Illustration of the prf-based reconstruction analysis. A, The standard model of prf-based reconstruction scales the prf properties for each voxel (x0, y0, s0) with the BOLD amplitude observed during each task condition. The reconstructed image is obtained by summing over the voxel dimension. B, The multivariate prf-based reconstruction approach developed in this study follows three steps. First, the given receptive field properties for every voxel are represented in a 2D pixel matrix (see Materials and Methods). Second, the pixel matrix is then regressed against the measured BOLD amplitude, separately for each task condition. Third, plotting the estimated regression weights reveals the stimulus reconstruction. Stimulus reconstruction shown for the group averaged (N = 17) left upper horseshoe from the localizer block. White contour lines illustrate the stimulus locations. C, Individual stimulus reconstructions for the summation method (top) and multivariate method (bottom). D, Stimulus reconstructions for different levels of sparsity, controlled by the L1/L2 ratio, ranging from 0 (L2, non-sparse) to 1 (L1, sparse). E, Determining the optimal L1/L2 ratio by comparing resulting reconstructions with the presented stimulus (left). Group averaged reconstruction accuracies indicated an optimal L1/L2 ratio of ∼0.2 corresponding to a very moderate level of sparsity. Error bars denote SD.