SENSE reconstruction with nonlocal TV regularization

Ill-conditioning is serious problem in SENSE reconstruction, especially when large acceleration factors are employed. For Cartesian SENSE, Tikhonov regularization and total variation have been commonly used. However, the Tikhonov regularized image usually tends to blur edges and total variation regularization has a blocky effect. In this paper, we propose a new SENSE regularization technique that is based on nonlocal total variation with Bregman iteration. It penalizes highly oscillatory noise and allows sharp edges and fine textures in reconstruction. The method is shown to be able to significantly reduce the artifacts in SENSE reconstruction.