© RSNA, 2012
The LOST algorithm was implemented in two stages as described in our previous study (20). For the first stage, a low-resolution image generated from fully sampled central k-space was used for identification of similarity clusters, with parameters block size, Nb = 8, and match threshold, λmatch = 0.1. For each reference block, comparison with other blocks was limited to a neighborhood of radius 8 voxels in the x-y direction and of radius 1 voxel in the z direction. The maximum number of blocks in a similarity cluster was limited to 16. For shrinkage in the first stage, hard thresholding was used, where the threshold τht was set to 0.04 times the largest (in absolute value) coefficient of the zero-filled coil image. For the second stage, the estimate from the first stage was used for identification of similarity clusters with parameters Nb = 4, lmatch = 0.05, and the same remaining parameters from the first stage. For shrinkage, LOST alternated between hard thresholding and Wiener filtering, where the thresholds τht and τwie were respectively set to 0.03 and 0.02 times the largest coefficient of the estimate from the first stage.