Biomedical Engineering Reference
In-Depth Information
8.3.2.8
Multimodal Non-Rigid Registration
We have proposed a multimodal version of Romeo [72] where the optical flow
is replaced by a more adapted and general similarity measure: mutual informa-
tion. Mutual information has been presented in the section dedicated to rigid
registration (8.3.2).
Let us note T w as the transformation associated with the deformation field w .
The total transformation T w
T 0 maps the floating volume B onto the reference
volume A . The field w is defined on S B , where S B denotes the lattice of volume B
(pixel lattice or voxel lattice). The cost function to be minimized then becomes:
< s , r > C B || w s w r ||
2
U ( w ; A , B ,
T 0 ) =− I ( A , ( T w
T 0 )( B )) + α
,
where C B is the set of neighboring pairs of volume B (if we note V a neighborhood
system on S B , we have: < s , r > C B s V ( r )).
A multiresolution and multigrid minimization are also used in this con-
text. At grid level and on each cube n , we estimate an affine displace-
ment increment defined by the parametric vector n : s n ,
d w s = P s n , with
P s = I 2 [1 x s y s ] for 2D images, and P s = I 3 [1 x s y s z s ] for 3D images (operator
denotes the Kronecker product).
To be more explicit, in 3D we have:
.
1 x s y s z s 00 0 000 0 0
00 0 01 x s y s z s 00 0 0
00 0 000 0 01 x s
P s =
y s
z s
Let us note T n , as the transformation associated with the parametric field
n . We have T
T d w and T n =
T d w | n , where T d w | n denotes the restriction
=
of T n to the cube n .
A neighborhood system V on the partition derives naturally from V :
n , m ∈{ 1 ··· N } , m V ( n ) ⇔∃ s n , r m / r V ( s ) . C being the set
of neighboring pairs on S k , we must now distinguish between two types of such
pairs: the pairs inside one cube and the pairs between two cubes:
n ∈{ 1 ... N } ,< s , r > C n s n , r n and r V ( s ) .
n ∈{ 1 ... N } , m V ( n ) ,< s , r > C nm m V l ( n ) , s n , r m
and r V ( s ) .
Search WWH ::




Custom Search