Iteratively refined nonlocal total variation regularization for Parallel variable density spiral imaging reconstruction

Parallel variable density spiral imaging is an efficient fast imaging technique that combines spiral-based fast k-space acquisition and sensitivity encoding reconstruction of parallel imaging. However, due to its ill-conditioned system matrix, sensitivity encoding reconstruction can suffer from severe noise amplification or residual aliasing artifacts. To solve the problem, a regularized reconstruction that is based on iteratively refined nonlocal total variation regularization was proposed in this study. The proposed method extends current popular total variation to non-adjacent pixels and iteratively improves the results based on Bregman scheme. The phantom simulation and in vivo experiments results demonstrate that this method can effectively suppress noise amplification in sensitivity encoding reconstruction while preserving image details. Compared with conventional total variation regularized images, images reconstructed by the new method are free of staircase artifacts and suffer less from structure loss.