Institute of Computer Science The Hebrew University of Jerusalem, Israel

Master of Science Thesis, 1998.

Fluoroscopy-based orthopedic procedures crucially depend on the ability of the surgeon to mentally recreate the spatio-temporal intraoperative situation from uncorrelated, two-dimensional X-ray images. Significant skill, time, and frequent use of the fluoroscope are required, leading to positioning errors and complications in a non-negligible number of cases, and to significant cumulative radiation exposure of the surgeon. To overcome these problems, our team at the Institute of Computer Science, the Hebrew University, Jerusalem, is currently developing a computer-integrated orthopedic system, called \mbox{\sc Fracas}, for closed medullary nailing of long bone fractures. We propose to replace the extensive use of fluoroscopic images with a virtual reality display of spatial bone fragment models created from preoperative CT and tracked intraoperatively in real time. {\sc Fracas} uses fluoroscopic images to establish a common reference frame between three-dimensional bone fragment models to the intraoperative situation and to verify that the correspondence is maintained.

This thesis describes the fluoroscopic image processing and anatomy-based 2D/3D registration techniques of {\sc Fracas}. Fluoroscopic image processing consists of image dewarping, camera calibration, and bone contour extraction. Anatomy-based 2D/3D registration consists of finding the rigid transformations that best matches the surface of a bone with its projected contour in the fluoroscopic X-ray images. Our approach focuses on bone imaging and emphasizes integration, full automation, simplicity, robustness, and practicality. We describe the experimental setup and report results quantifying the accuracy of our methods. For fluoroscopic image processing, we show that after dewarping and calibration, submillimetric spatial positioning accuracy is achievable with standard equipment. We present a new bone contour segmentation algorithm based on robust image region statistics computation which yields good results on clinical images. For registration, we present an extension of the iterative closest point algorithm for 2D/3D registration, and preliminary results with synthetic images of actual bone fragments.

Keywords: X-ray image processing, image correction and dewarping, camera calibration, conotur extraction, computer-assisted orthopaedics