NIH Workshop Summary
Summary: “Modeling & Simulation in Medicine: Towards an Integrated Framework”, held at the National Library of Medicine on July 20-21, 2000.
Full final version of report published in Computer Aided Surgery Volume 6, Issue 1, 2001 -
Final report of the meeting Modeling & Simulation in Medicine: Towards an Integrated Framework , Gerald Higgins, Brian Athey, James Bassingthwaighte, James Burgess, Howard Champion, Kevin Cleary, Parvati Dev, James Duncan, Michael Hopmeier, Donald Jenkins, Christopher Johnson, Henry Kelly, Robert Leitch, William Lorensen, Dimitris Metaxas, Victor Spitzer, Nagarajan Vaidehi, Kirby Vosburgh and Raimond Winslow
The goal of the workshop was to define a vision for computer-based modeling and simulation in medicine, including diagnostics, therapeutics, research and training. This vision includes a definition of where the field should be in 10 years, and the identification of technical and scientific “next steps” within a 5-year timeframe. Objectives included an assessment of the current state of model development in medicine, and the creation of shared technical resources in which to develop future models and simulations, including a common modeling language. Emphasis was placed on the development of a consensus among leading researchers and clinicians about model integration from different levels of human biology and the need to embed physiological attributes within computer models of anatomy and cellular morphology. Significant participation was encouraged from potential end-users of modeling and simulation technologies. These included military medical trainers, surgeons, radiologists, experimental biologists, pharmaceutical researchers and others who might benefit from the development of more usable and coherent software applications.
The meeting started with presentations from invited speakers, including leading researchers in biomedical visualization, diagnostic imaging, medical simulation, physiological modeling, multi-scale engineering in biology and the Visible Human Project. The list of speakers and presentations is shown as Appendix A of this report.
The second half of the meeting focused on breakout discussions, including the following groups:
➢ Image-based models – The use of diagnostic images, such as those derived from MRI, CT, X-ray, ultrasound, and functional imaging (PET, SPECT), to develop computer models and simulations that exhibit physical and physiological realism.
➢ Multi-level visualization and multi-scale modeling – Approaches to connect the simulations and visualization of multiple levels of cells, tissues, and organs. These include engineering studies based on hierarchal frameworks, as well as visual depiction of hierarchal systems based on integration of images from multiple levels of resolution and a variety of imaging devices.
➢ Functional anatomy simulation – Computational models of physiology embedded within computer graphics models and diagnostic image-based models of anatomy, to enable more realistic and accurate simulations.
➢ Cells and subcellular systems – Models of cellular processes and the creation of a framework for post-genomic analysis.
➢ End-User Applications - End-user applications, with an emphasis on training medical personnel in the military and the use of distributed visualization and simulation in medical education.
The meeting ended with a general discussion of next steps in the process, including the need to develop working groups led by 1-2 people in each domain.
TEN YEAR VISION
The participants felt that the tremendous potential of computer-based modeling and simulation in medicine could be realized within ten years given a significant commitment of resources.
Specific objectives that could be realized include:
(1) Predictive modeling of specific biological systems, for applications such as medical therapeutics and biomedical research. For example, predictive models of generic cell types such as red blood cells, eukaryocytes and prokaryocytes could be used to screen the effects of novel drugs in pharmacological research. Similarly, patient-specific models could be used to predict the effect of novel pathogens on the respiratory systems of individual patients.
(2) Synthesis of diagnostic imaging, modeling and simulation with therapeutics in real-time to improve healthcare delivery. For example, matching of a patient’s diagnostic images during surgery with a probabilistic atlas of normal anatomy encompassing the range of human variation could enhance accuracy and improve outcome.
(3) Validated and accurate simulations of major organs, organ systems and inter-related organs integrating anatomy, physiology, biomechanical properties, cell biology and biochemistry. These should include integrated models of the vertical organization of some of the major organs (heart, lung, muscle) as well as horizontally-integrated models of major physiological systems (circulatory, respiratory, immune). Visualization and simulation technology should be available to move seamlessly between different spatial resolutions (molecular to organ level) and different temporal states (development through aging; varying physiologic state) within an integrated simulation (See Figure #1).
