Intraoperative 3D Ultrasonography for Image-Guided Neurosurgery

Abstract

Stereotactic neurosurgery has evolved dramatically in recent years from the original rigid frame-based systems to the current frameless image-guided systems, which allow greater flexibility while maintaining sufficient accuracy. As these systems continue to evolve, more applications are found, and image guidance has become a valuable tool in the treatment of a variety of neurological conditions. However, the current systems still contain one major limiting characteristic that prevents optimal use in image-guided neurosurgery. This limiting factor is that the system only works accurately under a rigid-body assumption, making the system inaccurate for soft tissue surgery, in which tissue deformations can occur. Recent research shows that intraoperative imaging may be a good solution for solving this problem. In this thesis intraoperative 3D ultrasound acquisition, as a solution for solving the problem of soft tissue movement during image-guided neurosurgery, is researched. In Chapter 2, the evaluation of an interactive multi-scale watershed segmentation method is presented. This segmentation method is used for segmenting tumours in preoperatively acquired MR brain images of twenty patients scheduled for neuronavigational procedures. The accuracy, repeatability, and efficiency of the watershed method are compared with the, commonly used, method of manual delineation, which is time consuming and prone to interoperator variability. The aim is to determine whether the watershed segmentation method can replace the manual delineation method. In Chapter 3, the set-up of the image-guided surgery system, expanded with an intraoperative ultrasound system, is discussed, as well as the acquisition and 3D reconstruction of the ultrasound scans. For twelve patients ultrasound datasets, acquired prior to and directly after opening the dura, are quantitatively compared to the preoperatively acquired MR data to estimate the translational component of brain shift at the first stages of surgery. In Chapter 4, three different spatial calibration methods for 3D freehand ultrasound are evaluated, since spatial calibration can have an important impact on the accuracy of 3D ultrasound reconstruction. The calibration methods are evaluated using 3D point localization, distance measurements and volume measurements. In Chapter 5, it is investigated to what extent rigid registration of preoperatively acquired MR data and the first intraoperatively acquired ultrasound data is feasible. Two approaches toward intensity-based rigid registration are investigated. The first approach is based on pre-processing with linear and non-linear diffusion filters to remove speckle noise in the ultrasound image, and subsequent rigid registration with the MR image. The second approach is based on registering interfaces between different tissue types of the MR and ultrasound images, where the interfaces are defined by the gradient magnitude of both images. In Chapter 6, the performance of two non-rigid registration methods, a B-spline based free-form deformation method and an optical flow based registration method, are compared for the registration of intraoperatively acquired 3D ultrasound data. For this study, 3D ultrasound volumes acquired prior to and directly after opening the dura are non-rigidly registered for eight patients, in order to estimate, and correct for, the deformations of the tumour and the surrounding tissue. In Chapter 7, the research results presented in this thesis are summarized. Furthermore, it is described how the combination of all these results may solve the problem of brain deformations during image-guided neurosurgery. The thesis is concluded with a discussion of research issues that still have to be addressed before the technique can be used clinically on a routine basis.