File S1 - Multi-Compartment T2 Relaxometry Using a Spatially Constrained Multi-Gaussian Model

Contains Figure S1, rMSE of MWF computed using proposed method on brain simulation with different μS at SNR  =  100 and 300 (μN  = 0.013). The optimal value was found at 0.01 for both noise levels. Figure S2, Visual results for spatial constrained (μN = 0.013) on brain simulation at SNR of 100 (top) and 300 (bottom), where μS = 0.0001, 0.001, 0.01, 0.02, 0.1, respectively, from left to right. The ones within red boxes provide the best visual and numerical result, and show that the optimal value of this parameter is not overly sensitive to SNR. Figure S3, Convergence of the proposed algorithm. Left: MWF maps computed after 20, 30, 80 and 100 iterations on an in vivo example. Right: The numerical convergence of the cost function shows a classic pattern, whereby convergence is reached at 20 iterations, and further iterations do not appreciably reduce the cost. Although MWF maps begin looking reasonable in as few as 20 iterations, we chose 30 iterations to provide a margin of error. Figure S4, MWF maps computed from another in vivo MS patient scan. Top to bottom are MWF maps from conventional method, spatial constrained method and FLAIR images. Arrows in FLAIR images point to lesions. Figure S5, Left to right, single axial slice of a MS patient showing A] T2-weighted image, B] MWF map from conventional NNLS method, C] MWF map from spatially constrained Gaussian method, and D] MWF map reconstructed from sparse L1-regularized method.
(DOCX)