UCLA

Magnetic resonance imaging (MRI) offers superior soft tissue contrasts compared to computed tomography (CT) and has played an increasing role in both radiation therapy planning and real-time radiation therapy guidance. The versatile contrast information offered by MRI allows more precise tumor target and surrounding organs-at-risk (OARs) delineation and helps achieve hypo-fractionated radiation treatment with relatively high radiation dose per fraction, referred to as stereotactic body radiotherapy (SBRT). However, the application of MRI for radiation therapy in the abdomen is particularly challenging. First, the abdomen is constantly moving due to respiratory motion, making 3D acquisition of a clear image rely on breath holdings. Second, the number of organs present in the abdomen is high. Multiple MR protocols are required for a diverse set of contrasts for better lesion and OARs delineation, leading to inter-scan image misalignments. Third, for real-time treatment monitoring, only 2D real-time images are available to clinical settings with limited temporal resolution.

The primary goal of the work in this dissertation is to address all the aforementioned limitations. This work will discuss how one can utilize the MR Multitasking framework to achieve: 1) volumetric (3D) respiratory motion-resolved images with multiple contrasts for MR-based radiation therapy planning (RTP); 2) 3D real-time image generation with high temporal resolution and with simultaneous multiple contrasts; 3) an integrated abdominal image platform that produces qualitative 4D multicontrast abdominal images with multi-parametric 3D quantitative maps. This dissertation will also discuss the developments made and new theories discovered in the MR reconstruction pipeline that improve reconstruction results and speed. These include ones that are specific to the MR Multitasking framework and those that apply broadly to generic MR reconstructions.

The specific aims can be broken down into the following:

Aim 1. The development of a volumetric, motion-resolved, multi-contrast abdominal MR imaging protocol for radiation therapy planning under a free-breathing 8-minute scan. In this aim, we achieve T1-weighted, T2-weighted, and proton density-weighted images with multiple (8–11) respiratory phases with low-rank tensor modeling from the MR Multitasking framework. With 22 healthy volunteers and 5 patients, we demonstrate that the proposed protocol achieves better imaging quality compared to available clinical protocols. Initial clinical testing with interobserver analysis demonstrated acceptable target delineation quality.

Aim 2. The development of a 3D real-time MR imaging protocol capable of simultaneous multiple contrast display and with high temporal resolution (up to 160 fps) for radiation treatment monitoring. We demonstrate that the proposed real-time image protocol generates consistent image quality compared to in silico and in vivo references.

Aim 3. The development of an integrated abdominal imaging framework that, in addition to the capabilities introduced in Aim 1, can also produce T1 and T2 3D parametric maps that potentially enable organ function-guided radiation therapy planning. In addition to these extended capabilities, this aim also focuses on clinical feasibility by introducing GPU-based accelerated image reconstruction algorithms such that image reconstruction can be done within 5 minutes and map fitting can be done in 0.5 seconds.

Aim 4. Discussions on novel theories and empirical evidence of the direct inversion formula for the multi-coil MRI forward operator under arbitrary trajectories. We discovered that the forward operation in question satisfies the low displacement rank property and has an explicit formula for its inversion. In this aim, we provide theoretical justification and empirical evidence.