MRI Excitation Pulse Design and Image Reconstruction for Accelerated Neuroimaging

Abstract

Excitation pulse design and image reconstruction are two important topics in MR research for enabling faster imaging. On the pulse design side, selective excitations that confine signals to be within a small region-of-interest (ROI) instead of the full imaging field-of-view (FOV) can be used to reduce sampling density in the k-space, which is a direct outcome of the change in the underlying Nyquist sampling rate. On the reconstruction side, besides improving imaging algorithms’ ability to restore images from less data, another objective is to reduce the reconstruction time, particularly for dynamic imaging applications. This dissertation focuses on these two perspectives: Chapter II is devoted to the excitation pulse design. Specifically, we exploit auto-differentiation frameworks that automatically apply the chain rule on complicated computations. We derived and developed a computationally efficient Bloch-simulator and its explicit Bloch simulation Jacobian operations using such frameworks. This simulator can yield numerical derivatives with respect to pulse RF and gradient waveforms given arbitrary sub-differentiable excitation objective functions. The method does not rely on the small-tip approximation, and is accurate as long as the Bloch simulation can correctly model the spin movements due to the excitation pulses. In particular, we successfully applied this pulse design approach for jointly designing RF and gradient waveforms for 3D spatially tailored large-tip excitation objectives. The auto-differentiable pulse design method can yield superior 3D spatially tailored excitation profiles that are useful for inner volume (IV) imaging, where one attempts to image a volumetric ROI at high spatiotemporal resolution without aliasing from signals outside the IV (i.e., outer volume). In Chapter III, we propose and develop a novel steady-state IV imaging strategy which suppresses aliasing by saturating the outer volume (OV) magnetizations via a 3D tailored OV excitation pulse that is followed by a signal crusher gradient. This saturation based strategy can substantially suppress the unwanted aliasing for common steady-state imaging sequences. By eliminating the outer volume signals, one can configure acquisitions for a reduced FOV to shorten the scanning time and increase spatiotemporal resolution for applications such as dynamic imaging. In dynamic imaging (e.g., fMRI), where a time series is to be reconstructed, non-iterative reconstruction algorithms may offer savings in overall reconstruction time. Chapter IV focuses on non-iterative image reconstruction, specifically, extending the GRAPPA algorithm to general non-Cartesian acquisitions. We analyzed the formalism of conventional GRAPPA reconstruction coefficients, generalized it to non-Cartesian scenarios by using properties of the Fourier transform, and obtained an efficient non-Cartesian GRAPPA algorithm. The algorithm attains reconstruction quality that can rival classical iterative imaging methods such as conjugate gradient SENSE and SPIRiT. In summary, this dissertation has proposed and developed multiple methods for accelerating MR imaging, from pulse design to reconstruction. While devoted to neuroimaging, the proposed methods are general and should also be useful for other applications.