key feature of the method is that the measured data in a given direction is not directly used to estimate the denoised

We thank the reviewers for their positive comments on clarity, novelty, and convincing experiments. It was unexpected to us too that such a simple method would work so well. Sparsity in a learned basis is an important approach distinct from our own, and we will mention it. We train one regressor per held-out volume ( 2.2). Pastur (MP) are patch based algorithms, assembling voxels from a patch across volumes into a matrix.