Get documents about "WG3
Shared by: lanyuehua
-
Stats
- views:
- 0
- posted:
- 9/5/2012
- language:
- English
- pages:
- 19
Document Sample
Working group 3: Patient Modeling
and Simulation
• Ruzena Bajcsy—UC Berkeley
• Scott L. Bartow—Senatra Home Care Services
• Amit Bose—Tyco Healthcare
• M.Cenk Cavusoglu—Case Western Reserve Univ.
• Robert C. Kircher—Dose Safety Company
• Douglas Rosendale—VA
• Charles Taylor—Stanford Univ.
• Russ Taylor—Johns Hopkins
• Harvey Rubin—Univ. of Pennsylvania
• David Arney–Univ. of Pennsylvania
Why develop patient models?
• Improved health care—outcomes, quality
• Better utilization of health care costs
– Prevention, intervention, maximal value of EHR
• More efficient device development
– Human studies are expensive
– Device manufacturers need models
– High entry barriers to developing specific models
• More effective procedure execution
– Planning, monitoring, and control
• Training/professional certification
• Patient education and guidance in clinical decision making
• Research
Convincing successes in other fields
confirm the value of modeling
product development
safety
cost effectiveness
regulatory approval
examples:
aerospace industry
chemical plants
automotive
Lessons learned
• Lesson 1
• 1 a.Models exist at 5 levels of spatial scale:
• Biochemical/genetic
• Cell
• Organs
• Whole body
• In society
• 1.b Each model evolves on temporal scale
• 1. c At each scale the models involve
hetergeneous structures and physical processes
Examples of “tools”
• Biochemistry/Genes/Cells
Physiome project,
DARPA BioComp
• Organs/whole body
ITK open source NIH
funded image processing
toolkit. “digital astronaut”
in planning stage
DARPA Virtual Soldier
Lesson 1 continued..
1.d Models are incomplete
Incomplete or non-existing mathematical models for physiological
processes
Insufficient parameters for most biological processes
Incomplete data sets: e.g. quantitative postoperative data not collected
1.d Models must be accessible
to the community of practioners—large and heterogeneous
to the community of investigators
to the community of device developers
to the community of regulators
1.e Models must accommodate “uniqueness” of each patient but
also must permit aggregation of populations
Lesson 2
Convincing preliminary data show that image
based modeling is effective
• at procedural level—training, outcomes (seizure
focus ablation, arrhythmia focus ablation,
interventional radiology-image guided biopsies,
radiation therapy mapping)
• clinically cost effective
• at commercial level—some systems are already
in use
Lesson 2.a
• Convincing preliminary data show that
physiology based modeling is effective
critical care
intra-operative
home care
• Convincing preliminary data show that
patient-in-society based modeling is effective
home care
institutional care
vaccine strategies
Lesson 3
Mechanisms to share data, models, tools, results are necessary
Challenges:
2.a Interoperability
2.b Institutional barriers to sharing data, tools
2.c Maintenance of Privacy
2.d Academic reward system
2.e Commercial reward system
Demonstration cases:
(2-5 yr*) Create "Knowledge Portal"
Build a foundation for open source environment
ontology
links to available models, data and device sources
protocols for validation
Build and distribute anatomical atlases
data exists—VA may be best source
combine information from multiple patients
generate coordinate system to “place” patient
searchable
generate statistical analysis
predict outcomes based on individual characteristics and
statistical
outcomes
device companies can project scales and sizes
Create protocol manual
detailed written descriptions of specific interventions
metrics for evaluation
Statistical Atlases of Patient Anatomy
Average model +
Multiple resolution variation modes
models
Statistical
Segmentation
Analysis
Anatomical Labels
Biomechanics Electronic
Anatomical
General Surgical Plans Atlas
Training Outcome data
Data Sets
APPLICATIONS
