SlideShare a Scribd company logo
1 of 71
Low Complexity System Designs for
Medical Cyber-Physical-Human Systems
(CPHS)
Po-Liang Wu
University of Illinois at Urbana-Champaign
Ph.D. Preliminary Examination, May 2014 1
2
 The size and complexity of cyber-physical systems have increased
significantly.
 “These advances have not only made it possible to reach the
frontier faster; they have also increased by orders of magnitude the
levels of complexity open to exploration and experimentation.
Understanding complexity and learning how best to harness these
new capabilities are both a challenge and a responsibility.”
NSF Strategic Plan, September 2006
Control & Harnessing Complexity of
Cyber-physical Systems
Introduction
 Medical systems are both safety-critical and highly complex and our
goal is to have a formally verifiable safe system in spite of the
complexity, using the approach recommended by National
Academic Science:
• “One key to achieving dependability at reasonable cost is a serious and
sustained commitment to simplicity, including simplicity of critical
functions and simplicity in system interactions. This commitment is often
the mark of true expertise”
 Thus, we must understand the sources of complexity and their
interactions in medical systems.
3
Goal: reduce and control complexity
Complexity control is to control the interactions between the system
components, for example, preventing the non-safety-critical components
propagates the faults to the safety-critical components and cause failure of
the whole system.
Motivation – Complexity and Interactions
4
High mental workload
increases the likelihood
of medical errors
The medical
errors may
cause severe
complications.
The
degradation of
patient
conditions
further
increases the
cyber-physical
complexity
Physical complexity: more medical devices must be added to
handle the complications, more treatments must be performed
Cyber complexity: more exception handling routines must be
performed, more device interactions must be checked, etc
Further increases the human mental workload.
Dimensions and Measurements of Complexity
 From a computer science perspective, complexity reflects the workload of
performing tasks by measuring number of steps or elapsed time.
 Computational complexity
• Worst-case time requirements as a function of the size of its input (most usefully, in
terms of the O-notation)
 Verification complexity
• Measures the state space to verify the system properties.
 Cyber medical treatment complexity
• Measures the number of steps to perform a treatment at run-time.
 Mental workload complexity
• Measures human memory and mental computation demand for performing tasks.
• It can be measured by task completion time, accuracy, NASA-TLX, etc.
 Other complexity
5
Major Sources of Complexity
 Sources of verification complexity
• Message interleaving due to asynchronous communication
• Inconsistency of distributed system states due to race conditions
Summary: The combinatorial possible execution sequences and
message interleaving must be checked.
6
Communication Patterns (Remote Procedure Call)
Supervisor Child device
C1
Async
C2
R1
Supervisor Child device
C1
Sync
C2
R1
R2
7
• Maximum flexibility
• Easily cause state space explosion
• Low complexity
• Too restrictive
State Space Explosion
8
msg queue
size
1 2 3
# states 2336 122394 N/A
(out of memory)
 Model checking state space
 Count total number of states
Node A Node B
msg3 msg2 msg1
Channel objectFSM object FSM object
Size = 3
msgA msgB msgC
Channel object
The number of states is the Cartesian product of the states of node A,
node B, and the communication channels.
In our experiment, the state space quickly explodes as the more
message interleaving is allowed.
Major Sources of Complexity
 Sources of cyber medical treatment complexity
• Medical staff need to validate preconditions, monitor side effects, and check
patient responses when performing a treatment
• Incorrectly performing a treatment may cause severe medical complications.
9
ActivateDefibrillator
Rhythm ==
Shockable
InjectEPI
BloodPH >
7.4
UrineFlow
> 12 mL/s
Inject-
SodiumBicarbonate
Airway &
Breathing
Treatment
Preconditions
Major Sources of Complexity
 Sources of cyber medical treatment complexity
• Medical staff need to validate preconditions, monitor side effects, and check
patient responses when performing a treatment
• Incorrectly performing a treatment may cause severe medical complications.
10
ActivateDefibrillator
Rhythm ==
Shockable
InjectEPI
BloodPH >
7.4
UrineFlow
> 12 mL/s
Inject-
SodiumBicarbonate
Airway &
Breathing
AssistedVentilation
The side effect
of bicarbonate
may adversely
affect patient’s
breathing.
