Assumption-Free Assessment of Corpus Callosum Shape: Benchmarking and Application

Figure 3

Case study results. Note. Panel (a) depicts landmark (indicated in red) and semilandmark distribution of the mean real corpus callosum shape (large black points) and individual shapes (reduced opacity to show overplotting). Manual landmark assignment reflected the fact that the sliding scheme applied more landmarks to the genu (lower right), likely reflecting heightened detail in this area when compared with the midbody. Due to a combination of visual clarity and computational expense in overplotting, a random subset (650 of a total 2088) are shown. Panel (b) shows reconstructions of shapes positioned at -1 and +1 standard deviations on the first components in landmark and eFourier-based shape space demonstrates the similarity of this component. The positions of contours in the first two dimensions of these shape spaces are presented to the right. Panels (c) and (d) show the positions of shapes on the first two components of Landmark (c) and eFourier (d) PCA results, with greyed out axis numbers indicating scale (absolute position in tangent space for landmark, z score for eFourier). The scale of PC1 and PC2 is consistent within figures. Note: the first two components depicted explain 43.9% (Landmarks) and 62.1% (eFourier) of variability in global shape space—if these results were to be interpreted for theoretical outcomes rather than illustrative methodological comparison, additional dimensions would be required, and the results would likely be highly sample-specific.