UBC Theses and Dissertations
Feature-based registration of preoperative CT to intraoperative 3-D ultrasound in laparoscopic partial nephrectomy using a priori CT segmentation Bartlett, John George
Robotic laparoscopic partial nephrectomy is a state-of-the-art procedure for the excision of renal tumours. The challenges of this surgery along with the stereoscopic interface to the surgeon make it an ideal candidate for image guidance. We propose bringing pre-operative computed tomography data to the patient's coordinate system using three-dimensional intraoperative back ultrasound. Since computed tomography and ultrasound images represent like anatomical information quite differently, we perform a manual segmentation of the computed tomography before the operation and a semi-automatic segmentation of the ultrasound intra-operatively. The segmentation of the kidney boundary facilitates a feature-based registration strategy. Semi-automatic segmentation of kidney ultrasound images is difficult because the edges with large gradient values do not correspond to the capsule boundary seen in computed tomography. The desired edges are actually quite faint in ultrasound and poorly detected by common edge methods such as the Canny approach. After trying a number of approaches, the best results were obtained using a novel interacting multiple-model probabilistic data association filter to select edges from ultrasound images that were filtered for phase congruency. The manual segmentation of the prior is used to guide edge detection in ultrasound. Experiments on seven pre-operative patient datasets and one intra-operative patient dataset resulted in a mean volume error ratio of 0.80 +/- 0.13 from after registration to before registration. These results came after the implementation and evaluation of numerous other approaches, including radial edge filters, the covariance matrix adaptation evolution strategy, and a deformable approach using geodesic active contours. The main contribution of this work is a method for the registration of the pre-operative planning data from computerized tomography to the intraoperative ultrasound. For clinical use, this method requires some form of calibration with the laparoscopic camera and integration with surgical visualization tools. Through integration with emerging technologies, the approach presented here can one day augment the surgical field-of-view and guide the surgeon around important anatomical structures to the tissue that must be excised.
Item Citations and Data
Attribution-NonCommercial-NoDerivatives 4.0 International