Biomedical Engineering Reference
In-Depth Information
15.6
Model Inputs
One of the intriguing aspects of employing a computational model to update
images during image-guided neurosurgery is the opportunity to maximize
the utility of both preoperative and intraoperative data for neuronavigation.
This seems important at the present time given the wealth of preoperative
information and planning which accompanies most neurosurgical cases, yet
there is the potential for loss of or de-emphasis on this information with adop-
tion of intraoperative MR imaging. Preoperatively, image scans can be
employed to define the patient-specific computational model to be deformed
intraoperatively. The possibility of using additional MR sequences to derive
patient specific brain tissue properties through the emerging techniques of
diffusion tensor imaging
36, 37
and elastography also exists. Once in the OR,
several forms of reduced or incomplete data (short of volumetric imaging
with CT and MR) are also available for incorporation within the model. Spe-
cifically, cortical surface motion can be monitored with automated surface
digitalization techniques such as laser scanning. Coregistered ultrasound
provides another information source on the movement of some subsurface
structures; for example, the ventricular system, which can provide internal
constraints on the computed deformation field. In this section, examples of
model inputs available from both preoperative and intraoperative data are
briefly illustrated.
38, 39
15.6.1
Preoperative Data
Prior to surgery, high resolution MR imaging is routinely acquired and is ideal
for model discretization purposes. Following this image series, a 3D represen-
tation can be constructed from the MR slices by segmenting the brain from the
cranium using an image manipulation platform such as
Analyze AVW
(Bio-
medical Imaging Resource—Version 2.5, Mayo Foundation, Rochester, MN).
After the extraction, each voxel within the volume can be saturated to a con-
stant intensity value and a marching cubes algorithm
40
used to generate a sur-
face description characterized by triangular patches. Armed with the
resulting patch description, a 3D tetrahedral mesh can be generated.
41
If inter-
nal structures are to be preserved, the process of image segmentation and
boundary generation can be employed for each constituent of interest in
order to define a composite wire-frame boundary to guide the mesh genera-
tion process.
An alternate strategy which is particularly helpful for incorporating com-
plex structures such as gray
white matter boundaries is to use an image-to-
grid segmentation scheme. In this case, the average voxel intensity from the
MR image volume can be determined for each individual element within
the finite element mesh. An intensity-to-material-property map can then be