Urine flow rate
may become
lower than 12
due to kidney
function
degradation.
IncreaseIVFluid
Major Sources of Complexity
 Sources of mental workload complexity
• Physicians and nurses are unable to mentally keep track of physiological
measurements minute by minute.
• Most physicians can mentally correlate a small number of variables at a
time but unable to correlate many variables in depth.
• Unorganized and lack of context-dependent information
» Context is a set of state variables, such as patient’s organ status and current medical
procedure.
» Context information is important for physicians and system to validate treatments and
trigger alarms.
Summary: Physicians must memorize and correlate tremendous
information, which significantly increases their mental workload.
State of the current medical practice
11
Background – Existing Complexity Reduction
Methodologies and Mechanisms
 Verification complexity
• Model checking techniques
» Abstraction
» Partial-order reduction
• System design
» Atomic Transaction/Serializability
» Physically Asynchronous Logically Synchronous (PALS) architecture
» Simplex Architecture
 Cyber medical treatment complexity
• Machine learning, expert systems
• Closed-loop supervisory control
 Mental workload complexity
• Integrated display systems to improve situation awareness
• GOMS(Goals, Operators, Methods, and Selection rules) model
12
Technical Challenges
 Concurrency related complexity
• C1. In CPHS, asynchronous communication is commonly used due to
interrupts and exceptions.
» Ideas: Communication protocol to limit the possible message interleaving but still
provide adequate flexibility.
» Preliminary result: Interruptible Remote Procedure Call protocol
• C2. Concurrent control of distributed devices and controllers result in
complicated race conditions.
» Ideas:
– Architecture design to separate the concerns of controllers
– Communication protocol to prevent conflict race conditions
» Preliminary result: Hierarchical organ-based architecture, Consistency protocol
• Summary: Proposing run-time control protocols to reduce unnecessary
or inconsistent interleaving due to concurrency.
13
Technical Challenges
 C3. Medical complications due to incorrectly performed
treatments.
• Ideas: Validate treatments in respect of preconditions, side effects,
and expected responses
• Preliminary result: Treatment Validation Protocol
 C4. Situation awareness improved display and assistance
system.
• Ideas:
» Integrated display of physiological measurements, treatment status, and
medical records.
» Workflow-driven context information.
• Work in progress: Workflow navigation and guidance system
14
Technical Challenges
 C5. Deviation from the best practice guidelines
• Ideas:
• Work in progress:
» A protocol to adapt the workflows when deviation is detected in
collaboration with medical staff.
» Source code verification of the developed protocol using CBMC, which is
a Bounded Model Checker for C/C++ programs.
15
Hard (safe-critical) deviation Soft (non-safety-critical)
deviation
System controllable Automatically forbid a. Stop once
b. Request for confirmation
c. Adapt the workflow
Not system controllable Flag the deviation and
disengage
a. Flag the deviation
b. Adapt the workflow
REDUCING VERIFICATION COMPLEXITY:
INTERRUPTIBLE RPC ARCHITECTURAL
PATTERN
16
H. Yun, P.-L. Wu, M. Rahmaniheris, C. Kim, and L. Sha. A reduced complexity design pattern for
distributed hierarchical command and control system. In Proceedings of the 1st ACM/IEEE ICCPS, pages
42-49. ACM, 2010.
Communication Patterns
Supervisor Child device
C1
Async
C2
R1
Supervisor Child device
C1
Sync
C2
R1
R2
Supervisor
C1
stop
Interruptible
R1partial
Child device
C2
stop
R2
17
• Maximum flexibility
• Easily cause state explosion
• Maximum complexity reduction
• Too restrictive
• Control asynchronous interactions in terms of mesage passings.
• Allow bounded asynchronous interrupt and make the system back to
synchronous
• The increase of state space is also bounded
Init
B1
Next
cmd!
F_Result?
Ready
Exe
cmd?F_Result!
stop?
B2
stop!
stop?
Supervisor Child Device
FSM Definition of Interruptible RPC Pattern
(Normal Case)
18
P_Result!cmd
P_R
stop
Init
B1
Next
cmd!
F_Result?
Ready
Exe
cmd?F_Result!
stop?
