| Description |
Magnetic Resonance Imaging (MRI) is one of the most important medical imaging technologies in use today. Unlike other imaging tools, such as X-ray imaging or computed tomography (CT), MRI is noninvasive and without ionizing radiation. A major limitation of MRI, however, is its relatively low imaging speed and low spatial-temporal resolution, as in the case of dynamic contrast enhanced magnetic resonance imaging (DCE-MRI). These hinder the clinical use of MRI. In this thesis, we aim to develop novel signal processing techniques to improve the imaging quality and reduce the imaging time of MRI. This thesis consists of two parts, corresponding to our work on parallel MRI and dynamic MRI, respectively. In the first part, we address an important problem in parallel MRI that the coil sensitivities functions are not known exactly and the estimation error often leads to artifacts in the reconstructed image. First, we develop a new framework based on multichannel blind deconvolution (MBD) to jointly estimate the image and the sensitivity functions. For fully sampled MRI, the proposed approach yields more uniform image reconstructions than that of the sum-of-squares (SOS) and other existing methods. Second, we extend this framework to undersampled parallel MRI and develop a new algorithm, termed Sparse BLIP, for blind iterative parallel image reconstruction using compressed sensing (CS). Sparse BLIP reconstructs both the sensitivity functions and the image simultaneously from the undersampled data, while enforcing the sparseness constraint in the image and sensitivities. Superior image constructions can be obtained by Sparse BLIP when compared to other state-of-the-art methods. In the second part of the thesis, we study highly accelerated DCE-MRI and provide a comparative study of the temporal constraint reconstruction (TCR) versus model-based reconstruction. We find that, at high reduction factors, the choice of baseline image greatly affects the convergence of TCR and the improved TCR algorithm with the proposed baseline initialization can achieve good performance without much loss of temporal or spatial resolution for a high reduction factor of 30. The model-based approach, on the other hand, performs inferior to TCR with even the best phase initialization. |
