Open Collections

Wiley, Neale

University of British Columbia

Magnetic resonance imaging (MRI) is a technique that allows non-invasive investigation of the properties of tissue through manipulation of the properties of the hyperfine splitting of the hydrogen atom in a magnetic field. Diffusion MRI is a variant of MRI that sensitises the imaging process to diffusion of water molecules in tissue through the use of strong gradient magnetic fields to dephase the signal of any protons undergoing diffusion, leading to a reduction in signal magnitude in that voxel. There is increasing use of iterative reconstruction algorithms of MRI data using the principles of compressed sensing (CS) that rely on the sparsity of information in the image domain to improve reconstructions of undersampled MRI data. Further refinements to the CS algorithm not only rely on the sparsity of single MRI images, but take advantage of low rank properties of some MRI data along a further dimension to permit even higher undersampling. Diffusion tensor imaging (DTI) acquires many sampled volumes of the same field of view encoded along different diffusion vectors. A reconstruction using low rank properties in the diffusion dimension could improve DTI image reconstructions. However, the random phase of diffusion imaging due to subject motion during the diffusion sensitising gradients of the acquisition destroys the low-rank nature of diffusion datasets. Phase correction of a randomly undersampled k-space is impossible, but the application of a phase correction factor to the sensitivity maps used in multicoil acquisitions can correct the phase error to restore low rank properties to the data. Alternatively, one can use a similar technique to perform a secondary reconstruction of diffusion tensor image data, as the magnitude images of a diffusion scan already discard all phase information, which includes the motion corruption. This secondary reconstruction efficiently denoises and simultaneously interpolates the images to a higher resolution by leveraging a projection into a low-rank subspace that shares information across different diffusion shots. This type of denoising performs on par with other contemporary denoising algorithms while reducing upsampling blurring through the use of zero-padded interpolation.

Text

2024-08-21

Attribution-NonCommercial 4.0 International

http://hdl.handle.net/2429/88985

Physics

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc/4.0/