Constrained and Spectral-Spatial RF Pulse Design for Magnetic Resonance Imaging

Abstract

Magnetic Resonance Imaging (MRI) provides a non-invasive glimpse inside the human body, generates excellent soft tissue contrast, uses non-ionizing radiation, and has become a critical tool in diagnosis of disease in medicine. Radio Frequency (RF) pulses are an integral component of MRI pulse sequences and can be tailored to particular applications. This dissertation explores the MRI physics, convex optimization problems, and experimental methodologies required for the design of tailored RF pulses

First, we introduce constrained RF pulse design, a process that incorporates meaningful, physical constraints, such as peak RF amplitude and integrated RF power, and enables efficient RF pulse design. With this process we explore simultaneous multislice (SMS) imaging, a method used to accelerate MRI and combat notoriously long acquisition times. Compared to an SMS pulse designed without constraints, our constrained pulses achieved lower magnitude normalized root mean squared error (NRMSE) for an equivalent RF pulse length, or alternatively, the same NRMSE for a shorter pulse length. Constrained RF pulse design forms a basis for the rest of the dissertation.

Second, we show that prewinding pulses, a special class of RF pulses, help reduce signal loss due to intravoxel dephasing generated by magnetic field inhomogeneities. We propose a spectral-spatial prewinding pulse that leverages a larger effective recovery bandwidth than equivalent, purely spectral pulses. In an in vivo experiment imaging the brain of a human volunteer, we designed spectral-spatial pulses with a complex NRMSE of 0.18, which is significantly improved from the complex NRMSE of 0.54 in the purely spectral pulse for the same experiment.

Finally, we consider a slab-selective prewinding pulse, that extends spectral and spectral-spatial prewinding pulses to a common 3D imaging method. Here we integrate optimal control optimization to further improve the slab-selective spectral pulse design and see an in vivo improvement of excitation NRMSE from 0.40 to 0.37. In the context of a steady-state sequence small-tip fast recovery (STFR), we also show a major reduction in mean residual transverse magnetization magnitude after the STFR “tip-up” recovery pulse from 0.18 to 0.02 when adding optimal control. This method has the potential to connect prewinding pulse design from the MRI physicist engineering workspace to a clinical application.

In summary, we show that constrained RF pulse design provides an efficient way of improving MRI in terms of acquisition speed (via multislice imaging) and image quality (via signal recovery).