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

Figure 2

Shape manipulations, shape recovery, and shape space recovery. Note. Panel (a) shows shape manipulations of simulated shapes. These simulations induce variation based on biologically distinguishable regions (genu, splenium, and midbody), with some exaggeration of properties most common in the literature (volume, thickness, and curve). Panel (b) shows manipulations of nonshape properties: the distribution of points denoting the contour, the scale, rotation, and degree of local (rather than global) shape complexity. Panel (c) depicts Jaccard index comparing original and reconstructed shape across manipulations, with expanded view to the left due to strong ceiling effects. Panels (d), (e), and (f) depict the results of the principal components analysis undertaken. Panel (d) shows results when analysis includes only global shape manipulations. The background image shows hypothetical possible variation in corpus callosum shape across the first two principle components. The data points indicate where each shape is positioned within this space. The colour and label reflects which manipulation was undertaken: midbody curve (MC) in red; splenium size (SS) in green; midbody length (ML) in purple, and an admixture of these manipulations multiple features (MF) in black. Panel (e) shows results when global shape (MC, SS, ML, MF) and nonshape parameters of randomized points (RP in dark green), scale (S in teal), and rotation (R in orange) manipulations are included. Panel (f) shows results when global shape (MC, SS, ML, MF), nonshape (RP, S, R), and local complexity (LC) manipulations are included in analysis.