Biomedical Engineering Reference
In-Depth Information
uum physics should be developed which will insure the simulation to be real-time
and accurate. In the future, these models will allow at the same time coarse to fine
deformation depending on the loading frequencies.
MRI is one of the most widely used medical imaging techniques to visualize the
structure and function of the body. Early techniques such as region growing and
split-and-merge are mostly heuristic and may not be very robust. Modern techniques
use optimization which can be roughly classified into two categories: combinato-
rial methods [
14
,
15
] and variational methods [
16
,
17
]. Typically, medical image
segmentation is done by going through 2D MR slice images. However, even with
those advanced 2D methods, the necessity to analyze and edit each slice in every
MRI data set is a daunting task. Though the 2D sections convey all the information
without any ambiguity, some artifacts can be seen only on 3D views since they do
not contribute significantly to each individual 2D slice. Therefore, accurate and pre-
cise segmentation of the tissue boundary from 3D MR images is an essential and
crucial component for 3D medical computing. When the object of interest is chang-
ing, automatic segmentation becomes even harder [
18
,
19
]. We should explore how
to integrate various modern techniques, and shape and prior cues to achieve intel-
ligent segmentation [
20
-
22
]. After that, the main application will be the creation
of 3D mesh models of tissue for finite element modeling (FEM). The surface and
volume vector models can be used for further biomechanical processing and analy-
sis. It is thus essential to develop efficient algorithms for automated extraction and
reconstruction of patient specific models from medical images.
1.3.2 Segmentation, Deformation and 3D Reconstruction
Medical simulation using patient-specific MRI data requires the design of fast and
accurate segmentation techniques to extract, from the images, structures of interest
(e.g. muscles, and cartilages). The segmented structures (e.g. bones and ligaments)
can be used to reconstruct 3D models from patient-specific MRI data for applica-
tions such as stress and other biomechanical analyses. Direct segmentation involves
a detection step where regions are identified in images and a classification step
where regions are combined to form new regions. For complex problems, direct seg-
mentation is noise-sensitive, not robust and quite inaccurate. The addition of prior
knowledge can improve significantly the results, yielding sophisticated classification
methods [
23
-
25
] using registration algorithms [
26
-
29
] or deformable models [
30
,
31
].
Existing deformable models (deformable contours [
32
], parametric [
33
], implicit
[
34
] and discrete [
28
,
35
]) are typically controlled by external forces that attract
the model towards image features. New methods can extend previous work on
deformable models [
36
,
37
], e.g. the simplex mesh framework, in order to improve
mathematical definition and internal and external force computation. Complexity
will be reduced by using levels of details (LODs), medial axis representations and
efficient particle system simulation techniques. Eventually, the variety of acquisi-
