Constrained MR Image Reconstruction of Undersampled Data from Multiple Coils

While Magnetic Resonance Imaging (MRI) is recognized to be the leading diagnostic imaging modality for numerous diseases, it is often limited by the long examination times. The main reason for this is that for MRI encoding only a limited number of data points can be encoded from one MR signal (FID or echo) and the measurement has to be repeated for many times. Increasing imaging speed has always been a major research area ever since the first experiments in the 1970s, and improvements in this field have contributed significantly to pave the way for clinical application of MRI. One way to accelerate data acquisition is to reduce the number of data points that are used to reconstruct an image with a defined resolution. However, as this approach violates the Nyquist-Shannon sampling theorem, aliasing artifacts are introduced in the reconstructions which have to be eliminated. This can be done by incorporating additional a-priori knowledge during the image reconstruction process. Examples are the use of multiple channels for signal reception (parallel imaging) or the application of dedicated regularization terms which penalize aliasing artifacts in the reconstructed images. This work demonstrates the use of newly developed image reconstruction methods which allow to reconstruct images from highly accelerated data. Numerous examples based on simulations, phantom- and in-vivo-measurements are presented. It is also shown that parallelized implementations of these approaches can be achieved on dedicated graphics hardware which leads to a fundamental reduction of computation time