Figure 3.

We use the Jaccard overlap metric (intersection of coregistered labeled regions over their union) to compare performance in this evaluation. The data is visualized with a box and whisker plot, with notches. These plots show the median toward the center of the box. The edges of the box delimit the medians of the data above and below the median. The whiskers and points show the minimum and maximum of the data and any points that are plotted may be considered outliers. We used pairwise Student T-tests to determine whether performance differences are significant. In this figure, the MSQAff, CCAff and MIAff overlap results all report the quality of the affine mapping to the derived template from the (MSQ, MSQ), (CC, CC) and (MI, MI) results. Deformable results for these three runs are not shown. The MI-based affine registration gave the best performance for both brain and cortex labeling thus providing the best initialization for follow-up deformable registration. For this reason, the MSQDiff, CCDiff and MIDiff results all use the MI metric for the affine component and MSQ, CC and MI during deformable registration.