Sparsity regularized nonlinear inverse reconstruction for subsampled parallel Dynamic Cardiac MRI

Abstract

The emerging Compressed Sensing (CS) theory which synthesizes the sparsity priori, incoherent measurement, and nonlinear reconstruction of a desired signal comprehensively, has been employed in the Dynamic Cardiac Magnetic Resonance Imaging (MRI) to accelerate the imaging speed and to achieve the desired high temporal-spatial resolution further. Although the requirements of CS theory can't be guaranteed in the practical scenario of Dynamic Cardiac MRI strictly, the ideology behind the CS theory can still contribute significantly to the design of integrated sampling and reconstruction framework based on the sparsity priori of dynamic MR image. This work summarizes available sparsity priors for Dynamic Cardiac MRI which all based on the high correlation between adjacent frames of dynamic image. Specific attention is paid to the circumstances in which individual prior tends to perform well, to their effects on reconstructed image respectively, and to their mutual influences. Then the possibility of improving the reconstruction quality via exploiting two or more priors simultaneously is investigated, and the performance of diverse combinations are compared on cardiac cine data set. Based on the experimental investigation presented in this work, better understanding of the characteristics of each sparsity prior of dynamic cardiac MR image and their performance in subsampled Dynamic MRI inverse reconstruction can be achieved.