NSF Award
9631710
9631710 Kikinis In surgery it is necessary to recognize the boundary between healthy and diseased tissues. This is especially true for surgery in the brain where it is important to preserve as much healthy tissue as possible in the process of removing all of the diseased matter. Also, the boundary, which is irregularly shaped, must be identified in three dimensions. Surgeons typically examine medical images from magnetic resonance (MR) and computerized x-ray tomographic (CT) scanners to extract diagnostic information for surgical procedures. Currently surgeons are forced to imagine the three-dimensional situation for presurgery planning, something the human brain is not well suited to performing in a precise and reproducible manner. In order to provide a complete engineering solution to this problem using high performance computers, the following steps have to be performed: (1) acquisition of data in a suitable geometric raster; (2) generation of three-dimensional (3D) reconstructions; (3) registration of the 3D reconstructions to the patient on the operating room table; and, (4) labeling of relevant tissue structures in the data (segmentation). The technology required for achieving steps (1) - (3) is being developed outside of medicine. However, this is not true for segmentation, which is the least advanced of the four steps; this investigation seeks to develop core segmentation algorithms for solving the problem. Work during this investigation is to concentrate on template- based segmentation and operator-driven segmentation in neurosurgical applications. In template-based segmentation the goal is to improve the ability to establish correspondence structures automatically in cases with moderate pathology. The operator-based segmentation portion of the investigation would develop a working platform for the integration of a volume editor and a force feedback device needed for algorithmically more complex projects such as segmentation in three-dimension s. ***
