Radiation therapy has a significant role in the management of cancer patients. Computed tomography (CT) has been at the forefront of radiation therapy planning due to its widespread diagnostic use and its electron density information. Magnetic resonance (MR) imaging is another proven diagnostic modality, which can achieve superior soft tissue contrast and margin delineation, relative to CT. As such it has become a valuable tool for cancer diagnoses and staging. In this study, a centre-out radial acquisition using an off-resonance reception (co-RASOR) MR sequence, sensitive to magnetic field inhomogeneity, was applied to excite a broader frequency range of spins in the vicinity of metallic seeds. The resultant images display local hyperintensities around metallic markers. These images were then reconstructed with frequency offsets to rewind these hyperintensities to the geometric centre to obtain positive contrast. The contrast-to-noise ratio (CNR) was measured between a fiducial and its surrounding to compare Fourier and iterative reconstruction methods for undersampled co-RASOR. The motivation was to reduce the sequence acquisition time, while preserving sufficient CNR and resolution. For single slices, acquisition was 2.8 sec and multi-slice acquisition could acquire more than 50 slices in 73 sec, by reducing the acquired data by a factor of 8. This effectively encodes acquisition to 1.4 sec/slice. The noise present in undersampled images decreased significantly using iterative reconstruction methods, but a total variation based penalty better preserved the edges. Further extensions to the reconstruction method applied frequency-based filters that could isolate signals from different metallic compounds. The local hyperintensities rewind using opposite signed frequency offsets for diamagnetic and paramagnetic seeds. This allowed individual visualization of a low dose rate (LDR) brachytherapy seed and a gold fiducial marker. Phantom validation showed that each seed contains its maximal CNR in opposing frequency regions. The relative difference between global and local maxima in each frequency band ranged from 1.19 -- 3.73, and a single cut-off frequency was successfully applied for each acquisition plane. Image guidance systems rely on the position of these fiducial markers to compare daily setup images with CT and MR simulations. Phantom experiments and porcine tissue samples were used to assess the minimum separation of fiducials, geometric accuracy, and random errors associated with using the co-RASOR sequence. co-RASOR images were able to resolve fiducials separated by 0.5 - 1 cm, depending on image resolution. No systematic biases were observed by comparing co-RASOR to CT using rigid body registrations. The standard deviation of the systematic errors were \textless 0.5 mm between MR and CT registrations, and \textless 0.4 mm between MR scans without CT. These values are smaller than the current total systematic uncertainties, which should be limimed to <3 mm. The methods presented here can aid in understanding the trade-offs between acquisition speed and signal properties, differentiating cases where brachytherapy seeds are used in combination with fiducial markers for external beam boost, and aid in co-registration of multimodality imaging. / Thesis / Doctor of Philosophy (PhD)