Figure 4.
Results from the scan-rescan experiment in Sec. 3.1 showing the spatial distribution of the reliability a sagittal slice. The background image shows the template T1-weighted map. The left panels show the coefficient of variation (CV), and the right panels show the intraclass correlation (ICC). Within each side, the slices are organized to show a different method in each row and a different template type in each column. The results generally show large spatial variation across methods, with higher variation in voxel-based than region-based methods. Voxel-based analysis tended to have higher reliability in deep white matter and lower in superficial white matter. Region-based analysis tended to have more uniform error rates than methods analyzing individual voxels.