Biomedical Engineering Reference
In-Depth Information
To avoid optimizing all model parameters for each tumor case in the used
dataset, the first images after tumor growth (6th-day scan for dog cases, and the
first scan for the HC) are used to approximate κ r . This approximation involves the
assumption of negligible edema spread, tumor infiltration, and tissue death between
this and the final scan, which corresponds to κ t . Additionally, since tumors in the
starting images are small, the stresses and deformation due to the tumor may be
assumed negligible. The small deformation assumption was confirmed for the dog
cases by measuring the deformation of landmark points around the tumor in the
pre-transplantation and the 6th-day scans. These landmark points were selected
manually by an expert in the area around the tumor. Under the above-mentioned
assumptions, regions D r and T r were obtained from segmentations of the tumor
and edema in the starting images. The starting images already had some peri-tumor
edema. Therefore, e was treated as a parameter with e
[0 . 1 , 0 . 4].
In addition to performing simulations with different values of P and e for each
tumor case, simulations were also performed with varying material parameter
values for the brain tissues. For each simulation, the error in the deformation
predicted by the model compared to the actual deformation was computed. This
error calculation performed through the use of corresponding landmark points that
were found by a human expert in the used two images for each tumor case. The
overall procedure is summarized in the following steps.
1. Rigid registration of the target (final) scan to the respective starting scan
(used to define κ r ) was performed. Registration was carried out using a
stepped multiresolution minimization of normalized mutual information
via the CISG registration package [102].
2. Pairs of corresponding landmarks were manually identified by a human
expert (rater) in the starting and target images. The number of landmarks
for each case varied between 20 and 25 (see Table 2) and were selected in
the region around the tumor, where large deformation is observed.
3. A combination of manual and automatic segmentation of the starting image
into brain, ventricles, falx, tumor, and edema was performed. The falx
cerebri was also manually segmented in order to use it for the boundary
condition as explained above.
4. The automatic mesh generation approach explained in Section 4.3.6 was
used to generate tetrahedral FE meshes for each of the datasets. For visual
illustration, views of the mesh generated for DC1 are shown in Figure 15.
5. Based on the mesh generated in the previous step and the model equations
provided in Section 4.3.5, the FE package ABAQUS was used to obtain
the simulated tumor-induced deformation. Simulations were performed
with different values of P and e , and of the material parameter values.
Search WWH ::




Custom Search