Improved MRI Relaxometry through Statistical Signal Processing

TitleImproved MRI Relaxometry through Statistical Signal Processing
Publication TypeThesis
Year of Publication2018
AuthorsG. Ramos-Llordén
AdvisorJ. Sijbers, and A J. den Dekker
Academic DepartmentDept. of Physics
Degree Doctor of Science
Date Published02/2018
UniversityUniversity of Antwerp
Thesis TypePhD thesis

Magnetic Resonance Imaging (MRI) relaxometry is a quantitative MRI modality that deals with the estimation of the spin-lattice, T1, and the spin-spin, T2, relaxation times. Both relaxation times are fundamental parameters that describe the spin dynamics within a tissue during the relaxation process of the Nuclear Magnetic Resonance (NMR) phenomenon. During the last decades, spatial T1 and T2 maps have been analyzed to study and monitor the states of a multitude of human diseases. Those studies have shown that MRI relaxometry holds the promise of generating robust, objective image-based biomarkers for central nervous system pathologies, cardiovascular diseases and beyond. Unfortunately, quantitative biomarkers derived from MRI relaxometry are not yet sufficiently specific, sensitive, and robust to be routinely used in clinical practice. On top of that, high-resolution relaxation maps demand a clinically unfeasible long scanning time.

This PhD thesis tries to reduce the obstacles that preclude MRI relaxometry from being a fast, accurate, and precise quantitative MRI modality for clinical use by improving the way MR relaxometry data are acquired, processed and analyzed. In particular, by adopting a statistical signal processing approach, three contributions that address common problems of the field are given. Firstly, we present a unified relaxometry-based processing framework where the T1 map estimation and motion correction are accounted for in a synergistic manner, being both the T1 map and the motion parameters simultaneously estimated with a Maximum-Likelihood estimator. It is demonstrated that substantially more accurate T1 maps are obtained with our proposed integrated approach in comparison to the yet typical but suboptimal two-step approach: T1 model fitting after image registration. Secondly, we developed a fast, robust, T1 estimator for Variable Flip Angle (VFA) T1 mapping that can provide statistically optimal T1 estimates with unprecedentedly short computation time, thereby enabling optimal, real-time, VFA T1 mapping. Finally, we were able to reduce the long overall scanning time of MRI relaxometry studies using our novel k-space reconstruction technique that permits the reconstruction of individual MR images with less number of samples than it is commonly required.