assisted ultrasound has been extensively studied, it has been a challenge in the field to further automate the examination and make the robotic system more practical for primary use. In this study, we propose a visual navigation method based on the skill learning of the sonographer. The key information about the probe’s movement, including position and angle, was recorded using vision-based methods on 14 cases based on a professional sonographer’s practice. Then, an automatic scanning method based on visual guidance, human-machine collaboration, and path planning was designed using a novel parallel mechanismbased ultrasound robot. The strategy of traversal scanning is a key consideration in facilitating efficient coverage of desired ranges and involves two primary steps: waypoint selection and waypoint planning. To compare the efficiency of different schemes proposed for these two respective stages, a simulation study and an actual running time test of the robot were conducted to assess the weighted coverage, energy, and time consumption. The results demonstrate that the proposed traversal strategy can effectively cover the statistical area formed by the physician’s scanning experience, and the new workflow is promising to be tested in-vivo in our future work.