Loading...

Analysis and Processing of High Angular Resolution Diffusion Images

Afzali Deligani, Maryam | 2016

650 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 48408 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Fatemizadeh, Emadeddin; Soltanian Zadeh, Hamid
  7. Abstract:
  8. Diffusion Weighted Imaging (DWI) is a non-invasive method for investigating the brain white matter. Assuming the Gaussian model for diffusion process, diffusion tensor is constructed and Diffusion Tensor Images (DTI) are obtained. White matter is constructed from fiber bundles which have crossing in most of the regions. In the crossing regions, the Gaussian model cannot work. In this situation, DTI cannot reconstruct the fiber structures correctly. Therefore, High Angular Resolution Diffusion Imaging (HARDI) was proposed to solve this problem. Q-ball imaging is a new technique for HARDI reconstruction which is useful for estimating diffusion Orientation Distribution Function (ODF). ODF is a spherical function that its maximums are aligned to the fiber direction in each voxel. Considering that HARDI data contains a large amount of information about brain microstructure- like crossing fibers- they can be used for fiber tracking or fiber segmentation in the brain. Naturally, this information can be utilized for exact registration of brain structures. registration of HARDI data like other kind of images has a great importance in medicine. some applications of image registration are; group analysis in orther to find abnormal regions in a special kind of disease compared to the normal brain, or atlas construction purposes. Furthermore, interpolation of these kind of images is very important, because we should preserve the directional information of these images after interpolation. Our goal in this thesis is to propose some models for diffusion process and using the proposed models for interpolation and registration of diffusion weighted images. We propose a multi-tensor model in the ODF field. Our proposed method utilizes the geometrical shape of ODF to find the tensors and it does not need to preassume a value for eigenvalues of tensors. We use this model for interpolation and registration of DWI. Also, we propose a sparse representation for DWI. We utilize the diffusion signal and construct the sparse coefficients using K-SVD algorithm. This representation is used for interpolation and registration of DWI in the diffusion signal field. In this way, we can fit any model to the interpolated and registered images of this method. Finally, we compare our proposed methods on synthetic and real images with other state-of-the-art methods. Our framework shows better performance based on the angular error between ODFs (0.2 degrees improvement for interpolated images and 0.5 degrees improvement for registered images)
  9. Keywords:
  10. Sparse Representation ; Interpolation ; Image Processing ; Image Registration ; Diffusion Weighted Imaging (DWI) ; Multi-Tensor Model

 Digital Object List

 Bookmark

No TOC