B2
stop!
stop?
Supervisor Child Device
FSM Definition of Interruptible RPC Pattern
(Interleaving Handling)
19
P_Result!
F_R
stop
Evaluation
 Simplified distributed command & control system
• One manager and one or more child devices
• Modeling using Maude, a model checking tool
 Evaluation Strategy
• Compared Maude models following different patterns:
synchronous, interruptible, asynchronous RPC
• Varied number of child devices in the model
• For asynchronous RPC, we also varied maximum queue size.
20
Evaluation
#child
devices
sync Interrup
tible
async(2) async(3
)
async(4) async(5
)
1 5 29 70 121 195 275
2 11 133 1410 4426 11816 23732
3 22 701 28691 184418 782417 N/A(*)
21
State space (# states)
#child
devices
sync Interrup
tible
async(2) async(3
)
async(4) async(5
)
1 0 1 4 7 12 275
2 1 13 160 549 1571 23732
3 1 120 6586 49307 1096480 N/A(*)
Model checking time (msec)
(*) not finished in 12 hours
Summary
 Distributed command & control system model
• Interleaving problem
• State space explosion problem
 Interruptible RPC design pattern
• Allow asynchronous interruption
• Reduces state space for model checking
 However, this work does not address exception handling issues.
• Child devices may actively send exceptions due to unexpected events,
hardware failure, etc.
22
REDUCING VERIFICATION COMPLEXITY: A
LOW COMPLEXITY COORDINATION
ARCHITECTURE
23
P.-L. Wu, W. Kang, A. Al-Nayeem, L. Sha, R. B. Berlin Jr, and J. M. Goldman. A low complexity
coordination architecture for networked supervisory medical systems. In Proceedings of the
4th ACM/IEEE ICCPS, pages 89-98. ACM, 2013.
W. Kang, P.-L. Wu, M. Rahmaniheris, L. Sha, R. B. Berlin, and J. M. Goldman. Towards organ-
centric compositional development of safe networked supervisory medical systems. In
Computer-Based Medical Systems (CBMS), 2013 IEEE 26th International Symposium on, pages
143-148. IEEE, 2013.
State of the Art Medical Systems
 Integrated Clinical Environment (ICE) infrastructure is logically
centralized control system.
• The medical devices are coordinated by a single ICE supervisor controller.
24
http://www.mdpnp.org/
Motivating Example –
A Laparoscopic Surgery
 A camera is inserted into the
patient’s abdomen and the CO2 is
pumped into the abdomen to create
a space for surgery.
 The ventilator can be programmed
to compensate the SpO2 level
within the limits set by physicians
• Such as tidal volume, respiratory
rate, FiO2
 Excessive CO2 pressure may
inhibit the breathing volume and
result in low SpO2 level.
25Extreme pressure would collapse the vena cava/hepatic vein and cause the shock.
CO2
pump
Monitor
scalpel
http://images.google.com
Inconsistent Views and Actions May
Cause Safety Hazard
 SpO2 sensor raises a patient
adverse events (PAE), caused
by SpO2 level lower than the
threshold set by the physicians
 ICE supervisory controller
requests to decrease CO2
pump pressure without knowing
the status of the surgical
process.
 Decreasing CO2 pressure
might cause insufficient
operating space and a safety
hazard.
 Multiple devices coordination is
essential to avoid safety
hazards
26
Laparoscopic Surgery
ICE
Supervisory
Controller
CO2
Pump
SpO2
Sensor
Oxygen
Saturation
Low
Decrease
Pressure
Video
Camera
Operating
Space
Safety
Hazard
Complexity of Asynchronous
Interaction
Hierarchical Layers Asynchronous Design
State space Time (s)
Two Layers 12084 3
Three Layers 603056 143
Four Layers Out of memory Out of memory
27
 Complex interaction in medical systems can easily occur due to
patient adverse events (PAE) and cause state space explosion.
Hierarchical Organ-system
Architecture
 Medical Devices Layer
• Medical devices report the
physiological measurements
and raise PAEs
 Organ System Control Layer
• It is responsible for stabilizing
an organ-system by
coordinating the devices.
• If a PAE can be locally
compensated, the organ-
system controller will handle it
• Otherwise, the PAE must be
propagated to the high-layer
controller.
