Arterial Spin Labelling (ASL) MRI suffers from a phenomenon known as the partial volume effect (PVE), which causes a degradation in accuracy of quantitative perfusion estimates. The effect is caused by inadequate spatial resolution of the imaging system. Resolution of the system is determined by point spread function (PSF) of the imaging process and the voxel grid on which the image is sampled. ASL voxels are comparatively large, which leads to tissue signal mixing within an individual voxel that results in an underestimation of grey matter (GM) and overestimation of white matter (WM) perfusion. PV correction of ASL images is not routinely applied. When PVC is applied, it usually takes the form of correcting for tissue fraction only, often by masking voxels with a partial volume fraction below a certain threshold. There are recent efforts to correct for tissue fraction effect through the use of linear regression or Bayesian inferencing using high resolution tissue posterior probability maps to estimate tissue concentration. This thesis reports an investigation into techniques for PVC of ASL images. An extension to the linear regression method is described, using a 3D kernel to reduce the inherent blurring of this method and preserve spatial detail. An investigation into the application of a Bayesian inferencing toolkit (BASIL) to single timepoint ASL data to estimate GM and WM perfusion in the absence of kinetic information is described. BASIL is found to rely heavily on the spatial prior for perfusion when the number of signal averages is less than three, and is outperformed by linear regression in terms of spatial smoothing until five or more averages are used. An existing method of creating partial volumes estimates from low resolution data is modified to use a voxelwise estimation for the longitudinal relaxation of GM, which improves segmentation estimates in the deep GM structures and improves GM perfusion estimates. An estimate for the width of the PSF for the 3D GRASE imaging sequence used in these studies is made and incorporated into a complete solution for PVC of ASL data, which deblurs the data through the process of deconvolution of the PSF, prior to a correction for the tissue fraction effect. This is found to elevate GM and reduce WM perfusion to a greater extent than correcting for tissue fraction alone, even in the case of a segmented acquisition. The new method for PVC is applied to two clinical cohorts; a Frontal Temporal Dementia and Posterior Cortical Atrophy groups. These two populations exhibit differential patterns of cortical atrophy and reduced tissue metabolism, which remains after PV correction.
|Qualification||Doctor of Philosophy|
|Award date||12 Aug 2006|
|Publication status||Unpublished - 2015|
- Magnetic Resonance Imaging