(4) Creation of ‘body-double’, patient-specific image models that will serve as a repository for diagnostic, pathologic and other medical information about a patient. These will serve as a three-dimensional (3-D) template for enhancing communication between patient and physician, and provide a reference framework to examine pathologic and age-related changes that occur over time.
(5) High fidelity medical simulation for training and accreditation. These simulators will support true user interaction with simulated human organs, including validated physical and physiological properties, such as real-time tissue deformability, realistic bleeding and accurate haptics (“touch and feel”). High bandwidth access will facilitate distributed visualization and simulation of models for medical education and research and development applications.
THE NEED FOR DEVELOPMENT OF COMMON TECHNICAL RESOURCES
The breakout groups developed specific recommendations in each domain, and most of the groups produced overlapping recommendations. The common recommendations for sharing technical resources included the following:
1. The need to form a coordinating team, working groups and other commonly shared resources such as a web site that can provide information about biomedical models, simulations and images. This may require professional staff and coordination that require additional funding. Several individuals felt that this would be best accomplished through the use of a private (non-governmental), non-academic, non-profit entity if possible.
2. The need to create databases of biomedical models, simulations and images, which can be shared by developers and accessed via a common web site(s). This may include the development of standardized, extensible model formats for applications such as sharing of statistical image-based models.
3. Use of open source architecture, code and models where possible, using the sourceforge.net resource as a model/template. Develop methods for sharing models and simulations while preserving the intellectual property rights of authors/developers.
4. Development of a common modeling language for creation of biomedical models. Current examples include AnatML, CellML and the Common Anatomical Modeling Language (CAML).
5. The need to identify existing sources of physical and physiological data for building more accurate and realistic models and simulations. Creation of a database of the biomechanical properties of tissues and organs, for applications such as medical simulation.
6. For some applications, it may be necessary to develop an animal model for gathering additional data and testing predictive models. Suggested species include the pig, because of the similarity in cardiovascular characteristics between pigs and humans, and the mouse, which is used extensively for genetic analysis.
7. Creation of a web-based journal with a focus on medical modeling and simulation in medicine. Although there is currently a plethora of scientific journals, no single resource is devoted to publication of high quality research papers in this domain.
FIVE YEAR OBJECTIVES
Near-term scientific and technical objectives were identified by individual breakout groups. Emphasis was placed on the identification of critical technologies that need to be developed as well as strategies for enhancing multidisciplinary cooperation in the development of models and simulations. The objectives, participants and mission are listed below for each of the breakout groups.
1. Image-Based Modeling
a. Group definition: Image-based modeling, development of geometric and physics-based models, including statistical and deformable models.
b. Members of the imaging group included:
➢ James Duncan, Yale University
➢ Simon Warfield, Brigham and Women's Hospital
➢ James Burgess, Fairfax Inova Hospital
➢ Kevin Cleary, Georgetown University
➢ Christos Davatzikos, Johns Hopkins University
➢ Sohan Ranjan, Image-guided Neurologics, Inc.
➢ Raju Viswanathan, Image-guided Neurologics, Inc.
c. Five year objectives include:
1. Develop approaches for sharing algorithms/results for both researchers and end users (with a focus on generic, commonly-used algorithmic modules)
2. Develop robust testing and validation databases
3. Develop unified reasoning strategies to put models in touch with data (i.e. integrate data-driven / model-driven)
4. Include pathology in modeling/recovery process (not just normal conditions)
5. Help develop techniques to extract material parameters from images to feed models
6. Emphasize clinical applications / testbeds
2. Multi-Level Visualization and Multi-Scale Modeling
a. Group definition: Connecting models and simulations at different space/time scales, and developing approaches for visualization of models and biomedical model hierarchies.
