| In magnetic resonance (MR) imaging, the quest for faster acquisition and higher spatial resolution has prompted the development of several innovative methods aimed at improving the efficiency of data acquisition. The underlying idea is that one should be able to acquire fewer data points if certain prior information is already known about the object being imaged. Thus, these prior-information-driven methods have the potential to push the speed and spatial resolution of MR imaging beyond current limits. However, many of the existing methods are vulnerable to inaccuracies in prior information, thereby limiting their practical utility in a clinical setting.; The research in this thesis is aimed at addressing this issue by studying the properties and vulnerabilities of existing methods, developing new methods to overcome existing limitations, and applying new methods to biomedically relevant applications. By studying the connection among existing methods, an integrative framework was developed that facilitated the design of new reconstruction methods in a systematic fashion. One method in particular, Broad-use Linear Acquisition Speed-up Technique (BLAST), showed considerable improvement over existing methods in terms of reconstruction accuracy and its broad applicability. In addition to introducing several new reconstruction methods, this research led to the development of a consensus reconstruction approach, which adaptively combined several reconstruction methods together to improve the overall robustness. The feasibility of these reconstruction methods and approaches was demonstrated with a wide range of applications, including reduced field-of-view dynamic imaging, T2 mapping, MR thermometry, and MR spectroscopic imaging. |
