Biomedical Engineering Reference
In-Depth Information
In the context of image guided neurosurgery it is very important to be able to
predict the effect of certain procedures on the position of pathologies and critical
healthy areas in the brain. The most typical example is the prediction of the
displacement field within the brain due to craniotomy (so called “brain shift”).
This phenomenon can lead to large deformations within the brain, limiting the use
of preoperative images for guiding the surgery [
2
]. Acquiring new images during
surgery is only practical at lower resolutions (due to the acquisition time required),
and intraoperative MRI imaging is not widely available. Therefore, it is important
to be able to compute the deformation of the brain intra-operatively, based on some
sparse information (such as the deformation of the brain surface in the craniotomy
area) [
3
], and use the computed deformation field to register high quality preopera-
tive images to the current organ position.
In recent years many researchers have proposed the use of biomechanical
models for predicting intraoperative brain deformation [
2
,
4
-
8
]. While modeling
approaches can be different, the solution method of choice is usually the finite
element method (FEM), which implies that the computed deformation field is
expressed as displacements at every node of the FEM mesh.
Mesh-free methods can also be used for computing the solution [
9
]. While these
methods do not require a mesh (except the background mesh used for integration),
the computed deformation field is expressed in the same way as in the case of FEM,
as nodal displacements.
Recent developments in solution methods and parallel hardware implement-
ations made it possible to solve biomechanical models within the real time
constraints of neurosurgery [
8
,
10
,
11
]. Nevertheless, there are, to the best of our
knowledge, no detailed references in the literature discussing how the computed
deformation field should be used to update the preoperative image intraoperatively.
In this paper, we discuss the use of the deformation field expressed as nodal
displacements for warping high quality preoperative images. In the next section, we
will describe how a generic image based registration algorithm works and compare
it with a biomechanics-based one, identifying the requirements derived from the
specific form of the deformation field and from the application area (brain shift).
We present an implementation of this algorithm and a registration result in Sect.
3
and draw some conclusions and discuss possible real
time implementation
approaches in the last section.
2
Image Warping Using Biomechanical Models
A standard image based registration method includes the components presented in
Fig.
1
[
12
]. The moving (preoperative) image (M) is transformed using the chosen
transformer to obtain the transformed image T(M), which is then compared with the
fixed (intraoperative) image (F) based on a chosen measure (S). The result of
the comparison is a metric which is used by an optimizer to change the parameters of
the transformer in such a way that the difference between the transformed image and
the fixed image are minimized. An optimization loop is therefore needed, which
