Parallel Magnetic Resonance Imaging Reconstruction Using Similarity-Based Regularization

The parallel magnetic resonance imaging (parallel imaging) technique reduces the MR data acquisition time by using multiple receiver coils. Because of its ill-conditioned system matrix, the reconstruction suffers from noise amplification at high reduction factors, such as in the standard SENSE reconstruction. Total variation (TV) regularization is a popular technique for solving this problem. However, TV regularized images are vulnerable to staircase artifacts and texture loss. In this paper, we proposed a similarity-based regularization technique which enforces the consistence and similarity of pixel values within the image. The phantom simulation and in vivo experimental results demonstrate that this method can effectively suppress noise amplification in SENSE reconstruction while preserving image details. Compared with TV regularized images, images reconstructed by the new method are free of staircase artifacts and suffer less from structure loss.