Page 404 - Biomedical Engineering and Design Handbook Volume 2, Applications

382 SURGERY

Constructing realistic and efficient deformable models for soft tissue behavior is the main chal-
lenge in achieving realism in surgical training simulators. The deformable tissue models have to be
interactive, efficient enough to be simulated in real time, visually and haptically realistic, and able
to be cut and sutured. The state of the art for interactive deformable object simulation is not suffi-
ciently advanced to build realistic real-time simulations on consumer level computer systems, and
requires supercomputer level computational power and networked simulation technologies.

13.3.1 Patient-Specific Models in Surgical Simulation
A key requirement in development of virtual environment-based surgical simulators is the avail-
ability of visually and geometrically realistic models of the anatomy. It is essential to have a library
of anatomical models with relevant pathologies to be able to develop an educational curriculum. For
applications of surgical simulators beyond training, namely, surgical rehearsal and planning, it is
also necessary to have the capability to construct patient-specific models of the anatomy. As dis-
cussed above, the geometric models used in surgical simulations are constructed from medical diag-
nostic images, such as magnetic resonance and computerized tomography images. The medical
diagnostic images are first segmented to identify the image regions corresponding to the individual
anatomical structures. The segmented images are then processed through mesh generation algo-
rithms to create surface and volumetric geometric models needed for the visualization and physical
modeling, respectively. Both medical image segmentation and mesh generation are very active
areas of research. Level set methods are certainly the most popular techniques to capture the struc-
24
tures of interest during segmentation. The fast marching methods and dynamic implicit surfaces 25
are two classes of level set algorithms that are widely used. These algorithms are relatively insen-
sitive to noise, yield smooth surfaces, and have the topologies of the extracted surfaces robust to the
choice of the threshold. There are also a number of commercial software packages available (e.g.,
26
Amira ), which provide manual or semiautomated segmentation and mesh generation tools. In
spite of the availability of variety of tools for segmentation and mesh generation, these tasks remain
to be labor intensive, requiring significant user interaction during initial segmentation, and post-
processing for improving the quality of segmentation and generated meshes. Automation of image
segmentation and mesh generation are key requirements for the use of patient-specific models in sur-
gical simulators.

13.3.2 Deformable Object Modeling and Simulation
Most biological tissue is elastic. Therefore, a realistic simulation of surgical manipulation in a vir-
tual environment requires modeling of physical deformations of soft tissue. There are two widely
used types of physically based deformable object models used in the literature, namely, lumped ele-
ment models (also known as mass-spring-damper models) and finite element models.
Lumped element models are meshes of mass, spring, and damper elements. 27–29 They are the most
popular models for real-time surgical simulators, because they are natural extensions of other
deformable models used in computer animation, conceptually simple, and easy to implement. A
common problem with the lumped parameter models used in literature is the selection of component
parameters, spring and damper constants, and nodal mass values. There is no physically based or sys-
tematic method in the literature to determine the element types or parameters from physical data or
known constitutive behavior. The typical practice in the literature is somewhat ad hoc, the element
types and connectivities are empirically assumed, usually based on the structure of the geometric
model at hand, and the element parameters are either hand tuned to get a reasonable looking behav-
ior or estimated by a parameter optimization method to fit the model response to an experimentally
measured response. 30,31 There are several simulation libraries that are available for development of
surgical simulations using lumped element models. 32
Linear finite element models are used as a step to get closer using models with physically based
parameters. 33–35 Linear finite element models are computationally attractive as superposition can be