Patch-based, iteratively-reweighted compressive recovery for reconstruction of highly accelerated exercise stress cardiac cine

Background Real-time exercise stress cardiac magnetic resonance imaging is challenging due to exaggerated breathing motion and high heart rates; improvements in image reconstruction may help improve the reliability and diagnostic accuracy of this difficult imaging application. Cardiac images possess a rich structure that can be exploited to aid image reconstruction by enforcing sparsity in an appropriate transformed domain, e.g., in the undecimated wavelet transform (UWT) domain). When using UWT or its decimated counterpart, standard techniques achieve L1 regularization through the use of a single weighting rule (regularization strength) across different sub-bands [1]. Since the level of sparsity varies across sub-bands, it has been shown that iteratively adapting the individual regularization strength for each sub-band can improve the recovery process [2]. However, levels of sparsity may vary significantly not only between but also within sub-bands, and taking advantage of this finer-grained variation may further improve reconstruction results, especially in scenarios where severe motion is present. In this work, we demonstrate that the use of a patch-based iteratively reweighted approach, in which regularization strength is adapted for each spatiotemporal patch in the transformed domain, can improve image reconstruction of exercise stress cardiac images relative to standard compressive recovery techniques.