X3D IN RADIATION THERAPY PROCEDURE PLANNING
Felix G. Hamza-Lup, Ivan Sopin, Dan Lipsa
School of Computing, Armstrong Atlantic State University, Savannah, Georgia, USA
Omar Zeidan
M.D. Anderson Cancer Center Orlando, Florida, USA
Keywords: X3D, Web-based Simulation, 3D Modeling, Radiation Therapy.
Abstract: Radiation therapy, an increasingly available and effective cancer treatment solution, involves sophisticated
machinery as well as careful planning. Interactive 3D simulations combined with accurate 3D patient
specific data can improve the planning process saving time and resources in generating the optimal
treatment plan. We illustrate the potential of X3D in radiation therapy, specifically radiation treatment
planning. Embedding patient specific data (2D CT scans) in the interactive virtual setup improves the
radiation therapy planning by early detection of collision cases. The X3D system may be used to support the
decision making process as well as innovations in medical planning and training.
1 INTRODUCTION
While 3D simulation is just emerging as an accepted
scientific discipline for medicine, the majority of
applications are in the area of inter-operative
navigation and training. We present the potential of
X3D in radiation therapy, specifically radiation
treatment planning.
Radiation therapy, an increasingly available and
effective cancer treatment solution, involves
sophisticated machinery as well as careful planning.
An interactive X3D-based simulation combined with
accurate 3D patient data obtained from CT scans and
MRI will improve the planning process, saving time
and resources in generating the optimal treatment
plan. We illustrate the potential of X3D in radiation
therapy, specifically radiation treatment planning.
Embedding patient-specific data in the virtual setup
improves the radiation therapy planning by early
detection of collision cases.
A preliminary assessment leads us to believe that
an accurate 3D virtual representation of the radiation
therapy/surgery procedure supports the decision
making process (e.g. specific configurations in the
hardware for complex medical procedures) as well
as innovation in medical planning and training.
In Section 2 we provide a brief introduction to
the field of radiation therapy and associated
hardware components. We also identify significant
related work in radiation therapy simulation and
training. Section 3 describes our solution and
provides details of implementation. In Section 4 we
focus on a very important component that allows us
to bring patient-specific information inside the
simulator to improve its realism and efficiency. In
Section 5 we present the preliminary assessment
implemented in collaboration with M. D. Anderson
Cancer Center, Orlando.
2 BACKGROUND AND RELATED
WORK
Radiation therapy is the careful use of high-energy
radiation to treat cancer. A radiation oncologist may
use radiation to cure cancer or for palliative
purposes. About 50 to 60 percent of cancer patients
are treated with radiation at some time during their
disease (
RadiologyInfo, 2006). Radiation destroys the
ability of cancerous cells to reproduce, and the body
naturally gets rid of these cells in time. A cancer
patient may be treated with radiation alone (e.g.
prostate cancer); however, sometimes radiation
therapy is only a part of a patient's treatment.
Patients can be treated with radiation therapy before
359
G. Hamza-Lup F., Sopin I., Lipsa D. and Zeidan O. (2007).
X3D IN RADIATION THERAPY PROCEDURE PLANNING.
In Proceedings of the Third International Conference on Web Information Systems and Technologies - Web Interfaces and Applications, pages 359-364
DOI: 10.5220/0001292003590364
Copyright
c
SciTePress
surgery, enabling a less radical surgery than would
otherwise be required. For example, some bladder
cancer patients can keep their bladder if radiation
therapy is effective.
A radiation oncologist may use the radiation
generated by a machine outside the patient's body.
This procedure is called External Beam Radiation
Therapy. Radiation also may be given with
radioactive sources that are injected in the patient;
procedure called brachytherapy. Our contribution to
treatment planning focuses on the External Beam
Radiation Therapy (EBRT).
2.1 Time and Safety in the Treatment
Process
EBRT involves advanced machinery that rotates
around the patient, following a carefully planned
schedule. The radiation delivery unit, called a linear
accelerator (LINAC), consists of three main moving
components: the gantry, the collimator, and the
table/couch, as illustrated in Figure 1. The center of
rotation of all these components is a virtual point in
space called isocenter.
Figure 1: Linear accelerator with on-board imaging.
