Identification of Anatomical Landmarks for Registration of CT and Ultrasound Images in Computer-Assisted Shoulder Arthroscopy

This paper presents a phantom study that was conducted for an ultrasound-guided shoulder arthroscopy navigation system. The navigation system uses a surface model generated from pre-operative computed tomography images, which has to be registered to the patient during the procedure. The goal of this study was to determine the optimal regions on the scapula bone of the shoulder to achieve an acceptable registration. Experiments were performed to examine the robustness and suitability of these optimal regions by testing the sensitivity to variations in the initial alignment for two different registration algorithms, namely iterative closest point and sequential least squares estimation technique. The fiducial registration error was analyzed and compared for all experiments. Regions spread over the entire scapula result in significantly smaller registration error (p<0.001) than regions, concentrated around the shoulder joint and thus accessible during the shoulder arthroscopy. However, the results also showed that the registration is still acceptable for the image- guided navigation system when these accessible landmarks are used