Real MRI data

  

Figure 3: Real data: ANIMAL-only vs ANIMAL+sulci

Standard ANIMAL: When applied to real data, the image in Fig. 3-b shows that that the standard   non-linear registration is not enough to align cortical structures. In fact, it does not appear to be much better than the linear registration. This is confirmed by the quantitative results in Table 2, where the average ${D_{\mathsf{surf}}}$value for the standard   registrations is only slightly better than that for linear registrations.

ANIMAL+selected sulci: The use of 5 paired sulci improves cortical registration significantly in the neighbourhood of the chosen sulci as shown in Fig. 3-c. The central sulcus (purple), superior frontal (orange), middle frontal (red), olfactory (black) and Sylvian (red) (arrows) are well defined, and this is confirmed by improved ${D_{\mathsf{surf}}}$ values in Table 2, when compared to the standard   registrations. Unfortunately, the effect appears to be relatively local in that alignment of neighbouring sulci are only somewhat improved (e.g., postcentral sulcus). This is probably because there is no explicit representation of the other sulci in the registration, so they do not participate in the fitting process. However, using some labelled sulci yields a 0.50mm average ${D_{\mathsf{surf}}}$ reduction (from 4.5 to 4.0mm) that is highly significant (p<0.0001, T=157, d.o.f.=320, paired T-test).

ANIMAL+selected sulci+all sulci: When the previous result is used as input for a registration using all sulci as features, then some of the other sulci (e.g., postcentral) come into alignment, however others may be attracted to non-homologous sulci, since no explicit labelling of sulci is used in this step.