• Treatment planning, outcomes analysis, basic
research, …
R. Taylor & J. Yao
One Application: Bootstrapping Atlas
Average model +
Multiple resolution variation modes
models
Statistical
Segmentation
Analysis
Atlas-assisted Electronic
segmentation Anatomical
Atlas
Training
Data Sets
APPLICATIONS
• Treatment planning, outcomes analysis, basic
research, …
R. Taylor & J. Yao
Statistical Atlases of Physiology
Average model +
Analytical models variation modes
Signal X i F ( X , S , i ) Statistical X F ( X , S, )
processing Analysis
Dead Pig
Avg Force Magnitue: 11.59 N
0.34
0.32
Signal features
0.3
Red/IR
20 per. Mov. Avg. (Red/IR)
0.28
Ratio
0.26
0.24
Biology info Electronic
0.22
0.2
Atlas
400 600 800 1000 1200
Time
1400 1600 1800 2000
Lab data
Training Outcome data
Data Sets
APPLICATIONS
• Device design, treatment monitoring, planning,
outcomes analysis, basic research, …
R. Taylor & J. Yao
Fused Statistical Atlases
Average model +
Multiple resolution variation modes
models
Statistical
Segmentation
Analysis
X F ( X , S, )
X i F ( X , S , i )
Dead Pig
Avg Force Magnitue: 11.59 N
0.34
0.32
0.3
Red/IR
Anatomical Labels
20 per. Mov. Avg. (Red/IR)
0.28
Ratio
0.26
0.24 Lab data Fused Atlas
0.22
0.2
General Surgical Plans
400 600 800 1000 1200 1400 1600 1800 2000
Time
Training Outcome data
Data Sets
APPLICATIONS
• Treatment planning, outcomes analysis, basic
research, device design, control, …
R. Taylor & J. Yao
Another Application: Filling in
information
Patient-specific Patient-specific
images model
Atlas-assisted
segmentation
Dead Pig
Avg Force Magnitue: 11.59 N
0.34
0.32
0.3
Red/IR
Augmented
20 per. Mov. Avg. (Red/IR)
0.28
Ratio
0.26
models 0.24
Fused
0.22
0.2
400 600 800 1000 1200 1400 1600 1800 2000
Time
Electronic X i F ( X , S , i )
Atlas
APPLICATIONS
• Treatment planning, outcomes analysis, basic
research, …
R. Taylor & J. Yao
Research needs
• Understand abstraction
– domain specific
– technical fix
• Improved techniques for assessing clinically
relevant variability in measurements
• Experimental validation of models using:
ex vivo and bio-mimetic materials and
systems
animal models
clinical data
• Policy—privacy, security, legal, regulatory
Specific recommendations
• (2 yr*) common ontologies
descriptions of blood vessel branching for predicting
cardiovascular surgery outcomes
descriptions of activities of daily living for safe
performance in the home by the elderly
• (5 yr*) Statistical/analytical tools—
“on the fly” analysis of randomized trials
risk analysis – procedure/outcome,
statistical methods for characterizing variability,
abnormality, anatomical variance.
Specific recommendation
• (2-5 yrs*) Build teams for the production of high confidence medical devices:
work plan-
1) multidisciplinary academic and industry teams develops model
2) team does trials to validate model, publishes studies
3) FDA approves model for medical device validation
4) team maintains model
5) device manufacturer uses model for FDA submissions
Example: SRI / Stanford consortium with 7 medical device manufacturers
to develop model of femoral artery stent.
Consortium does data acquisition and modeling. Consortium publishes work,
can use for certification, companies buy in and get pre-publication data. Data
generated a redesign of stent testing methods and FDA using results in
regulatory process
Other examples:
Diabetes—insulin pump design
Chemotherapy-infusion/intralesional design
Pacemaker—control and validation
Long term oxygen therapy—delivery systems and monitoring
Recomendation: FDA, NSF, NIH, NIST, encourage public/private partnerships
academic/industry/government
Example: insulin pump device
Model-based Medical Device Software development
Plant
Software
Organ Sensor
model
Organ App
development
s
models
s/w
phase
Effector
Bio chem
processes
Plant
Sensor Medical Device
Sensor
Organ
Lab test
Organ
models
models
App
phase Effector
s/w
Bio chem Effector
processes
Medical Device
Clinical Organs
Sensor
trials Metabolic
App
s/w
processes
Effector
