Table 1.

Comparison of different methods on the dataset, with affine registration used for reference.

Method	Dice (%)	|JD ≤ 0|
Affine Only	56.33 $\pm$ 0.04	-
VoxelMorph-1	73.07 $\pm$ 0.04	186 $\pm$ 38
VoxelMorph-2	73.94 $\pm$ 0.05	392 $\pm$ 69
SYMNet	71.89 $\pm$ 0.31	0.5 $\pm$ 0.4
LapIRN	71.43 $\pm$ 0.04	0
LapIRNdisp	74.89 $\pm$ 0.18	1757 $\pm$ 259
MF-Net (ours)	75.38 $\pm$ 0.05	332 $\pm$ 22

Dice measures registration accuracy (higher values are better), and |JD ≤ 0| indicates the number of folding voxels that do not conform to the anatomical structure. Please note that the best-performing results are highlighted in bold.