| Popis: |
Cardiac Magnetic Resonance (CMR) provides unique anatomical and functional information of the structures of the heart. To quantitatively assess cardiac functions, clinicians typically segment CMR slices in 2D. These can be used for 3D reconstructions allowing to better understand cardiovascular functions as well as predict different cardiac processes. These reconstructions can be used in the area of precision medicine or for shape analysis among populations, as well as serve as the input for biological simulations. However, CMR acquisition times usually necessitate slices to be acquired at different breath holds, which results in potential misalignment of the acquired slices. These artifacts cause any 3D model and subsequent evaluations to become distorted. Correcting for this spatial misalignment is required for accurate 3D reconstruction of the heart. In order to go from cardiac imaging to the geometric definition of patient-specific anatomies, many steps must be taken to achieve accurate geometric representations. Imaging, segmentation, and meshing challenges make the generation of anatomically realistic volumetric meshes a complex task. The main aim of this Thesis is to develop tools and methodologies to generate geometrically correct cardiac meshes in the presence of spatial misalignment between slices, to represent the anatomy of patients in a personalized manner. Three main novel contributions are presented. The first concerns the development of a complete framework for the correction of translational and rotational misalignments between CMR slices, using only the information from the images, allowing the method to be used directly in the clinics. The method aligns the respective slices together by using the line intensities at their intersections. We introduce the normalized cross correlation of local phase vectors as a similarity measure making the method applicable to different CMR protocols, and perform a complete validation using manually traced contours, including the estimation of contouring errors. The second concerns the development of a 3D surface reconstruction method capable of dealing with typical cardiac segmentations which are non-parallel, crosssectional, sparse, heterogeneous, and non-coincidental contours. An evaluation of the method is performed and a parameter study achieve to tune the method parameters to cardiac applications. Finally a third methodology is presented which includes both prior works and incorporates the individual steps of segmentation, alignment and mesh reconstruction in one method. This method provides a fully automatic way to go from images to 3D personalized models. |
