kt-WiSE-MR: ECG-Free, Free-Breathing Cardiac CINE MRI: High Quality Reconstruction from Subsampled k-space using a Groupwise Motion Correction Paradigm
Cardiac CINE magnetic resonance imaging (MRI) has become a key medical image modality, since it can offer high and selective contrast and high spatial resolution, being accepted as reference for myocardial anatomical and functional imaging. It is known, however, that both cardiac and respiratory induced motions of the heart produce severe image artifacts that constitute major challenges in cardiac MRI. Breath-hold (BH) acquisitions and electrocardiogram (ECG) synchronizations either prospective or retrospectively are common solutions that are routinely used in clinical practice. For those cases in which breath cannot be held, some solutions exist, but they are rather inefficient as well as time consuming. On the other side, the compressive sensing (CS) theory has been successfully applied to MRI reconstruction to get good reconstructions from a small number of incoherent measurements in k-space. CS results have improved in dynamic images by adding a motion compensation term so that results are sparser in a time-space transformed domain. The CS theory has also been applied to find efficient imaging procedures in the free-breathing case. However, both for the breath hold and the free breathing cases, the proposals described so far compensate motion by using pairwise compensation approaches, either making use of a reference image or by using a sequential compensation. We have observed that a groupwise methodology for motion compensation is a better choice for the breath-hold case. Moreover, methods proposed so far for free-breathing scans do not account for both the cardiac and the respiratory motion in their reconstruction algorithms. In this proposal, we intend to develop a methodology that achieves high quality ECG-free and free-breathing MRI cine acquisition by making use of a groupwise-oriented motion estimation and compensation approach. We will also study the appropriateness of different k-space sampling procedures as well as different groupwise metrics. ECG-like information and breathing state will be estimated for low quality images which will be aligned after an optimization procedure, to end up with high quality images that will efficient use of the redundancy in k-space. Extensions of this approach to 3D real time imaging will also be studied. Finally, given that the problem at hand can be parallelized to a great extent, the developed algorithms will be implemented in graphical processing units (GPUs), which have shown to be able to significantly speed up off-line computations.