Table 2.

B‐spline transformation parameters for the learned transformation.

Transformation	Grid size	Distribution
${\mathbf{t}}_1$	2 × 2 × 2	$\mathcal{U}{(-6.4,6.4)}^3$
${\mathbf{t}}_2$	4 × 4 × 4	$\mathcal{U}{(-3.2,3.2)}^3$
${\mathbf{t}}_3$	8 × 8 × 8	$\mathcal{U}{(-1.6,1.6)}^3$
${\mathbf{t}}_4$	16 × 16 × 16	$\mathcal{U}{(-0.8,0.8)}^3$

$\mathcal{U}{(a,b)}^n$ is the multivariate uniform distribution that samples vectors with n components in the [a,b] interval in voxels. The resulting transformation is defined as ${\mathbf{T}}_{\mathrm{learned}}={\mathbf{t}}_1\circ {\mathbf{t}}_2\circ {\mathbf{t}}_3\circ {\mathbf{t}}_4$.