Strategies for Correcting Respiration-Induced B0 Variations in Oscillating Steady-State Functional MRI (OSS-fMRI)

Abstract

Blood-oxygen-level-dependent (BOLD) functional MRI (fMRI) has become an essential tool for non-invasively studying brain function, allowing scientists to measure brain activity by detecting changes in blood flow. However, its limited spatial resolution makes it challenging to capture fine-scale neural activity, such as depth-specific signals in the cortex or subtle variations in brain networks. High-resolution fMRI has the potential to reveal these intricate dynamics, but achieving such resolution necessitates a high thermal signal-to-noise ratio (SNR), which diminishes with smaller voxel sizes. Traditional gradient echo (GRE)-based fMRI techniques face difficulties with this trade-off due to T2* decay and thermal noise. One possible solution is the use of ultra-high-field (UHF) scanners; however, these scanners are costly, making them impractical for routine research or clinical use. Oscillating Steady-State Imaging (OSSI) is a novel fMRI acquisition method which leverages a large oscillating steady-state signal, achieving twice the SNR compared to GRE imaging under matched acquisition parameters. However, OSSI is sensitive to off-resonance, making it susceptible to physiological noise, particularly respiratory motion which can reduce the temporal SNR (tSNR) of the signal. To address these challenges, this dissertation presents multiple dynamic B0 correction methods to mitigate respiration- and drift-induced signal fluctuations in OSSI fMRI. First, we developed a two-stage deep learning-based retrospective correction method for OSSI called OSS-NET. In the first stage, the network predicts B0 changes directly from the OSSI signal. In the second stage, these predicted B0 changes are integrated with the OSSI signal to estimate the underlying BOLD response. Our findings showed that OSS-NET-generated B0 field maps exhibited strong spatial agreement with GRE-based double-echo field maps and achieved higher tSNR compared to the l2-norm combination method. These results establish OSS-NET as a potential post-processing technique for OSSI, providing a means to combine the OSSI signal and reduce respiration-induced signal fluctuations simultaneously. Temporal B0 changes can cause shifts in the OSSI steady state, leading to time-varying functional contrasts. These shifts can diminish or completely eliminate the functional contrast particularly for volumetric acquisitions, making it impossible to recover through post-processing alone. This limitation underscores the inadequacy of retrospective correction methods and highlights the need for prospective correction strategies. In the second part of this thesis, we introduced a zeroth-order real-time correction approach using free induction decay navigators (FIDNavs). This method compensates for the B0 changes as they occur, maintaining temporal stability and enhancing functional contrast. In vivo experiments demonstrated the efficacy of this approach, with real-time tracking yielding over a 100% increase in active voxels and more than a 50% improvement in mean tSNR. Finally, we introduced a proof-of-concept for first-order real-time correction using FIDNavs. This approach addresses the limitations of zeroth-order correction, which assumes uniform B0 shifts across the imaging volume and may be ineffective in handling spatially varying B0 field changes. These variations commonly occur in non-axial slices or 3D imaging scenarios. We explored two strategies for rapidly estimating first-order B0 changes using FIDNavs acquired from each coil. The first approach used a coupling matrix derived from a calibration scan, while the second leveraged the geometric centroids of coil sensitivity maps to spatially encode the FIDs. We implemented the calibration-based approach in a real-time feasibility study on a phantom. Our results demonstrated substantial improvements in tSNR maps and a marked reduction in signal fluctuations.