Figure 2.

Model selection demonstrates that A, in this fixed sample, the tensor‐sphere model best described the data (lowest AIC) in most voxels. In some regions later found to contain more lumen space, simpler models were sufficient, but B, a boxplot of the relative AIC values confirms that, for most voxels in both fresh and fixed cases, the tensor‐sphere model explained the data better than a two‐compartment model without restriction (tensor‐ball) and a two‐compartment model without anisotropy (ball‐sphere)