Feasibility of Pointcloud-based Ultrasound-CT Registration towards Automated, Robot-Assisted Image-Guidance in Spine Surgery^*

Abstract

Image-guidance has been shown to improve spine surgical accuracy and patient outcome. Such image guidance can be achieved by registering low-cost, realtime, and radiation-free intraoperative ultrasound (US) with preoperative computed tomography (CT). Employing robotic US system (RUSS) allows automated acquisition of a wide 3D volume, facilitating efficient and accurate registration with CT. However, the registration between CT and robotic 3D US remains an open problem mainly due to (i) the lack of relevant testbed and dataset, and (ii) the modality discrepancy between CT and US. To address these challenges, we present a custom-built lumbar spine phantom for multimodal imaging with rigidly attached fiducials. The phantom is scanned by CT and in-house RUSS. A novel pointcloud-based registration pipeline is presented to register CT and robotic 3D US data of the spine phantom. Preliminary experiments demonstrate efficient (0.53 ± 0.02 seconds) registration of the lumbar spine with an accuracy of 3.57 mm in terms of fiducial registration error (FRE), which is robust to varying initial alignment of the pointclouds. These results show initial feasibility of adopting the proposed registration pipeline in robotic US guided spine surgery. Future work includes further evaluation of registration performance in ex vivo spine samples with denser US acquisitions to further improve registration accuracy.