Fast magnetic resonance imaging by data sharing: Generalized series imaging and parallel imaging

This research is concerned with high-speed magnetic resonance imaging (MRI), a longstanding problem in imaging of time-varying objects. Due to the inherent physical and physiological constraints, the physics-based fast imaging methods are reaching a bottleneck at 30 ms/image. This research is focused on an emerging data-sharing approach to fast imaging, which can be used in combination with physics-based methods to achieve higher imaging speeds.;The first part of this research is concerned with motion correction in data-sharing imaging. Specifically, we present an analysis of the mutual information metric for rigid-body registration of two digital images and propose several preprocessing methods to improve its performance. This analysis is useful for improving the estimation accuracy of motion parameters.;The second part of this thesis addresses the problem of data acquisition and image reconstruction in data-sharing imaging. Specifically, two imaging techniques based on a generalized series (GS) model and a parallel data acquisition strategy are developed to improve imaging speed with a minimal concomitant loss of image quality. First, a new GS imaging method is presented which can effectively share information between different time frames of an image sequence. By optimizing the data acquisition scheme and the image reconstruction algorithm, we have successfully addressed the challenging problem of capturing transient dynamic features in the presence of nonrigid-body motions. Second, a new algorithm is proposed for image reconstruction from undersampled data acquired using multiple phased-array receiver coils. Specifically, wavelet denoising and motion compensation techniques are used to make the parallel imaging model more accurate. A support vector machine (SVM) regression method is developed to overcome the ill-conditioning of the model matrix. An adaptive scheme for choosing the regularization parameters is also proposed. Experimental results show that the proposed algorithm can reconstruct images with fewer artifacts and higher signal-to-noise ratios. This is particularly useful when data-sharing imaging is pushed to achieve very large acceleration factors.