Abstract:
Magnetic Resonance Imaging (MRI) is a powerful modality that offers noninvasive imaging of biological tissue without any harmful radiation, but is based on the spin properties of hydrogen and other nuclei. Besides providing pure anatomical information, different
techniques have been developed that make MRI a versatile tool for imaging tissue properties such as flow, diffusion or relaxation times. As nuclei experience different shielding of the external magnetic field dependent on their molecular environment, it is possible to
derive spectroscopic information with Magnetic Resonance (MR) as well. The appearance of the spectra depends on factors such as spin-spin coupling, thus information about molecular structure can also be inferred. This information is of great interest especially
when it comes to alterations of metabolism. Still, as all nuclei other than hydrogen have a relatively low gyromagnetic ratio and hydrogen in molecules other than water has low abundance in the human body, spectroscopic modalities require reduced spatiotemporal
resolution to compensate for the low intrinsic Signal to Noise Ratio (SNR). A relatively new technique to retrieve metabolic information indirectly is Chemical Exchange Saturation Transfer (CEST) MRI. With specific Radio Frequency (RF) preparation, hydrogen
protons that are not part of the bulk water pool are selectively labeled depending on their resonance frequency shift relative to bulk water. The labeled protons will subsequently exchange with protons in the bulk water pool and accumulate there. As the preparation
might be executed over a longer time course, the magnetization of the bulk water can be continuously altered. The resulting signal changes in the bulk water pool are orders of magnitude larger than the direct signal from off-resonant protons. While CEST is
advantageous compared to spectroscopic imaging in terms of SNR, it cannot provide the same spectral resolution. Moreover it is significantly slower than pure anatomic imaging due to repeated image acquisitions. Fast imaging is therefore crucial for applications
of CEST MRI. Additionally, the lower spectral resolution of CEST causes entangled signal origins that require sophisticated data evaluation and design of the RF preparation schedules. While CEST MRI has high sensitivity, its specificity is low and false
conclusions regarding the signal origin hamper the application as a diagnostic tool.
In the first step of this PhD project the signal origin in model solutions that mimic in vivo conditions is investigated in more detail. Experiments reveal that CEST MRI not only at spectrometers but even at whole body MR scanners is more sensitive than
commonly assumed. Also, a novel background signal origin in model solutions is confirmed, which may help to avoid false interpretations during the transition from model solutions to in vivo applications. In the second project, the feasibility of a fast imaging
method for CEST MRI at 3T is investigated. The optimized MR sequence provides decent spatiotemporal resolution with high reproducibility for CEST MRI at clinical field strength of 3 T. To facilitate the reproducibility of CEST MRI across different sites and MR scanners, in the third project, an open source sequence definition standard is proposed and implemented for both human and pre-clinical scanners. In the final project, the entire acquisition and reconstruction of MRI data is formulated as a model-free,
supervised learning problem. This top-down approach autonomously optimizes both data acquisition and evaluation on a real MR scanner. Exemplary mapping of creatine concentration via CEST is learned without any analytical model. The proposed method
also enables investigating whether MRI can be exploited as a tool for mapping arbitrary contrasts.
In the following sections, the main projects during the course of this PhD are briefly summarized.
Print-on-Demand