b. Members of the Visualization and Modeling group included:
➢ William Lorensen, GE Corporate Research
➢ Raghu Raghavan, I.E. Med
➢ Christopher Johnson, University of Utah
➢ David Lindisch, Georgetown University
➢ Rakesh Mullick, National Institutes of Health
➢ Donna Rounds, Physiome Sciences
➢ Victor Spitzer, University of Colorado
➢ Art Wetzel, Pittsburgh Supercomputing Center
➢ Brian Worley, Oak Ridge National Laboratory
➢ Terry Yoo, National Library of Medicine
c. Five year objectives include:
1. Extend current models and visualizations to their limits (and understand their limits and uncertainties)
2. Computational workbenches (integration, interaction, collaboration)
3. Use of Common Software Architectures (standards, plug and play, open source, languages)
4. Easier access to remote resources (the “Grid”)
5. Interactive visualization techniques for multi-level data
6. Customizable (“just for you – right now”) visualization
7. An animal model to test hypotheses – pig? mouse?
8. Define the interfaces between the models, software
3. Functional Anatomy Simulation
a. Group definition: Development of simulations of functional anatomy by integrating computational models of physiology with computer graphics and image-based models.
b. Members of the Simulation group included:
➢ Kirby Vosburgh, CIMIT (Harvard)
➢ Gerald Higgins, WashCAS
➢ Stephane Cotin, CIMIT (Harvard)
➢ David Deerfield, Pittsburgh Supercomputing Center
➢ Don Hilbelink, University of South Florida
➢ Kara Krause, Oak Ridge National Laboratory
➢ Mike Leventon, Agrose, Inc.
➢ Ken Lutchen, Boston University
➢ Dwight Meglan, Virtual Presence, Ltd.
➢ Tom McKracken, Visible Productions, Inc.
➢ Dmitris Metaxas, University of Pennsylvania
➢ Paul Segars, University of North Carolina
➢ Ben Tsui, University of North Carolina
c. Five year objectives include:
1. Standards for geometrical model databases
2. Standards for integration of models
3. Strategies for mapping function onto anatomy
4. Begin development of “suites” of hierarchal cardiothoacic models – channel, cellular, tissue – each has identified set of capabilities , publicly accessible and interactive
5. Development of good, publicly available segmentation tools
6. Models of patient variability
7. Detailed research agenda for material properties data in vivo
8. Development of modeling language for describing anatomy, physiology, mechanical properties – biological objects as software objects
9. Development of interfaces including haptics
4. Cells and Subcellular Systems
a. Group definition: Models of cellular processes and the creation of a framework for post-genomic analysis.
b. Members of the Cell Modeling group included:
➢ James Bassingthwaite, University of Washington
➢ Mike Hopmeier, Unconventional Concepts, Inc.
➢ Juan Cebral, George Mason University
➢ Tom Colatsky, Physiome Sciences
➢ Jenny Freeman, Agrose, Inc.
➢ Jay Snoddy, University of Tennessee
➢ Karin Willis, WashCAS
➢ Raimond Winslow, Johns Hopkins University
c. Five year objectives include:
1. Gene/protein > functional phenotyping - Need to choose selective topics
2. Develop databases with focus on:
➢ OORDB (object-oriented/relational)
➢ Technical standards
➢ Scientific standards
➢ Search engines for distributed databases
3. High throughput phenotyping from genomic intervention - How to select exemplary topics?
4. A few example models - Cell level, organ level
5. End-User Applications
a. Group definition: End-user needs in medical education, including distributed learning and military training.
b. Members of the Applications group included:
➢ Howard Champion, Tech Med, Inc.
➢ Larry Hettinger, Logicon Technical Services, Inc.
➢ Brian Athey, University of Michigan
➢ Sue Bogner, Institute for the Study of Medical Error
➢ Parvati Dev, Stanford University
➢ Robert Johnston, Cinemed, Inc.
➢ Robert Leitch, Combat Casualty Care
➢ Harvey Magee, Telemedicine and Advanced Technology Research Center, U.S. Army
➢ Victor Wong, University of Michigan
➢ Lynn Yaffe, ITT Research Institute
c. Five year objectives include:
1. Grand Challenge – Establish Human Simulation Project as a national/international strategic endeavor (Human Genome Model) – identify and include all the “key players”
2. Focus on studies of effectiveness of simulation on outcomes – transfer of trainin
3. Develop a set of peer-reviewed models
4. Centralized model and data repository
5. Define information, control, and scoring requirements and integrate in real-time
6. Develop a distributed, interactive simulation prototype