326
H. Courtecuisse et al.
the cortical or ventricles surface on iMRI), homologous landmarks, or 3D vessels.
Ideally, segmenting the intra-operative images should be fully automatic. However,
semiautomatic methods are commonly used as long as interactions remain very
limited.
13.2.2.3 Intra-operative Boundary Conditions and Loads
Many authors have used imposed displacements on FE nodes to match the model
with the structures segmented from intra-operative images. In this case, the computational model could be mostly considered as a representation of the warping
function in a registration problem. The brain shift phenomenon is not simulated
directly; instead the model is used to extrapolate the deformations of the brain from
a sparse set of data, on a mechanical basis.
As opposed to this approach, several authors have proposed to actually model
the influence of gravity, loss of CSF of tissue resection on the brain, to simulate
its deformation [13, 19]. In that context, the collected intra-operative data are not
derived as loads for a forward simulation but are used to find the parameters leading
to the optimal solution of an inverse problem.
13.2.2.4 Efficient Computational Methods
Computational efficiency is a crucial factor in the operating room. While the
entire process must be considered, including image acquisition and processing,
the computational model resolution itself can be one of the main limiting criteria.
Real-time or interactive simulation is generally not required in a neurosurgical
context, but overall computation time should be limited to a couple of minutes. This
additional duration could be especially acceptable if the surgery can be resumed just
after data acquisition, while the model is ran and the pre-operative MRI images are
updated.
13.2.2.5 Modelling Tissue Resection
Neurosurgery often requires the retraction and resection of the brain tissue, for
instance, to remove a tumour, entailing many additional issues. Topological modifications (such as re-meshing or element suppression) must then be performed on
the model. These modifications invalidate any precomputations, raising additional
concerns about the computation time. An iterative approach is, for example,
proposed in [20], involving several simulations and re-meshing steps providing
updated images in approximately 7 mn. However, from the best of our knowledge,
a functional, robust, and validated solution has not yet been proposed to address the
problem of resection while meeting the constraints of the operating room.
H. Courtecuisse et al.
the cortical or ventricles surface on iMRI), homologous landmarks, or 3D vessels.
Ideally, segmenting the intra-operative images should be fully automatic. However,
semiautomatic methods are commonly used as long as interactions remain very
limited.
13.2.2.3 Intra-operative Boundary Conditions and Loads
Many authors have used imposed displacements on FE nodes to match the model
with the structures segmented from intra-operative images. In this case, the computational model could be mostly considered as a representation of the warping
function in a registration problem. The brain shift phenomenon is not simulated
directly; instead the model is used to extrapolate the deformations of the brain from
a sparse set of data, on a mechanical basis.
As opposed to this approach, several authors have proposed to actually model
the influence of gravity, loss of CSF of tissue resection on the brain, to simulate
its deformation [13, 19]. In that context, the collected intra-operative data are not
derived as loads for a forward simulation but are used to find the parameters leading
to the optimal solution of an inverse problem.
13.2.2.4 Efficient Computational Methods
Computational efficiency is a crucial factor in the operating room. While the
entire process must be considered, including image acquisition and processing,
the computational model resolution itself can be one of the main limiting criteria.
Real-time or interactive simulation is generally not required in a neurosurgical
context, but overall computation time should be limited to a couple of minutes. This
additional duration could be especially acceptable if the surgery can be resumed just
after data acquisition, while the model is ran and the pre-operative MRI images are
updated.
13.2.2.5 Modelling Tissue Resection
Neurosurgery often requires the retraction and resection of the brain tissue, for
instance, to remove a tumour, entailing many additional issues. Topological modifications (such as re-meshing or element suppression) must then be performed on
the model. These modifications invalidate any precomputations, raising additional
concerns about the computation time. An iterative approach is, for example,
proposed in [20], involving several simulations and re-meshing steps providing
updated images in approximately 7 mn. However, from the best of our knowledge,
a functional, robust, and validated solution has not yet been proposed to address the
problem of resection while meeting the constraints of the operating room.