Table 6.

Evaluation result for multi-modal image-to-image tests.

	Deformation Error w.r.t LDDMM optimization on T1w-T1w data [mm]
Data percentile for all voxels	0.3%	5%	25%	50%	75%	95%	99.7%
Affine (Baseline)	0.1664	0.46	0.9376	1.4329	2.0952	3.5037	6.2576
T1w-T1w LP	0.0348	0.0933	0.1824	0.2726	0.3968	0.6779	1.3614
T1w-T1w LPC	0.0289	0.0777	0.1536	0.2318	0.3398	0.5803	1.1584
T1w-T2w LP	0.0544	0.1457	0.2847	0.4226	0.6057	1.0111	2.0402
T1w-T2w LPC	0.0520	0.1396	0.2735	0.4074	0.5855	0.9701	1.9322
T1w-T2w LP, 10 images	0.0660	0.1780	0.3511	0.5259	0.7598	1.2522	2.3496
T1w-T2w LPC, 10 images	0.0634	0.1707	0.3356	0.5021	0.7257	1.1999	2.2697

Deformation error (2-norm) per voxel between predicted deformation and optimization deformation. Percentiles over all deformation errors are shown to illustrate the error distribution. LP: prediction network. LPC: prediction+correction network. 10 images: network is trained using 10 images (90 registrations as training cases).