Optimal path planning for autonomous robotic ultrasound scanning (reinforcement learning)

Abstract

To realize autonomous ultrasound (US) scans for different anatomies, it is important to quickly compute the optimal path for scanning from a generic image atlas. Due to the properties of US, it cannot pass the rigid object, e.g., bone. This makes it challenging to properly visualize the organs coved under the ribs, e.g., kidney and liver.

The objective of this project is to compute the optimal scanning direction for robotic ultrasound scanning. A continuous trajectory including both rotation and translation movement is generated to eliminate the shadow caused by the ribs and improve the overall image quality by placing the probe close to the objective anatomies.

The first task is to develop an effective user interface to annotate the area of the object in a 3D view. The second part is to compute the optimal path that can fully cover the selected area. It could be solved as an optimization problem with multiple objectives or using the reinforcement learning method.

Keywords: Autonomous ultrasound screening, path planning, reinforcement learning, optimization