This project was supervised by Dr. Gabor Fichtinger from the School of Computing at Queen’s University.
Image guidance is the current standard of care when performing neurosurgery. Through real-time tracking systems, surgeons are able to navigate around vasculature and identify internal structures. Skull registration methods differ depending on patient orientation. Although a noninvasive surface method is sufficient fro patients in a supine orientation, surgeries requiring a prone orientation use a more invasive standard of care. The current standard of care for the registration of prone position patients is to insert several titanium screws into the skull that can be used to hold a stereotactic frame or to act as individual markers. This method can achieve accuracy to under one millimeter between the skull surface and a surgical tool, and is necessary due to the lack of anatomical landmarks present and easily accessible on the posterior skull. This method of registration is not favoured by patients due to its invasive nature, and is also disfavoured by surgeons as this increases preoperative time and imaging required. We hypothesized that by incorporating an optically tracked ultrasound into the registration method, we could achieve a similar level of accuracy while being non-invasive. The use of a B-mode ultrasound would allow us to visualize bone surfaces that were previously inaccessible, and that the addition of these surfaces would be specific enough to break symmetry and ambiguity when implementing a surface registration. This was tested for its feasability using the following experimental setup: a computer running the open source 3D Slicer platform and the Plus Server Toolkit, a phantom skull, a Polaris Optical Tracker, and a Telemed MicrUs ultrasound, and a plastisol sheet to simulate skin. The registration method being proposed combined two registration methods; landmark-based and surface-based registration. The initial registration was performed using approximated anatomical landmarks around both mastoid processes, and the external occipital protuberance. A surface-based registration was then performed by selecting bony surface points in the ultrasound image, while scanning around the skull cap, the posterior base of the skull, and both mastoid processes. Using ultrasound provided more access to skull surfaces than processes utilizing a stylus or pointer. The target registration error (TRE) was measured by selecting a point within the skull and comparing the experimental registration to a ground truth registration. This phantom study, exploring the feasibility of tracked ultrasound, underwent five trials (n=5), for which three different error metrics were recorded. For each trial the initial registration TRE, mean surface model to point distance, and final TRE were recorded. The registration method proposed had an average TRE of 1.6+/-0.1mm, and average surface model to point distance of 0.6+/-0.1mm, and an average initial registration TRE of 2.5+/-0.5 mm. The use of tracked ultrasound could address the problems presented in prone patient registration. Throughout all trials performed, bone surface points were collected manually and non-invasively. The demonstrated registration method was subject to rotational error and as a result does not succeed in achieve similar levels of accuracy to that of the current standard of care. By exploring automated bone surface point placement and optimizations for the ultrasound scanning protocol we hope to further reduce the error and speed up the registration time.