| dc.description.abstract |
Magnetic Resonance Imaging (MRI) is a fascinating example of how basic research in physics, studying nuclear spin and magnetism, has evolved into a highly successful technology that enables non-invasive medical imaging. As such, it provides insight into both anatomy and physiology, and aids in the diagnosis of a wide range of pathologies. In addition, functional MRI is a central tool for addressing neuroscientific questions that seek, for example, a better understanding of the human brain. The basic principle of Magnetic Resonance (MR) technology is that nuclear spins can be manipulated and probed by electromagnetic fields. When done properly, the interactions between the spins, as well as with their microscopic electronic environment and the external electromagnetic fields, allow a large variety of different tissue properties to be accessed and imaged without the use of harmful ionizing radiation. This makes MRI an inherently multimodal imaging technique, yielding information about, for example, proton density, relaxation times, diffusion and perfusion, blood oxygenation, molecular structure, temperature, magnetic susceptibility, electrical and even mechanical properties. Despite the versatility and advantages of this technology, probably the most striking limitation of MRI lies in the comparatively long acquisition times required to obtain the necessary information, especially to achieve sufficient Signal to Noise Ratio (SNR).
To address this problem, the aforementioned manipulation and probing of nuclear magnetization has been subject to extensive research and engineering efforts. In particular, approaches to optimize and speed up MRI can be broadly classified into three categories: 1) faster imaging sequences such as Fast Low Angle Shot (FLASH) or Echo Planar Imaging (EPI), 2) hardware-based improvements such as better gradient coils and so-called Parallel Imaging (PI), and 3) reconstruction methods based on tailored signal processing. This thesis covers two distinct topics in the context of optimizing MRI acquisition and reconstruction.
As the first topic, a novel concept for improving PI at Ultra-high Field (UHF) MRI is presented. Among all technical developments to speed up MRI acquisition, PI is one of the most successful and widely applied techniques. It relies on the additional localization information provided by the spatial sensitivity profiles of multiple Radio Frequency (RF) receive coils used to detect nuclear magnetization. However, the maximum achievable PI acceleration factor that still provides acceptable image quality is fundamentally limited by the number and spatial independence of the coil sensitivity profiles. The proposed novel concept to improve PI is based on electronically modulated time-varying receive sensitivities enabled by custom-built reconfigurable RF coils. It is investigated and demonstrated how these can be realized experimentally and used advantageously in acquisition and reconstruction. This work thus can be seen as falling into categories 2 and 3 above.
As a second topic, Chemical Exchange Saturation Transfer (CEST) is considered as one of the MRI contrast mechanisms listed above that promises to provide insights into the molecular microenvironment and to detect low concentration metabolites. Due to the required sequence structure of repeated RF-prepared readouts, CEST MRI suffers from long scan times. In addition, the extraction of the desired contrast information usually requires computationally complex processing steps. In this context, a data-driven linear projection method for CEST parameter estimation from the acquired raw data is proposed, which allows simple and fast contrast generation and a potential reduction of acquisition times by providing insight into which parts of the CEST data contain relevant information and which parts could be omitted. This project can be seen as belonging to categories 1 and 3 of the MRI optimization methods mentioned above.
Finally, a novel and experimental method of optimizing MRI contrast generation is proposed, where both acquisition and contrast mapping are treated as a joint numerical optimization problem. Departing from the conventional way of basing such optimizations on theoretical models and numerical simulations, a model-free framework is implemented here that optimizes both acquisition parameters and contrast extraction schemes purely based on automated exploratory acquisitions running on a real MRI scanner. The method can also be seen as belonging to categories 1 and 3. A proof-of-principle demonstration of this framework is given in the context of CEST MRI. The approach may be particularly useful in situations where a theoretical description of the targeted problem is not available, such as hardware system imperfections. This in turn suggests a possible link to category 2 of the MRI optimization methods mentioned above. Overall, all of these projects highlight the potential that lies in synergistically considering all aspects of MRI related to hardware, acquisition, and reconstruction in order to strive for optimization of acquisition times and information retrieval. |
en |
