Robot-assisted Motion Compensation based on Optical Flow in Ultrasound Images

The compensation of target motion is crucial to perform medical tasks, such as localized therapy delivery. For example, abdominal organs can periodically move up to few centimeters in a couple of seconds due to respiration. This sets the need for accurate and fast motion estimation strategies, as well as for their integration into control routines to manage the possible mismatches with respect to the treatment area. In this paper, we combine optical flow estimation in ultrasound imaging with robotic control to tackle this problem. The motion estimator based on optical flow (Farneback method) was implemented in a custom Matlab Simulink Real-Time, which was in communication with a ROS network to control the robot that handles the ultrasound probe. The motion estimation accuracy was evaluated by robotically moving the ultrasound probe along known trajectories that were concurrently estimated by the proposed method, reporting a maximum root mean square error of 1.18 mm. Also, motion compensation was tested on a phantom attached to a motorized slide to simulate breath-induced motion. Robotic compensation allowed for target stabilization in the imaging plane, reporting a median normalized cross-correlation across images of 0.95 during compensation. We believe that these results justify further investigation on the effective integration of US optical flow in the framework of robot-assisted targeted therapy.