Towards Fully Autonomous Ultrasound Scanning Robot With Imitation Learning Based on Clinical Protocols

Ultrasound scanning plays an important role in modern clinical examinations. Thanks to its small footprint, low cost, and popularity, it has been widely used in annual physical examinations and many other diagnosis and intervention procedures. However, the scanning results depend heavily on the clinician operators' skills, causing inconsistency and even false detection. A fully autonomous ultrasound scanning robot can be a promising solution, enabling consistent and accurate operation. In this letter, we propose a procedure-specified imitation learning framework based on clinical protocols to implement autonomous scanning tasks. Based on the targeted carotid artery examination procedure, we design a feature-based visual servo controller for both in-plane and out-of-plane feature tasks. A one-step-exploring (OSE) method is also designed to improve the robustness and transition smoothness in between tasks. The proposed methods are tested and validated with experiments on both medical phantom and human subjects. The results show that the proposed appraoch is able to significantly improve the success rate of procedure completion from 22.2% to 84.6% for the out-of-plane feature-based operation.