1. Testbed for Medical Cyber-Physical Systems
Insup Lee
PRECISE Center
Department of Computer and Information Science
University of Pennsylvania
NSF PI Meeting
October 17, 2013

2. Why Testbed for Medical CPS?
A typical medical systems brings together multiple facets of the CPS field
Physiology
Sensor technology
Networking
Dynamic changes
Human in the loop
MCPS Testbed is needed for development and evaluation of safe, secure, and effective clinically relevant MCPS technologies.
Create a MCPS Test Bed for the evaluation of safety, security, and effectiveness of a variety of clinically relevant technologies, such as networked medical devices, clinical workflows and executable guidelines, data fusion and learning algorithms, patient and caregiver models, etc.

3. MCPS Ingredients MCPS Clinical guidelines Patient modeling
Decision support
and smart alarms
Medical
devices
Caregiver
interfaces
Interoperability
protocols

4. Research Challenges in MCPS
Development of (new) medical devices
Medical device interoperability
Compositional safety platform
Ad-hoc assembly at bedside
Security & Privacy
Closed-loop control
Data-driven personalized care
Smart alarms & clinical decision support
Physiological closed-loop control
Safety Assurance
Usability
Certification
We are trying to develop high-confidence medical software systems using formal methods. We are working on interoperability platform that facilitates compositional safety. We are also developing smart medical alarms as well as closed loop control algorithms to manage medications that administer to the patients.
OS:
General challenges:
High-assurance medical devices
Interoperability between medical devices
Network introduces new patient hazards
Ad-hoc assembly at bedside
Security vulnerabilities
Usability challenges
Challenges in closed-loop control
Control design under extreme uncertainty
Safety of closed-loop systems
RM:
Patient models needed at appropriate abstraction levels
Realistic dynamics, parametric variability
Capable of sophisticated control-theoretic analysis
Safety in dynamically created scenarios
Compositional reasoning?
Safety case construction
Modeling of clinical scenarios
Workflows, requirements for devices, safety criteria
Sensor fusion of physiological sensors
Better reliability, faster detection
Safe operation with unreliable networks
Glucose
Meter
Network
Patient
Caregiver
Controller
Alert
Infusion pump
Network 4

5. Organization and Deployment
A collection of open-access repositories
Data repository
Physiological data and device traces
Patient model repository
Device repository
Device models and interfaces (virtual devices)
Remote access to web-enabled devices
Certification/regulatory artifacts
Regulatory submissions
Open-access computing infrastructure
Simulation of ICU/OR environment
The data and device models should be available for both research as well as education.
Data:both clean and dirty data with/without annotations, time synched, etc. Include context information (condition of patients, …)
Model: High level models and detailed models that can be parameterized for specific patients
The science of MCPS should include the certification dimension.
Certification
- Samples, standards

6. Testbed for Technology Components
Modeling and analysis tools and artifacts
Models of medical devices
Patient physiology
Clinical workflows
Medical device integration
Physical or virtual implementation of multi-device treatment scenarios
Real-time network/middleware infrastructure
Data analysis, mining, and visualization
Evaluation of individual projects
Cross-correlation of results between projects

7. Testbed for Cross-domain Fertilization
Access to caregivers for
clinically relevant problems
operational feedback
Access to regulators
Access to medical device industry
Community challenge problems
e.g., pacemaker challenge problem
Regulators
how to improve the process and/or help the process
what are relevant evidence and how to collect and use them

8. Acknowledgements John Hatcliff (KSU) Margie Fortino (UPHS)
Steve Patek (UVA)
PRECISE center members (Penn)

9. THANK YOU!