Table 2.

Image registration performance of Attention-Reg under various training strategies. The initial Dice is 0.619±0.17 and the initial SRE uniformly distributes between 0 and 20 mm.

Training settings	Result SRE	Result Dice	#Parameters
Attention-Reg without pre-training	4.66±2.49 mm	0.789±0.07	1,248,905
MIND-SSC + Attention-Reg	5.23±2.76 mm	0.778±0.08	1,187,173
Attention-Reg co-trained with segmentation	4.98±2.76 mm	0.783±0.08	1,722,995
Constrastive loss pre-trained Attention-Reg	4.27±2.38 mm	0.802±0.07	1,248,905