Results and Validation

This section gives validation results on synthetic brain MR images with a wide range of noise and bias values as well as real MR data. The computation for each iteration is $O (\vert\mathcal{A}_t\vert \vert\mathcal{T}\vert \vert\mathcal{N}_t\vert)$. We have found empirically that, with simulated MR images from the BrainWeb [31] database, ICER produces the largest reduction in RMS errors after a single iteration itself. Subsequent iterations converge at a nearby RMS-error value. All results in this paper employ a single iteration of ICER.