The principle of operation for such devices is
fairly simple. Based upon the patient’s tumor
location, the hardware components change position
and orientation while delivering pre-planned
radiation doses to the patient on the table. However,
in some cases, the complex relative orientations of
all these components may cause a collision with the
components themselves or between the components
and the patient. In addition, LINAC head
attachments and patient immobilization devices may
also be a source of collision concerns. Therefore, a
radiation treatment planner may generate a
seemingly optimal plan, only to result in a collision
when a “dry run” (i.e. an execution of the plan
without radiation for collision checking) is
performed on the LINAC. This results in a delay to
the patient treatment, since the plan has to be revised
to account for these unforeseen collisions.
Additional time and resources must be invested to
adjust the existing or create an alternative treatment
plan.
Analytical methods for collision detection for
LINAC-based radiation surgery have been proposed
as a means to improve the EBRT planning process
(Beange and Nisbet, 2006; Hua, Chang, and Yenice,
2004; Humm, Pizzuto, and Fleischman, 1995;
Purdy, Harms, and Matthews, 1993). Analytical
methods, even though accurate, are based on the
hardware rotational and translational numerical
values disregarding patient-specific geometry.
Investigating previous work concerning graphical
simulations of the LINAC system, we have
identified the following limitations:
(1) sophisticated setups and additional software
and hardware components;
(2) The simulations involve only generic patient
body representations (Tsiakalos, Scherebmann, and
Theodorou, 2001); hence collisions with patients are
not accurately modelled or predicted;
(3) The simulations are local and can not be
deployed via the web for potential collaboration with
remote experts during treatment planning.
Researchers at the Hull Immersive Visualization
Environment proposed a VR environment for
training purposes (Beavis et al, 2006), using a virtual
patient based on the visible human female dataset.
Their prototype is targeted mainly at medical
training. Our efforts are directed towards a
distributed X3D-based system that will allow easy
access from a web browser to the virtual room and
will improve the actual planning process of the
EBRT by providing a high-resolution model of all
treatment components, including patient-specific
geometry.
2.2 Web-based Systems in Medical
Planning
Simulation-based Medical Planning has been
recently investigated for cardiovascular disease
(Steele et al, 2003). The Virtual Reality Modeling
Language (VRML) has been employed to provide
the visual web-based interface in the past. The
European Institute of Telesurgery has proposed a 3D
anatomical structure visualization and surgical
planning system that allows manipulation and
interaction on virtual organs extracted from CT-scan
or MRI data (Chirstophe, Luc, and Jacques, 2002).
With the advent of the X3D standard and its
extended functionality, the Internet-based systems
for simulation gained momentum. Currently, the
Medical Working Group, part of the Web3D
WEBIST 2007 - International Conference on Web Information Systems and Technologies
360
consortium, is taking steps in developing an open
interoperable standard for the representation of
human anatomy based on input from a wide variety
of imaging modalities. Our research and
development efforts presented in the next section
take into consideration and build on the existing
technology, available in the public domain, allowing
us to offer medical personnel free of charge access
to the X3D simulation and associated paradigms.
3 AN X3D TOOL FOR
COLLISION DETECTION
As we pointed out in Section 2.1, time-consuming
“dry runs” are necessary to avoid collisions during
the EBRT process. The “dry runs” do not take into
consideration the patient’s geometry. Depending on
the size of the patient, unforeseen collisions might
occur.
Providing a virtual 3D environment that will let
the medical personnel view the execution of the
treatment plan allows them to avoid potential
collision cases and even experiment with new
configurations. Our preliminary efforts in designing
such an environment (Hamza-Lup, Davis, and
Zeidan, 2006) provided a set of useful observations
and guidelines for the current implementation.
The simulator implementation takes advantage of
two technologies, Java and X3D. X3D (Web3D
Consortium, 2006) is an ISO standard with an open
architecture and a rich range of capabilities for real-
time graphics processing that is employed in a wide
array of domains and user applications. A successor
to VRML, X3D is being developed by the Web3D
Consortium as a refined standard (Lau et al, 2003).
Figure 2: X3D Environment (3DRTT).
Besides X3D, in the development process we
employ several software tools for 3D modeling.
Figure 2 illustrates a snapshot of the virtual room
(denoted 3D Radiation Therapy Treatment —
3DRTT) which models the real environment,
depicted in Figure 1.
3.1 User Interaction
The simulator provides an intuitive floating
graphical user interface (GUI) for controlling the
angles and locations of the machine’s parts
(presented in Figure 2). The user may rearrange the
GUI components to avoid important objects
occlusion.
Volumetric slides and scrolls keep controlling
operations simple and naturally fit in the 3D scene.
The user can also show/hide a patient or the
radiation beam generated by the collimator by
turning designated switches on/off.
Besides the mouse manipulation capability, the
user’s viewpoint can be easily changed on the
orthogonal axes. As a part of X3D, preset
viewpoints provide handy control and fast overview
potential and can be switched with a single
keystroke.
An important component of the simulator is the
ECMAScript (ECMA International, 1999), a
JavaScript-like language that introduces additional
functionality into X3D. The ECMAScript functions
provide the GUI’s interactivity. For instance, a user
shows or hides the patient or the beam by clicking
the appropriate buttons, or the values change when
the user rotates the gantry either by dragging it or
scrolling the associated control element. The
position of the object on the screen is synchronized
with the position of the corresponding scroll wheel.
Figure 3: Client-Server Interaction.
The X3D model containing the ECMAScript and
the optimized polygonal objects is embedded in a
web page that the user accesses from a local
machine. When the user submits new values on the
web page, a Java servlet processes them. The server
does not only generate a new web page, but also
deploys a new X3D model, as described in Figure 3.
Every user of the simulator is assigned a unique
session ID. Thus, clients using the same model avoid
conflicts in controlling the simulator. So, 3DRTT
X3D IN RADIATION THERAPY PROCEDURE PLANNING
361
has the distributed functionality of a 3D distributed
visual system on the web and can easily be shared
between researchers and remotely located medical
personnel.
3.2 3D Models Acquisition and
Processing
A 3D laser scanner from Faro™ Technologies is
employed to collect point clouds from several
viewpoints. Once the point clouds are collected, they
are merged into one cloud based on a set of specially
designated markers. We filter the noise and wrap the
valid points into a polygonal model.
To improve the rendering process, we decimate
the model by removing redundant polygons in flat
areas while securing a sufficient number of polygons
in the regions with complex geometry. The
polygonal model is exported into an X3D object and
employed as a part of the scene. All filtering,
wrapping, smoothing, and exporting operations are
performed using the built-in algorithms in the
commercially available Raindrop GeoMagic™
software.
Considering the geometrical complexity and the
high-polygonal resolution of the model, we have to
optimize the polygonal model such that adequate
frame-rates (25 FPS or more) are obtained on
machines with less rendering power. To improve the
rendering speed and reduce the file size, we are
using textures, simulating the geometry of complex
areas. Textures also save development time because
instead of processing a complex region we rather
substitute it with a texture that does not require a
long time to generate. Another positive side effect is
the network traffic reduction.
3.3 Collision Detection - Special
Collision Situations
Currently, the X3D standard supports only avatar
collision detection, meaning that users may “collide”
with virtual objects as they travel through the scene.
Only a few research groups attempted to
implement the object-to-object collision detection
directly in a VRML or X3D environment. For
instance, the V-COLLIDE library (Hudson et al,
1997) provides fast collision detection for arbitrary
polygonal objects based on the Robust and Accurate
Polygon Interference Detection (RAPID) algorithm
(Gottschalk, Lin, and Manocha, 1996) and is
potentially capable of computing collisions in
VRML. However, to the best of our knowledge,
none of these collision methods have been fully
implemented and tested in a VRML/X3D player.
Besides the visual check for collisions in
3DRTT, we are currently developing an automatic
collision detection system via bounding boxes that
approximates the shapes of virtual objects.
4 PATIENT 3D MODELS FROM
CT DATA
An important issue we address with this work is
real-patient data embedded in the X3D simulation.
We have to efficiently convert a set of CT scans of a
patient to a polygonal model of the patient’s body.
The set of CT scans used is stored using Digital
Imaging and Communication in Medicine (DICOM)
standard.
We process CT scans using the Visualization
Toolkit (VTK) (Schroeder, Martin, and Lorensen,
1996), a software system for 3D computer graphics,
image processing, and visualization. The first step in
our processing is to select from CT scans a volume
that eliminates the table on which the patient is
positioned. This operation reduces the number of
polygons obtained in our 3D patient model.
We apply the marching cubes algorithm
(Lorensen and Cline, 1987) with a value that selects
the isosurface of the patient’s skin. We used the
same value for the skin as it was done in (Lorensen,
2006). The marching cubes algorithm checks if the
corner values of a cube are above or below the
isosurface value. If some corner values are above
and some are below, then the isosurface obviously
intersects the cube, and we can compute the
intersection polygon by using a lookup table. The
lookup table provides a fast way to find the edges
intersected by the isosurface based on what corners
values are above and below the isosurface.
The application of the marching cubes algorithm
produces a polygonal model that includes the skin of
the patient as well as internal organs. We have
applied this algorithm on a set of 133 CT scans of a
patient torso. The number of polygons obtained
(approximately 500,000) significantly slowed down
the interaction with our web-based application. To
reduce the number of polygons, we apply a
decimation operation (Schroeder, Zarge, and
Lorensen, 1992) which reduces the number of
polygons to approximately 100,000. Decimation
works by evaluating the distance between each point
and the average plane of the triangles using the
point. If that distance is small, the point and all
WEBIST 2007 - International Conference on Web Information Systems and Technologies
362
triangles using it can be deleted. Obviously, the
“hole” left has to be re-triangulated. Once re-
triangulation is accomplished, we apply a Laplace
smoothing operation (Field, 1988) that modifies the
position of each vertex to be the average of the
neighbouring vertices.
To obtain a “clean” polygonal model that
contains only the skin of the patient (remember, we
are interested in visual collision checks between the
hardware and the patient), we apply a polygon
connectivity filter that further reduces the number of
triangle strips to about 3,000. A challenge in
applying this algorithm is specifying the connected
surface. Choosing the maximum connected surface
isolates the skin of a patient in most cases. The result
is illustrated in Figure 4.
Figure 4: 3D Torso Model from CT Scans.
A final step in the process is the conversion of
the 3D model into an X3D object. The X3D object is
embedded in the scene on the LINAC table in the
same position and orientation as the real patient
(Figure 5).
Figure 5: Patient-specific 3D model on the Web.
Now, the medical personnel can virtually execute
the treatment plan and check for collisions with the
patient body. Of course, the CT data is limited to a
specific body part (in this case the torso). We are
planning an improvement to this approach, i.e. to
generate a full body patient model from incomplete
CT scan data. We think that such an improvement
will help detect potential collisions with other
patient body parts, e.g. hands, legs, etc.
5 PRELIMINARY ASSESSMENT
We have deployed the system on a secure web site
and allowed medical personnel from M.D. Anderson
Cancer Center, Orlando, to remotely access the web-
based simulation environment. Their first subjective
reaction was that the X3D world is very realistic and
that it improves their confidence in running the
LINAC in a real scenario, since they can now
visualize the relative position of all the components
involved, including the patient.
To objectively test collision scenarios, we asked
a radiation therapy technician and a therapist to
simulate a plan that contains collisions among the
system components (illustrated in Figure 6). The
result was that the X3D simulation provides an
accurate representation of the LINAC (specifically,
the Varian 23iX LINAC) that can predict any
collision scenarios with centimeter accuracy.
Figure 6: Visual Collision Validation.
The preliminary assessment using visual
inspection from different angles, illustrated in Figure
6, provides an early validation for the accuracy of
the simulator. The tool can be used off-line by
planners to inspect their patient-specific beam
arrangements before simulation runs. We are in the
process of measuring the impact of the X3D
simulator on the EBRT planning and execution
efficiency.
6 CONCLUSIONS AND NEAR
FUTURE
We have presented the development of an X3D
Web-based simulator for External Beam Radiation
Therapy treatment simulation. The ability to
detect/predict a possible collision between all
LINAC components for a given patient eliminates
the need for backup plans and saves planning time.
In addition, it enables the planner to explore
X3D IN RADIATION THERAPY PROCEDURE PLANNING
363
different and unconventional gantry-couch-
collimator combinations for treatment that may give
rise to better quality plans.
We are developing a database, containing X3D
objects corresponding to various EBRT equipment
models, to meet the demands of a broader range of
medical facilities and applications.
We hope that our research and development
efforts will advance the quality of the radiation
therapy process and consequently will improve
cancer patients’ treatment and recovery.
ACKNOWLEDGEMENTS
Support for this work came from M. D. Anderson
Cancer Foundation. We thank Anne Ocheltree and
Chris Morabito from FARO™ Technologies
Orlando for their support in obtaining the high
resolution 3D models for the EBRT room.
REFERENCES
Beange, I., Nisbet, A., 2000. A collision prevention
software tool for complex three-dimensional isocentric
set-ups. In British Journal of Radiology, vol. 73, pp.
537-541.
Beavis, A., Ward, J., Bridge, P., Appleyard, R., Phillips,
R., 2006. An Immersive Virtual Environment for
Training of Radiotherapy Students and Developing
Clinical Experience. In American Association of
Physicists in Medicine, vol. 33, pp. 2164-2165.
Chirstophe, K., Luc, S., Jacques, M., 2002. PACS-based
interface for 3D anatomical structure visualization and
surgical planning. In Proc. SPIE — Medical Imaging
2002: Visualization, Image-Guided Procedures and
Display, vol. 4681, pp. 17-24.
ECMA International, 1999. ECMAScript Language
Specification. In Standard ECMA-262, vol. 2006, 3
ed. Geneva.
Field, D. A., 1988. Laplacian smoothing and Delaunay
triangulations. In Communications in Applied
Numerical Methods, vol. 4, pp. 709-712.
Gottschalk, S., Lin, M., Manocha, D., 1996. OBB-Tree: A
Hierarchical Structure for Rapid Interference
Detection. In SIGGRAPH '96: Proceedings of the 23rd
Annual Conference on Computer Graphics and
Interactive Techniques, New Orleans, Louisiana.
Hamza-Lup, F., Davis, L., Zeidan, O., 2006. Web-based
3D Planning Tool for Radiation-Therapy Treatment.
In Web3D 2006: 11th International Symposium on 3D
Web Technology, Columbia, Maryland.
Hua, C., Chang, J., Yenice, K., 2004. A practical approach
to prevent gantry-couch collision for linac-based
radiosurgery. In Medical Physics, vol. 31, pp. 2128-
2134.
Hudson, T., Lin, M., Cohen, J., Gottschalk, S.,
Manocha, D., 1997. V-COLLIDE: Accelerated
Collision Detection for VRML. In Proceedings of the
Second Symposium on the Virtual Reality Modeling
Language, Monterey, California.
Humm, J., Pizzuto, D., Fleischman, E., 1995. Collision
detection and avoidance during treatment planning. In
International Journal of Radiation Oncology, Biology
and Physics, vol. 33, pp. 1101-1108.
Lau, R., Li, F., Kunii, T., Guo, B., Zhang, B., Magnenat-
Thalmann, N., Kshirsagar, S., Thalman, D.,
Gutierrez, M., 2003. Emerging Web Graphics
Standards and Technologies — Web Graphics
Tutorial. In Computer Graphics and Applications, vol.
23, pp. 2-11.
Lorensen, B., 2006. Make Your Own Visible Woman,
[Online], Available:
http://www.crd.ge.com/esl/cgsp/projects/makevw
[9 Jan 2007].
Lorensen, W. E., Cline, H. E., 1987. Marching cubes: A
high resolution 3D surface construction algorithm. In
SIGGRAPH '87: Proceedings of the 14th Annual
Conference on Computer Graphics and Interactive
Techniques, Anaheim, California.
Purdy, J. A., Harms, W. B., Matthews, J. W., 1993.
Advances in 3-dimensional radiation treatment
planning systems: room-view display with real time
interactivity. International Journal of Radiation
Oncology, Biology and Physics, vol. 27, pp. 933-944.
RadiologyInfo, 2006. Introduction to Radiation Therapy
(Radiation Oncology), [Online], Available:
http://www.radiologyinfo.org/en/info.cfm?pg=intro_o
nco&bhcp=1 [10 Jan 2007].
Schroeder, W. J., Zarge, J. A., Lorensen, W. E., 1992.
Decimation of triangle meshes. In SIGGRAPH '92:
Proceedings of the 19th Annual Conference on
Computer Graphics and Interactive Techniques,
Chicago, Illinois.
Schroeder, W. J., Martin, K. M., Lorensen, W. E., 1996.
The design and implementation of an object-oriented
toolkit for 3D graphics and visualization. In VIS '96:
Proceedings of the 7th conference on Visualization
'96, San Francisco, California.
Steele, B., Draney, M., Ku, J., Taylor, C., 2003. Internet-
Based System for Simulation-Based Medical Planning
for Cardiovascular Disease. In IEEE Transactions on
Information Technology in Biomedicine, vol. 7, pp.
123-129.
Tsiakalos, M. F., Scherebmann, E., Theodorou, K., 2001.
Graphical treatment simulation and automated
collision detection for conformal and stereotactic
radiotherapy treatment planning. In Medical Physics
vol. 28, pp. 1359-1363.
Web3D Consortium, 2006. What is X3D? [Online],
Available: http://www.web3d.org/about/overview/ [10
Jan 2007].
WEBIST 2007 - International Conference on Web Information Systems and Technologies
364