Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Robust processing of diffusion weighted image data

Parker, Greg 2014. Robust processing of diffusion weighted image data. PhD Thesis, Cardiff University.
Item availability restricted.

[thumbnail of 2014parkergdphd.pdf]
PDF - Accepted Post-Print Version
Download (24MB) | Preview
[thumbnail of ParkerGD.pdf] PDF - Supplemental Material
Restricted to Repository staff only

Download (74kB)


The work presented in this thesis comprises a proposed robust diffusion weighted magnetic resonance imaging (DW-MRI) pipeline, each chapter detailing a step designed to ultimately transform raw DW-MRI data into segmented bundles of coherent fibre ready for more complex analysis or manipulation. In addition to this pipeline we will also demonstrate, where appropriate, ways in which each step could be optimized for the maxillofacial region, setting the groundwork for a wider maxillofacial modelling project intended to aid surgical planning. Our contribution begins with RESDORE, an algorithm designed to automatically identify corrupt DW-MRI signal elements. While slower than the closest alternative, RESDORE is also far more robust to localised changes in SNR and pervasive image corruptions. The second step in the pipeline concerns the retrieval of accurate fibre orientation distribution functions (fODFs) from the DW-MRI signal. Chapter 4 comprises a simulation study exploring the application of spherical deconvolution methods to `generic' fibre; finding that the commonly used constrained spherical harmonic deconvolution (CSHD) is extremely sensitive to calibration but, if handled correctly, might be able to resolve muscle fODFs in vivo. Building upon this information, Chapter 5 conducts further simulations and in vivo image experimentation demonstrating that this is indeed the case, allowing us to demonstrate, for the first time, anatomically plausible reconstructions of several maxillofacial muscles. To complete the proposed pipeline, Chapter 6 then introduces a method for segmenting whole volume streamline tractographies into anatomically valid bundles. In addition to providing an accurate segmentation, this shape-based method does not require computationally expensive inter-streamline comparisons employed by other approaches, allowing the algorithm to scale linearly with respect to the number of streamlines within the dataset. This is not often true for comparison based methods which in the best case scale in higher linear time but more often by O(N2) complexity.

Item Type: Thesis (PhD)
Status: Unpublished
Schools: Computer Science & Informatics
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Date of First Compliant Deposit: 30 March 2016
Last Modified: 20 Oct 2023 12:55

Actions (repository staff only)

Edit Item Edit Item


Downloads per month over past year

View more statistics