Evaluation of atlas-based brain shift model for improved adaptation to intraoperative neurosurgical conditions

Abstract

Image guidance utilizes a rigid registration between the preoperative images and physical space of the operating room and it is now the standard of care in neurosurgical procedures. The fidelity of the image guidance system is known to be compromised by the extensively studied phenomenon of brain shift. A considerable body of work in literature has focused on solving this problem either through intraoperative imaging or by updating preoperative images with mathematical models. The factors that affect the magnitude and the direction of tissue deformation cannot be predicted to exact precision before the procedure and are often difficult to measure during the procedure. In previous literature, to account for this uncertainty, a statistical atlas-based method was used to capture the range of possible solutions. This work was validated using postoperative magnetic resonance (MR) data. Postoperative MR images are typically acquired after a lapse of 24 hours of surgery, during which period a shift recovery is known to occur. As a result, the postoperative measurements are typically smaller than what would be observed intraoperatively. Moreover the surgical environment is quite dynamic due to active tissue resection and retraction, which affect the observed displacements in the region of the craniotomy. The goal of this work was to systematically study the differences between the pre- and post-operative MR deformation and preoperative MR and intraoperative laser range scan deformation and devise strategies to better adapt the atlas-based model for intraoperative conditions. Strategies for improving the subsurface accuracy by accounting for the dural septa were studied. In order to make the atlas-based method feasible for intraoperative implementation, methods for automation of brain and dural septa segmentation were developed. Sensitivity analysis was performed to determine the impact of atlas resolution on accuracy. Lastly, preliminary studies for integration of intraoperative forces of retraction and resection into the atlas-based model were conducted. The results of these studies provide important conclusions to advance the goal of implementation of an efficient and cost-effective brain shift correction strategy for the neurosurgical image guidance.