Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Medical Biophysics

Supervisor

Robert Z Stodilka

Abstract

The hybridization of magnetic resonance imaging (MRI) with positron emission tomography (PET) or single photon emission computed tomography (SPECT) enables the collection of an assortment of biological data in spatial and temporal register. However, both PET and SPECT are subject to photon attenuation, a process that degrades image quality and precludes quantification. To correct for the effects of attenuation, the spatial distribution of linear attenuation coefficients (μ-coefficients) within and about the patient must be available. Unfortunately, extracting μ-coefficients from MRI is non-trivial. In this thesis, I explore the problem of MRI-based attenuation correction (AC) in emission tomography.

In particular, I began by asking whether MRI-based AC would be more reliable in PET or in SPECT. To this end, I implemented an MRI-based AC algorithm relying on image segmentation and applied it to phantom and canine emission data. The subsequent analysis revealed that MRI-based AC performed better in SPECT than PET, which is interesting since AC is more challenging in SPECT than PET.

Given this result, I endeavoured to improve MRI-based AC in PET. One problem that required addressing was that the lungs yield very little signal in MRI, making it difficult to infer their μ-coefficients. By using a pulse sequence capable of visualizing lung parenchyma, I established a linear relationship between MRI signal and the lungs’ μ-coefficients. I showed that applying this mapping on a voxel-by-voxel basis improved quantification in PET reconstructions compared to conventional MRI-based AC techniques.

Finally, I envisaged that a framework for MRI-based AC methods would potentiate further improvements. Accordingly, I identified three ways an MRI can be converted to μ-coefficients: 1) segmentation, wherein the MRI is divided into tissue types and each is assigned an μ-coefficient, 2) registration, wherein a template of μ-coefficients is aligned with the MRI, and 3) mapping, wherein a function maps MRI voxels to μ-coefficients. I constructed an algorithm for each method and catalogued their strengths and weaknesses. I concluded that a combination of approaches is desirable for MRI-based AC. Specifically, segmentation is appropriate for air, fat, and water, mapping is appropriate for lung, and registration is appropriate for bone.

Download

Share

COinS