| The core of the thesis is the idea of deformable-model-based information recovery from medical images, with the aim to reduce the impact of soft tissue deformation, noise, and image artifacts. Soft tissue deformation is the common denominator of many medical imaging problems. For this reason the main part of the thesis addresses the problem of soft tissue deformation recovery. Two volumetric deformable models based on soft tissue biomechanics are presented and used for deformation compensation. Experiments reported by other researchers as well as ones done by our group suggest that the complexity of soft tissue deformation renders deformable-model-based recovery very difficult. These findings lead to the concept of deformable model guidance. Rather than letting the model predict the soft tissue deformation based only on pre-deformation data, the approach we take is to guide the model by information available during the deformation. The models are guided by limited surface information with the goal to recover the deformation in the full volume. Another deformable-model-based information recovery is presented for the case of extraction of 2D structures embedded in 3D medical image volumes. The deformable model is based on the physical properties of the 2D structures, which significantly reduces the search space and enhances the quality of recovered information.; These methods are applied to image guided neurosurgery, where the top priority is the accuracy of surgical navigation systems. In particular, we describe intraoperative brain deformation compensation, with a stereo system used for model guidance. In addition, we show how deformable-model-based information recovery can be used to help localize implanted electrodes from postoperative 3D image volumes. Both applications are a part of a larger project aimed at unifying anatomical; functional, and electro-physiological data into one coordinate system. |
