PROCESSING TECHNIQUES FOR DIFFUSION- WEIGHTED MAGNETIC RESONANCE IMAGES

Abstract

In the last few years, diffusion-weighted imaging has presented itself as a powerful and reliable contrast generating technique in magnetic resonance imaging (MRI). A number of past studies have shown that images of diffusing water, probing tissue structures at an unprecedented microscopic scale, provided insight on tissue pathophysiology complementary to that obtained through the most basic contrast generating mechanisms such as spin density, and T1 and T2 differences between tissues. Amongst its applications, the most successful has been brain ischemia detection in the early 1990s up to the most advanced application to date that being fiber tracking in the brain, which in combination with functional MRI would open a window onto the important issue of functional and structural neuronal connectivity.

In this thesis, we started from the well-established linear model characterizing the attenuating effect of diffusion on the MR signal as characterized in the addition of diffusion terms to the Bloch equations. A thorough review was done on the concepts of diffusion-weighted and diffusion-tensor imaging (DTI) with focus primarily on DTI post-processing issues. Quantitative diffusion anisotropy indices, as the fractional anisotropy (FA), the relative anisotropy (RA) and the volume ratio (VR), were utilized to characterize the diffusion anisotropy in the images. We verified the detrimental effect that background noise has on the estimation of the diffusion tensor components, eigenvalues and eigenvectors that in tum lead to erroneous measures of the diffusion anisotropy as measured by the anisotropy indices. Using Monte Carlo simulations, on true tissue data, we confirmed that below a certain signal-to-noise ratio (SNR), the estimated eigenvalues of the diffusion tensor diverge from their true values rendering isotropic tissue anisotropic and anisotropic tissue more anisotropic. Finally, with the proper adjustments to the noisy real images in the post-processing phase, color-coded maps depicting the nerve fiber spatial orientations in white matter of the brain were constructed based on experimental data obtained from normal human volunteers. The results of this thesis can be considered as a profound step in the path of addressing the problem of fiber tracking more accurately in the future taking into account the practical limitations outlined in this work.