Page 424 - Biomedical Engineering and Design Handbook Volume 2, Applications

402 SURGERY

example of this type of data is the Visible Human Project, which has detailed CT, MRI, and photo-
graphic data of a male and a female. The data are carefully segmented and labeled, and a database
of organs is constructed from the data. The model can then be inspected, for example, using the
11
VOXEL-MAN software, or used to match to other patient data.

14.3.3 Preoperative Analysis and Planning
Once the diagnosis has been made and it has been decided that surgery is necessary, the next step is
to carry preoperative analysis and elaborate a surgical plan of action. This plan can range from sim-
ple tasks such as determining the access point of a biopsy needle, to complex gait simulations,
implant stress analysis, or radiation dosage planning. Because the analysis and planning is specific
to each surgical procedure and anatomy, preoperative planning and analysis software is usually
custom to each clinical application. These systems can be viewed as medical CAD systems, which
allow the user to manipulate and visualize medical images, models of anatomy, implants, and surgical
tools, perform simulations, and elaborate plans. To give the reader an idea of the current scope of
these systems, we will briefly describe two planning systems, one for orthopedics and one for radi-
ation therapy.
In orthopedics, planning systems are generally used to select implants and find their optimal
placement with respect to anatomy. For example, a planning system for spinal pedicle screw inser-
tion shows the surgeon three orthogonal cross sections of the acquired CT image (the original xy
slice and interpolated xz and yz slices) and a three-dimensional image of the vertebrae surfaces. The
surgeon selects a screw type and its dimensions, and positions it with respect to the anatomy in the
three cross-sectional views. A projection of the screw CAD model is superimposed on the images,
and its position and orientation with respect to the viewing plane can be modified, with the result
displayed in the other windows. Once a satisfactory placement has been obtained, the system stores
it with the screw information for use in the operating room. Similar systems exist for total hip and
total knee replacement, which, in addition, automatically generate in some cases machining plans
(cut files) for intraoperative surgical robots. Other systems also extract kinematic or finite-element
models and perform gait and stress analysis that help surgeons estimate the effectiveness of the pro-
posed solution.
Another example of a complex planning system is in the field of radiation therapy. The goal of
radiation therapy is to kill tumor cells by exposing them to a radiation beam while affecting as little
as possible the surrounding healthy cells. One way of achieving this is to expose the tumor cells to
radiation beams from different directions so that the cumulative radiation effect on the tumor cells
destroys them while preserving the surrounding healthy cells. The planning task consists of identi-
fying the tumor and the critical areas where no radiation should be present from MRI images, and
then selecting the number of beams, their radius, intensity, duration, and placement that maximizes
the radiation to the tumor cells while minimizing the radiation to other cells, especially those in the
critical areas. This problem is formulated as a geometric minimum-maximum constrained optimiza-
tion problem, and solved with a combination of geometric and nonlinear optimization techniques.
The planning system includes a data visualization and volume definition module, and outputs a
series of location commands to the robotic arm carrying the radiation source, and the beam infor-
mation at each location.

14.3.4 Registration
Multimodal registration is one of the key steps for information integration in CIS systems. The goal
of the registration process is to allow the combination of data from several modalities, possibly taken
at different times, so that they can be viewed and analyzed jointly. Registering two data sets consists
of finding a transformation that aligns common features in two modalities, so that their spatial loca-
tions coincide. Registration is necessary for many tasks such as