Skip to main content
. 2024 Jun 25;2:1–33. doi: 10.1162/imag_a_00197

Fig. 12.

Fig. 12.

Deformable 3D registration accuracy as mean Dice scores over the 21 largest brain regions (large-21), 10 fine-grained structures not optimized at SynthMorph training (small-10), and image similarity. Each violin shows the distribution across the skull-stripped cross-subject pairs from Table 1. For comparability, we initialize all deformable tools with affine transforms estimated by NiftyReg. The asterisk indicates SynthMorph performance without skull-stripping. Downward arrows show median scores outside the plotted range. Higher Dice and lower MSE-MIND are better.