Abstract
Reliable and accurate radiographic image registration that quantitatively compares the similarities or differences between the source and target images is critical for application of medical imaging in cancer diagnosis, treatment assessment and therapy planning. Before such quantitative comparisons can be done, it necessitates aligning the source and target images that typically acquired for different postures, or at different times of the patient (i.e. patient’s stature/organ geometry are often affected by the therapy and disease progression). A large number of medical image registration methods solely relying on image processing technologies have been successfully developed over the years. However, many of them have been proven effective for a single organ or a selected segment. Problems that involve large differences between the source and target images caused by motion of articulated bones/skeletons and deformations of soft body organs/tissues, such as whole-body Computed Tomography (CT), still remain a challenge. Therefore, biomechanical models, in which predicting the deformation filed of body organs/tissues is treated as a computational problem of solid mechanics, have been introduced in past decades.
The deformation field predicted by the patient-specific biomechanical models is used to warp source images to target configuration. This needs to compute position/location of all voxels within the discretised geometry from the nodal displacements and coordinates through interpolation within the element using element shape functions. The number of voxels in a typical whole-body CT image dataset is over 3107, therefore, an efficient numerical inverse isoparametric mapping algorithm to calculate the local coordinates of arbitrary points within the 8-noded hexahedral finite element has been developed.
Fully non-linear (i.e. accounting for both geometric and material non-linearity) finite element (FE) analysis has been historically used for computation of soft organ/tissue deformation. To facilitate rapid generation of patient-specific finite element models, time-consuming image segmentation that divides the problem domain into non-overlapping constituents with different material properties is abandoned in this thesis. Instead, the Fuzzy C-Means (FCM) algorithm is used for tissue classification and assigning material properties automatically at integration points of the computation grid direct form the CT images. Despite the 4-noded tetrahedral element is the most popular choice for computational biomechanics, the 8-noded under-integrated (with one Gauss point) hexahedral elements are used for spatial discretisation of the 3-D whole-body geometries as they do not exhibit volumetric locking. Furthermore, the 8-noded hexahedral meshes tend to offer better computational efficiency than tetrahedral meshes (for the same characteristic size of an element, less hexahedrons than tetrahedrons are needed to mesh a given volume).
Generation of hexahedral meshes for the 3-D whole-body geometries requires time-consuming manual correction even with the advanced software packages. To eliminate tedious mesh generation, the meshless (also known as mesh-free) method is used for predicting deformations of body organs/tissues. The meshless method utilizes an unstructured cloud of points for spatial discretization and, therefore, is much less demanding when building computational grids than the finite element discretization. The patient-specific meshless models are solved by a suite of meshless algorithms (Meshless Total Lagrangian Explicit Dynamics MTLED). The deformations of body organs/tissues computed using non-linear meshless models are verified against the results computed using the non-linear finite element models.
The registration accuracy of whole-body CT images using the patient-specific biomechanical models is qualitatively and quantitatively evaluated. For quantitative evaluation, I use edge-based Hausdorff distance (HD) to quantify the spatial differences between the registered (i.e. source images warped using the deformations predicted by biomechanical models) and target images. The results confirm that accurate whole-body CT image registration can be achieved using a patient-specific non-linear biomechanical model (i.e. finite element models and meshless models) constructed with fuzzy tissue classification for material property assignment.
The deformation field predicted by the patient-specific biomechanical models is used to warp source images to target configuration. This needs to compute position/location of all voxels within the discretised geometry from the nodal displacements and coordinates through interpolation within the element using element shape functions. The number of voxels in a typical whole-body CT image dataset is over 3107, therefore, an efficient numerical inverse isoparametric mapping algorithm to calculate the local coordinates of arbitrary points within the 8-noded hexahedral finite element has been developed.
Fully non-linear (i.e. accounting for both geometric and material non-linearity) finite element (FE) analysis has been historically used for computation of soft organ/tissue deformation. To facilitate rapid generation of patient-specific finite element models, time-consuming image segmentation that divides the problem domain into non-overlapping constituents with different material properties is abandoned in this thesis. Instead, the Fuzzy C-Means (FCM) algorithm is used for tissue classification and assigning material properties automatically at integration points of the computation grid direct form the CT images. Despite the 4-noded tetrahedral element is the most popular choice for computational biomechanics, the 8-noded under-integrated (with one Gauss point) hexahedral elements are used for spatial discretisation of the 3-D whole-body geometries as they do not exhibit volumetric locking. Furthermore, the 8-noded hexahedral meshes tend to offer better computational efficiency than tetrahedral meshes (for the same characteristic size of an element, less hexahedrons than tetrahedrons are needed to mesh a given volume).
Generation of hexahedral meshes for the 3-D whole-body geometries requires time-consuming manual correction even with the advanced software packages. To eliminate tedious mesh generation, the meshless (also known as mesh-free) method is used for predicting deformations of body organs/tissues. The meshless method utilizes an unstructured cloud of points for spatial discretization and, therefore, is much less demanding when building computational grids than the finite element discretization. The patient-specific meshless models are solved by a suite of meshless algorithms (Meshless Total Lagrangian Explicit Dynamics MTLED). The deformations of body organs/tissues computed using non-linear meshless models are verified against the results computed using the non-linear finite element models.
The registration accuracy of whole-body CT images using the patient-specific biomechanical models is qualitatively and quantitatively evaluated. For quantitative evaluation, I use edge-based Hausdorff distance (HD) to quantify the spatial differences between the registered (i.e. source images warped using the deformations predicted by biomechanical models) and target images. The results confirm that accurate whole-body CT image registration can be achieved using a patient-specific non-linear biomechanical model (i.e. finite element models and meshless models) constructed with fuzzy tissue classification for material property assignment.
Original language | English |
---|---|
Qualification | Doctor of Philosophy |
Supervisors/Advisors |
|
Publication status | Unpublished - 2015 |