| Statistical shape models have considerably improved the robustness and accuracy of image analysis methods by incorporating a-priori shape knowledge learned from a set of examples. These models have primarily been applied to segmentation of densely sampled 3D image data, where they serve to restrict the search space to the trained shape variations. In this proposal, we aim to generalize 3D statistical shape models for shape extraction from sparse and/or unorganized datasets. Such sparse data acquisitions frequently occur in biomedical applications. In particular, we will focus on the use of 3D statistical shape and deformation models for providing 3D anatomical guidance in minimally invasive, image-guided interventions in which considerable intra-operative deformations occur. During these interventions, intra-operative imaging is often limited and we aim to relate detailed 3D information as provided by pre-operative imaging or 3D anatomical shape models to the intra-operative situation. The proposed research contains three main contributions. First, methods will be developed that utilize prior information on object shape and deformation to fit 3D anatomical models to unorganized and sparsely sampled imaging data. Second a detailed investigation is conducted into the accuracy with which 3D models can be fitted, given a prior shape model and different sparseness scenarios. This can be used to give realistic estimates of the accuracy that can be achieved when fitting models to sparse imaging data, or to design image sampling strategies to achieve a desired accuracy. Third, the method will be evaluated on two applications; providing 3D image guidance during liver and cardiac ablation. |
