Online undersampled dynamic MRI reconstruction using mutual information

We propose an algorithm based on mutual information to address the problem of online reconstruction of dynamic MRI from partial k-space measurements. Most of previous compressed sensing (CS) based methods successfully leverage sparsity constraint for offline reconstruction of MR images, yet they are not used in online applications due to their complexities. In this paper, we formulate the reconstruction as a constraint optimization problem and try to maximize the mutual information between the current and the previous time frames. Conjugate gradient method is used to solve the optimization problem. Using Cartesian mask to undersample k-space measurements, the proposed method reduces reconstruction error from 3.41% in ModCS, 1.57% in ModCS_Res and 1.16% in CaNNM to 0.61% on average per frame. Moreover, fast reconstruction of images at the rate of 2 to 10 frames per second makes our method a good alternative for current CS based methods in online dynamic MRI applications.