Robotic ultrasound has the potential to make musculoskeletal imaging more repeatable, standardized, and less operator-dependent. In our group, we have developed a Franka-based robotic ultrasound platform for human arm scanning and 3D reconstruction. We also have ultrasound image data from 15 human subjects.
This project aims to investigate how reinforcement learning can be used to improve robotic ultrasound scanning. Instead of following a fixed scanning trajectory and fixed probe settings, the robot should learn how to locally adjust the ultrasound probe during scanning. In particular, the project will focus on adapting probe contact force and probe orientation based on ultrasound image feedback and force sensing.
The goal is to improve ultrasound image quality, bone and muscle visibility, and the reliability of 3D reconstruction. The student will work with an existing robotic ultrasound system and explore learning-based strategies for adaptive probe control.
The project is suitable for students interested in robotics, reinforcement learning, medical imaging, computer vision, and control.
Keywords: robotic ultrasound, reinforcement learning, force control, musculoskeletal imaging, 3D reconstruction, Franka robot
Contact:
Dezhi Sun
dezhi.sun@utwente.nl