Biomedical Engineering Reference
In-Depth Information
a
Optimizer
M
T
Transformer
Moving image
Point
T(M)
No
F
S(F,T(M))
Yes T
opt
Terminate
?
Measure
Fixed image
b
Boundary
conditions,
materials
M
Segmentation,
Meshing
Biomechanical
model
Preop. image
Loading
T
Transformer
Solver
Intraop. info
T(M)
Fig. 1 Main components of a general image-based registration solver (a) and of a registration
solver based on a biomechanical model (b)
changes the parameters of the transformer, in order to find the best agreement
(as defined by the chosenmeasure) between the transformed image and the fixed image
In case of a biomechanics based registrationmethod, the workflow is very different.
The preoperative image is segmented in order to extract the anatomical features of
interest. Based on this segmentation, a computational grid is generated (meshing). The
boundary conditions (such as contacts between brain and skull) andmaterial properties
for each type of tissue are defined. The loading definition is the last step required to
obtain a complete biomechanical model. Intraoperative information may be involved
in defining the loading (typically very sparse information is sufficient, such as the
surface deformation in the area of craniotomy [
3
,
7
,
13
,
14
]). Once the biomechanical
model is constructed, a solver is used to compute the transform, which is then applied to
the preoperative image [
15
]. Unlike image-based registration, computing the transfor-
mation does not always require a fixed image. This makes biomechanics-based
registration methods very attractive for intraoperative applications.
Registration is generally computationally expensive, the transformer being the
computational bottleneck. This is one of the reasons registration has been largely
confined to preoperative applications [
12
].
The transformer maps points in the moving image to new locations in the
transformed image. The intensity of the points in the transformed image is determined
by interpolating intensity values of the corresponding points in the moving image.
Therefore, the transformation of the moving image can be seen as a two steps process:
