Grand Challenges in Medical Modeling and Simulation

advertisement
Grand Challenges in Medical Modeling and Simulation
R. Bowen Loftin, Ph.D.
Professor of Electrical and Computer
Engineering
Professor of Computer Science
Director of Simulation Programs
Old Dominion University
Norfolk, Virginia
USA
bloftin@odu.edu
Introduction
The practice of medicine is thousands of
years old. Although medical education
training has made some use of “simulation”
(for example, the use of live animal models
and cadavers), only recently has the field
begun to embrace computer-based modeling
and simulation. This paper presents “grand”
and “not-so-grand” challenges in medical
modeling and simulation.
Two broad
categories of challenges are identified—
challenges that are relatively specific to the
medical domain and challenges that are
shared with other modeling and simulation
application domains.
Background1
“See one, do one, teach one.” This simple
set of phrases has characterized medical
education and training for over 4,000 years.
Today’s physician is largely the product of
an apprenticeship program that uses patients
in hospitals as the primary elements of the
“classroom”.
Little of the training
technology developed in the past century has
changed this historic process. During this
same century, however, we saw both the
This section was adapted from Bowen Loftin, “Med
School 1.0: Can Computer Simulation Aid Physician
Training,” Quest 5 (2), pp. 16-19 (July, 2002).
1
Executive Director
Virginia Modeling, Analysis & Simulation
Center
7000 College Drive
Suffolk, VA 23435 USA
757-686-6200 (voice)
757-686-6214 (fax)
www.vmasc.odu.edu
invention of the airplane and the maturation
of flight simulation as the primary training
tool for the aviator. Today, every military
and commercial pilot masters a new aircraft
in simulation. We have reached the point
where the best flight simulators are virtually
indistinguishable from the real thing. This is
not the case in medical simulation.
The foundations of medical modeling and
simulation were laid down over the past
twenty to thirty years. A major impetus in
transferring modeling and simulation
technology to the medical field has been the
U.S. military [Elliot, 1994]. During the Gulf
War, a senior U.S. Army medical officer
(COL Richard Satava, MD) realized that
many of the medical personnel under his
command (largely reservists recalled from
their civilian practices for the war) had little
recent experience with treating the type of
casualties typically found in war.
Fortunately, the Gulf War was an almost
bloodless one for the United States. COL
Satava, however, returned home convinced
that the same approach to training fighter
pilots and tank crews for the Gulf War
should be adapted to training (and refreshing
the skills of) military medical personnel.
Ultimately, the Defense Advanced Research
Projects Agency and, more recently, the
U.S. Army Medical Research and Materiel
Command [Moses, 2001] have developed
programs to create the needed technologies.
The Challenges2
At the outset it is important to note that
some of the challenges in medical modeling
and simulation are grand and some are not
so grand [Ota, 1995]. In addition, some of
the challenges described below are identical
to or reasonably represented by challenges
in other application domains. Below we
describe those “unique” and “grand”
challenges in some detail and mention other
challenges that have common elements with
those that are already well known.
Tissue Modeling
This author’s candidate for the “grandest”
challenge in medical modeling and
simulation is that of modeling human
tissues. The human body is composed of
many different tissue types and most
medical procedures involve interactions with
a number of different tissues. Moreover,
each tissue type is typically heterogeneous
and likely exhibits highly non-linear and
anisotropic behaviors. Finally, a successful
model or models must be computable in
real-time so that its responses to user inputs
are immediately available. [Chen, 2000]
Consider a simple incision. The scalpel
must pass through the skin (which itself is
comprised of several tissue types), through
adipose tissue, through muscle tissue (which
is highly anisotropic), and, perhaps, into or
through an organ. In the process the scalpel
will sever blood vessels, releasing blood (yet
another type of tissue). In a matter of a few
seconds, then, the scalpel may pass through
2
One portion of this section has been adapted from
R. Bowen Loftin and Mark Phillips, “’Smart’
Simulations for Medical Training and Planning”. In
Proceedings of the 2001 Western Multiconference.
more than ten tissue types of tissue, and, in
each case, the tissue response (visual, haptic,
olfactory, . . .) must be both appropriate to
the training objectives and occur with no
discernable (to the user) delay.
Physics-based approaches to tissue modeling
can produce realistic behaviors but are
unable to achieve the required real-time
performance. The gap between actual and
desired performance is so great that even
Moore’s Law will not, in the foreseeable
future, provide a solution. At present, we
have abandoned the effort to generate
complete physics-based models and, instead,
are tackling the problem by developing firstorder physics models that are “hybridized”
with mathematical models, based on
empirical data [Fung, 1993] and validated
through expert evaluation. Thus, these
models can be both computationally
tractable and “good enough” in the judgment
of experts. Even though this method can
provide a solution for some tissue types, we
have not been able to address the large
number of tissue types that may be
simultaneously or successively involved in
common medical procedures.
Multi-Modal Simulation
Another area of great challenge lies in the
need for medical simulators to be truly
mutli-modal. While it is certainly true that
many simulators employ non-visual sensory
displays (notably auditory and vestibular) to
augment their visual displays, useful
medical simulators may require visual,
auditory, haptic, and olfactory displays.
Assuming that the relevant display
technologies actually exist and provide
adequate fidelity, the challenge here lies in
the correlation of the different displays so
that the user perceives sensory cues in
proper relation to each other and to user
interactions. This problem is exacerbated by
the inherent variable (due primarily to
distance and ambient air movement) delay in
delivering an olfactory stimulus to the user
and the bandwidth limitations of current
haptic displays.
Transfer of
Environment
Training to
the
Clinical
This challenge of medical modeling and
simulation has been met in many other
domains. Why should it be regarded as a
grand challenge here? The answer lies in
the culture of medicine. In spite of many
efforts, few studies have examined the
degree to which simulator training transfers
to the clinical environment. Not only is this
difficult because of the typically subjective
nature of the evaluation of clinical
performance but it is also difficult because
the medical community is often resistant to
such evaluations. The reasons for this
resistance may be manifold. For example,
the innate conservatism of medicine brings
an imputed risk to new approaches to
training. Thus, the supervising physician
may be reluctant to allow a student (or a
resident or a fellow . . .) to perform a
procedure if his or her training has been
carried out solely via a simulator. This
raises a “catch-22” barrier in that physicians
may demand objective evidence of a
simulator’s efficacy as a prerequisite to the
simulator’s evaluation in a clinical setting.
Seamless Integration of the Virtual and Real
Medical schools are increasingly utilizing
“standardized patients” to train their
students in clinical skills. These “patients”
are actors/actresses trained to present the
symptoms of a variety of diseases.
Typically, the student is limited to a patient
interview and does not physically examine
the “patient” since they do not actually have
the disease. For example, a “patient” can be
trained to provide a history and describe the
symptoms of coronary artery disease.
However, the student cannot perform an
examination or administer tests that would
support a diagnosis since the patient does
not actually have the disease. The value of
this use of human “simulators” would be
greatly enhanced if the student could
seamlessly transition from a “real” human
that can be interviewed to a “virtual” human
(with an instantiation of the anatomical and
physiological indicators of a disease). This
can only be accomplished if one can rapidly
(and cheaply) generate a virtual copy of a
real person and “edit” that copy so that it
contains the proper anatomical and
physiological clues. [Kakadiaris, 1998]
Potential of Simulation as a Means of
Examining the Practitioner
The final challenge that is relatively unique
to medical modeling and simulation lies in
the potential of the simulator to be used as
an evaluator of competence on the part of
current practitioners. In general, once a
physician is licensed, there is no
requirement for periodically demonstrating a
given level of continuing competence in a
procedure.
Simulators can provide a
convenient and objective means of regularly
confirming a physician’s ability to perform a
given procedure at or above an agreed upon
level of competence. This is particularly
useful
for
infrequently
performed
procedures and for physicians who have
reached a certain age or who have
experienced medical problems that could
impair their performance.
“Common” Challenges
In this section we briefly note challenges in
medical modeling and simulation that are
(more or less) the same as those found in
other modeling and simulation application
domains.
Fidelity always rears its ugly head. As do
many other modeling and simulation user
communities, the medical community wants
all the fidelity it can get.
This
“requirement” is offered without supporting
data. While it is very likely that the training
of some (many, most, . . . ) medical
procedures demands the highest available
fidelity, at least some, especially basic,
skills, should be trainable with more modest
levels of fidelity. This is a topic as worthy
of research as it is in other application
domains.
Human-Simulator Interfaces are often, as in
other domains, critical. In order to avoid
negative training and to provide a
“complete” training solution, a medical
simulator designed to train a procedure, as
opposed to a basic skill, should have an
interface that replicates that of the real
world. Thus, the student should touch
instruments that provide the same look and
feel as their real world counterparts. If other
devices are used (for example, an
endoscopic camera), they should also be
replicated in the simulator.
This
requirement can be challenging, especially if
the simulator addresses open surgery where
the degrees of freedom and the numbers of
instruments are significantly larger than in
minimally-invasive surgery. [Baur, 1998]
In order to provide a spectrum of simulators
that spans the medical education and
training continuum, attention must also be
given to Interoperability. Unfortunately,
most medical simulators have been
developed independently with little or no
thought given to their integration into a
family of simulators designed to train a
complete procedure (for example, a
dissection simulator coupled to one for
suturing). Thus, as in other application
domains, there is a need for standardization
that
supports
interoperability
and
integration. Another element of integration
involves the Medical Curriculum. This
curriculum was developed before simulators
were available and does not explicitly
include them. Thought, therefore, must be
given to revising the medical curriculum to
properly exploit simulation.
Embedding Intelligence in simulators is not
a new idea. Clearly, given the value of
medical expertise and the demands of the
clinical environment on this expertise, there
is great value in embedding an expert
“coach” into many simulators.
Such
embedded coaches would allow students to
benefit from simulator-based training
without the need for an expert to be present
as a guide, mentor, and evaluator.
Data Representation is a particularly acute
issue for medical modeling and simulation,
as it is for other application areas. Raw data
is often volumetric in nature (from CAT or
MRI sources) but can also be photographic.
Models derived from this data have
historically been polygonal surfaces so that
rendering performance is maximized. More
recently, true volumetric data sets have
become popular [Bro-Nielsen, 1999].
Composability of medical models and
simulations poses a significant barrier to the
wider use of the technology for education
and training as well as for procedure
planning.
So long as each model or
simulation is “hand crafted”, cost (see the
next paragraph) and development time will
continue to be problems.
Cost, of course, is a major issue for medical
modeling and simulation. Although the
popular perception may be that the medical
enterprise is well funded, the reality,
especially in medical education, is quite the
opposite. The cost of simulators must be
significantly reduced if they are to become
commonly-available tools within the
medical school curriculum.
[Cotin, 1998] Cotin, S., Delingette, H., and
Ayache,
N.
“Real-Time
Elastic
Deformations of Soft Tissues for Surgery
Simulation.”
Technical Report 3511,
INRIA, October, 1998.
Conclusions
[Delingette, 1998] Delingette, H. “Towards
Realistic Soft Tissue Modeling in Medical
Simulation.”
Technical Report 3506,
INRIA, 1998.
In this position paper we have identified five
grand (or not-so-grand) challenges in the
field of medical modeling and simulation.
Chief among these is the problem of tissue
modeling. In addition, a number of other
challenges have been noted that are shared,
to a greater or lesser degree, with other
applications of medical modeling and
simulation. The medical domain clearly
provides a rich set of problems the solution
to which are of high value to both the
military and civilian sectors.
[Elliot, 1994] Elliot D, Combat Casualty
Care: Current Capabilities and Inadequacies
– A Surgeon’s Perspective. In Proceedings
of the Advanced Technology Applications to
Combat Casualty Care Workshop, pp. 1314, May 25-26, 1994, pp. 13-14.
[Fung, 1993] Fung, Y.C. Biomechanics:
Mechanical Properties of Living Tissues.
New York: Springer-Verlag, 1993.
References
[Baur, 1998] Baur, C., Guzzoni, D, and
Georg, O. “VIRGY: A Virtual Reality and
Force Feedback-Based Endoscopic Surgery
Simulator. In Proceedings of Medicine
Meets Virtual Realty 6, San Diego, CA,
January 28-31, 1998. IOS Press.
[Bro-Nielsen, 1999] Bro-Nielseen, M. and
Cotin, S.
“Real-Time Volumetric
Deformable Models for Surgery Simulation
Using Finite Elements and Condensation.”
In Proceedings of Medicine Meets Virtual
Realty76, San Francisco, CA, January 2831, 1999. IOS Press.
[Chen, 2000] Chen, D., Kakadiaris, I.,
Miller, M., B. Loftin and Patrick, C.
Modeling for plastic and re-constructive
breast surgery.
In Proceedings of the
Medical Robotics, Imaging and Computer
Assisted Surgery Conference, Pittsburgh,
PA, October 11-14, 2000.
[Kakadiaris, 1998] Kakadiaris, I.A. and
Metaxas, D. “3D Human Body Model
Acquisition
from
Multiple
Views.”
International Journal on Computer Vision
30 (3), pp. 191-218 (1998).
[Moses, 2001] Moses G, Magee J, H, Bauer
J, Leitch R. Military Medical Modeling and
simulation in the 21st Century. In
Proceedings of Medicine Meets Virtual
Reality 2001 Conference, pp. 322-328.
[Ota, 1995] Ota, D., Loftin, R.B., Saito, T.,
Lea, R. and Keller, J. Virtual Reality in
Surgical Education. Computers in Biology
and Medicine 25 (2), pp. 127-137 (1995).
Download