Biomedical Engineering Reference
In-Depth Information
to the inability of the suggested tumor growth model in reproducing physically
realistic brain tissue deformations, which therefore derails the image matching
process.
A good model for deformations induced by tumor growth is essential for
achieving a registrationwith accuracy that is acceptable for purposes of surgery and
therapy planning. AlthoughKyriacou et al. [79] used a biomechanical model of the
deformation caused by tumors to register images of tumor patients to anatomical
atlases, this approach was only implemented in 2D and relied on a computationally
expensive regression procedure to solve the inverse problem of estimating the
tumor location in the atlas.
Here an approach for establishing the deformable registration of a normal
anatomical atlas to a tumor-bearing brain scan is presented. This approach re-
quires the integration of three components. The first is a biomechanical 3D model
for the soft-tissue deformation caused by the bulk tumor and peri-tumor edema.
This model is implemented using the finite-element (FE) method and is used to
generate a number of examples of deformed brain anatomies due to tumors starting
from normal brain images. The second component is a statistical model of the de-
sired deformation map that approximates this map via the sum of two components
in orthogonal subspaces with different statistical properties. For any particular tu-
mor case that should be registered to the atlas, a partial observation of the desired
deformation map is obtained via a deformable image registration method devel-
oped for dealing with two normal brain images (HAMMER), which is the third
component of the presented approach. Based on the constructed statistical model
of the deformation, this partial observation is used to estimate the corresponding
mass-effect model parameters that would have produced such a deformation. Fi-
nally, the desired deformation is obtained by applying the mass-effect model to
the atlas image and the use of deformable image registration to match it to the
subject's image.
The rest of this section is organized as follows. In Section 4.2 we provide a
detailed overview of the whole approach. In Section 4.3 we provide a description
of the biomechanical model for tumor mass-effect and show the results of compar-
isons between the actual deformations and deformations predicted by this model
in four real brain tumor cases. In Section 4.4 we describe the developed statistical
approach for estimating the biomechanical model's parameters. Deformable reg-
istration results for a real and a simulated brain case are presented in Section 4.5.
We conclude with a discussion of the approach and suggestions for future work.
4.2. Overview of the Approach
The proposed deformable registration approach is best explained with the aid
of Figure 12. The subject's brain
B
SD
includes regions
T
SD
(bulk tumor), and
possibly
D
SD
(peri-tumor edema). The main goal of the deformable registration
