Usage
  • 202 views
  • 498 downloads

Diffeomorphic Deformable Image Registration for Cardiac Magnetic Resonance Imaging (MRI)

  • Author / Creator
    Sheikhjafari, Ameneh
  • Deformable registration of cardiac magnetic resonance imaging (MRI) is one of the crucial tasks in medical image analysis. It aims to find the unique transformation between images taken from the same scene at different times, from different views, and by different imaging modalities such as MRI and CT. The main goal of this thesis is to develop automated deformable registration methods, particularly improving the accuracy and robustness of image registration while preserving the topology and invertibility of the deformation.

    Although deformation fields related to the point correspondence between a pair of images are high-dimensional, we propose a method that generates deformation fields from low-dimensional latent variables by minimizing a dissimilarity metric between a fixed and a warped moving image. This low-dimensional manifold formulation avoids the intractability associated with the high-dimensional search space that most other methods face during image registration.

    Moreover, we propose an end-to-end learning-free multi-resolution framework. This method eliminates the need for a dedicated training set while exploiting the capabilities of neural networks to achieve accurate deformation fields. Since it is capable to share the parameters through the architecture, it can be used for Groupwise registration as well as pairwise registration. We integrated Gaussian filters with the GMCNet to impose a smoothness constraint which relaxes the need for an explicit regularization term and its corresponding weight in the cost function.

    Additionally, we propose a learning-free diffeomorphic recursive framework, which models the changes in the deformation over multiple resolutions as opposed to the deformation itself. The final deformation is estimated by a solution to an ordinary differential equation (ODE). Thus, the resulting algorithm is recursive. Following this recursion, the moving image is warped successively and enables the final prediction to be decomposed into smaller displacements.

    Finally, we present an end-to-end unsupervised diffeomorphic framework based on moving mesh parameterization. This new parameterization of the deformation field has three significant advantages; firstly, it relaxes the need for an explicit regularization term and its corresponding weight in the cost function. Secondly, it guarantees diffeomorphism through explicit constraints applied to the transformation of the Jacobian determinant. Finally, it is suitable for cardiac data since it parameterizes the deformation using radial and rotational components.

  • Subjects / Keywords
  • Graduation date
    Fall 2022
  • Type of Item
    Thesis
  • Degree
    Doctor of Philosophy
  • DOI
    https://doi.org/10.7939/r3-zddc-6w88
  • License
    This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.