Most of the unrolling-based deep learning fast MR imaging methods learn the parameters and regularization functions with the network architecture structured by the corresponding optimization algorithm. In this work, we introduce an effective strategy, VIOLIN and use the primal dual hybrid gradient (PDHG) algorithm as an example to demonstrate improved performance of the unrolled networks via breaking the variable combinations in the algorithm. Experiments on in vivo MR data demonstrate that the proposed strategy achieves superior reconstructions from highly undersampled k-space data.
This abstract and the presentation materials are available to members only; a login is required.