Functional magnetic resonance imaging (fMRI) is a neuroimaging technique that provides an unparalleled ability to non-invasive study brain activity. Since its inception in the early 1990s, fMRI has become a dominant tool in studying neurological responses to tasks and stimuli and has been critical in our evolving understanding of brain mapping. These achievements in neuroscience would not be possible without critical breakthroughs in MRI theory and hardware advancements, which continue to increase the speed and resolution of fMRI acquisitions. This dissertation explores a highly signal efficient fMRI imaging strategy known as Oscillating Steady-State Imaging (OSSI) and presents specialized artifact compensation strategies for addressing the practical challenges of the OSSI method.

First, we develop analytical models and simulations of OSSI, which describe how the signal magnitude varies as a function of frequency. These simulations are then used to study how respiration-induced frequency changes cause artifactual signal fluctuations to a signal timecourse. Our simulations show that the severity of respiration artifacts changes with initial off-resonance. Furthermore, we show that respiration artifacts are primarily caused by transient signal effects rather than changes to steady-state magnitude. These findings inform the two correction strategies proposed in the remainder of the dissertation.

The second portion of this work describes "OSSCOR," a retrospective method to correct timecourse magnitude changes caused by temporally varying frequency. We show how the OSSI signal exhibits a frequency-time duality that can be used to reshape structured physiological noise into a low-rank matrix. We then use principal component analysis in a data-driven correction strategy to create nuisance regressors for subsequent fMRI analysis. We also describe a variation of our method where free induction decay (FID) signals can be used to create nuisance regressors, referred to as ``F-OSSCOR.'' Both OSSCOR and F-OSSCOR were found to significantly improve the functional sensitivity and signal stability compared to polynomial detrending alone, and OSSCOR was also found to significantly outperform a standard data-driven correction method.

Finally, we present a prospective correction method which utilizes FID measurements to estimate and correct for B0 changes in real-time. Prospective correction has the potential to outperform retrospective correction methods by directly reducing perturbations to steady-state magnetization during acquisition. We first present the results of a feasibility analysis where simulation was used to determine how scan parameters would affect correction performance. We then developed a prospective correction application using the RTHawk platform to perform data analysis and parameter adjustment in real-time. Our initial fMRI proof-of-concept shows that real-time correction can increase the number of activated voxels and improve overall image stability as measured by temporal SNR.