Biomedical Engineering Reference
In-Depth Information
Fig. 2 Comparison between FE model and statistical meshless model constructed from the
medical image. (a) T2 MRI of the brain with the tumor and ventricles present, notice that no
clear boundaries can be easily defined, especially for the tumor, (b) finite element model of
ventricles, (c) finite element model of the tumor, (d) soft classification of ventricle, (e) soft
classification of tumor, (f) statistical meshless model of ventricle and (g) statistical meshless
model of tumor, dots represent nodes while crosses represent integration points. Notice that no
specific tissue class is defined in the domain. Material properties are assigned directly to the
integration points based on soft classification results
All mass in the model is located at the nodes. Each integration cell is allocated a
mass based on its volume and density. This mass is split evenly to the n nodes in the
support domain of that cell. Many nodes will thus have different masses propor-
tional to the number of support domains in which they are included.
2.2 Material Property based on Soft Tissue Classification
Traditional “hard” segmentation required in FE modeling divides intracranial area
into nonoverlapping constituent regions with discretely defined boundaries. The
segmented image data contains less information than the original image as contin-
uous intensity data has been replaced with discrete label maps. Furthermore,
malignant brain tumors can spread within the brain and spine, as shown in
Fig.
2a
. They lack distinct borders and therefore are very difficult to segment. We
use a modified fuzzy c-mean soft classification in the intracranial regions, as we
only require an approximate tissue classification results that are used to assign
material properties to integration points where stress is evaluated.
Fuzzy c-mean has been used with some success in the soft or fuzzy segmentation
of brain MR images [
10
]. It clusters similar intensity data by computing the
