Robot-Assisted US-Based Vertebral Segmentation for Pedicle Screw Trajectory Identification

A standardized robot-assisted US-Scan of the vertebra L1 to the S1 is executed. The two probe trajectories are generated automatically by an optical camera or RedGreenBlue-Depth (RGB-D) camera. Then the robot will scan at two mm/s, automatically following the trajectory. A 6 degrees of freedom (DoF) force sensor (Nano25, Ati Industrial Automation Inc.) was assembled at the US probe. During the scanning, it measures the interaction force and torque between the US probe and volunteers' skin to ensure volunteer safety. The US images and corresponding robot poses are recorded for the reconstruction. A previously developed algorithm is used to obtain a preliminary 3D reconstruction that will be used to adapt the scanning pattern in case of incomplete scanning.

A handheld US-Scan of the vertebra L1 to S1 is executed. An optical marker for an Atracsys Camera is mounted to the US probe to track its position. Simultaneously, the optical tracking allows for standardization of the US-Scan through navigation. The acquired US images and corresponding poses are used to reconstruct the lumbar spines and compared to the ground truth (CT scan). Handheld US scanning will be performed at around two mm/s in two scanning trajectories. The US probe orientation will be adjusted during the scanning by comparing the optical marker orientation with the initial probe orientation. This combination leads to different scanning patterns. For each pattern, the scanning procedure will be repeated three times. At least once per volunteer, each handheld scanning pattern will be performed with a higher manual speed to assess the reconstruction quality with faster scanning. A graphical user interface (GUI) will synchronize and store the data during the scanning.

A ULD CT scan (CT, Siemens Naeotom Alpha: A-207883-62) of the lumbar spine (L1-S1) is performed at the Swiss Center for Musculoskeletal Imaging (SCMI) at Balgrist Campus, Zurich. The total duration estimate for the CT examination is 30 minutes, whereas the scan takes 15 minutes. The volunteers lie on their abdomen during this procedure to simulate the spine's position during the subsequent US examination. The vertebrae (L1-S1) in the CT Scans are manually segmented with global thresholding and the region growing tool in a standard segmentation software (Materialise Mimics, Leuven, Belgium). The 3D surface models are then exported as Standard Triangle Language (STL) files.