28
SpO2 low
Cmd
Hierarchical Organ-system
Architecture
 Surgery and Anesthesia
Control Layer
• Surgical and anesthetic
controller are responsible for
inter-organ system
coordination.
• The anesthetic controller is the
primary controller for
anesthesia related events.
• The surgical controller focuses
on handling surgery related
events.
29
SpO2 low
Cmd
Hierarchical Organ-system
Architecture
 Benefits
• It naturally fits into the ICE infrastructure and enables localized PAE
handling. If PAEs can be handled by different subtrees, the higher level
controllers are not interfered.
• Capture human homeostasis. (An organ system can compensate
moderate abnormalities without interfering other organ systems)
 Engineering challenges
• Localized PAE handling may introduce message races to the controllers
of different layers.
• A communication protocol is required to guarantee consistency.
30
Consistent View Generation and
Coordination (CVGC) Protocol
• Design goals:
» The protocol ensures each controller has all its child nodes’ status.
» Avoid conflict race conditions
• The device setting is locked during the consistent view
generation to avoid inconsistency.
» Each device and controller has a lock to protect its setting.
• A controller can send commands to change the device
settings if and only if it holds all its child devices’ locks.
31
Like traditional locking mechanisms, such as tree locking, the
proposed protocol aims to prevent race conditions that may cause
safety hazards.
Unlike traditional locking mechanisms, the system needs to handle
both top-down and bottom-up message races.
Consistent View Generation and
Coordination (CVGC) Protocol
 The CVGC protocol has
three phases
• PAE propagation phase
• Locking and consistent
view generating phase
• Command execution and
unlocking phase
32
Ca
Cb
PAE1&CV&
LockSet
Consistent View Generation and
Coordination (CVGC) Protocol
 The CVGC protocol has
three phases
• PAE propagation phase
• Locking and consistent
view generating phase
• Command execution and
unlocking phase
33
Ca
Cb
PAE1&CV&
LockSet
Cc
lockAndReport
lockAndReport
lockAndReport message is used for locking a subtree and
gathering the consistent view.
A controller can send a command to change device settings if
and only if the lockAndReport is processes and the controller
receives the locks and consistent view.
Consistent View Generation and
Coordination (CVGC) Protocol
 The CVGC protocol has
three phases
• PAE propagation phase
• Locking and consistent
view generating phase
• Command execution and
unlocking phase
34
Ca
Cb
PAE1&CV&
LockSet
Cc
CV&
LockSet
CV&
LockSet
After processing the lockAndReport message and gathering
consistent view and locks, Cb can send a command to change
its child devices’ settings.
Race Condition Handling
 A race condition occurs when
lockAndReport message is propagated
downward while another PAE2 is
propagated upward.
 When Cc receives PAE2, its child
devices’ settings have been locked.
Therefore, Cc does not need to further
propagate the lockAndReport message.
 Cc sends PAE2&CV&LockSet to Cb.
 Cb has the consistent view and all the
locks.
 How to handle multiple PAEs is pre-
specified by the physicians.
35
Ca
Cb
PAE1&CV&
LockSet
Cc
PAE2&CV&
LockSet
lockAndReport
Consistent View Generation and
Coordination (CVGC) Protocol
 The CVGC protocol has
three phases
• PAE propagation phase
• Locking and consistent
view generating phase
• Command execution and
unlocking phase
36
Ca
Cb
Cc
Cmd
CmdCmd
Cmd
If Cb can handle PAE, it sends the command to change the
device settings and release the locks.
Correctness of the CVGC Protocol
 Model checked architectural properties
• CVGC protocol is deadlock free.
• A controller can send a command iff it receives the consistent view and
the locks.
• The locking mechanism prevents multiple controllers command the
same device simultaneously.
 Timing analysis
• µ: message transmission delay
• WCET: worst case execution time
37
Complexity Evaluation
 Verified safety properties
• The pulmonary controller will increase the respiratory rate of the
ventilator when the SpO2_low PAE is received if the respiratory rate is
below the maximum rate. (Layer 2)
• If the surgery is at a critical point, the supervisory controller should not
allow a reduction of the laparoscopic CO2 pump pressure. Instead, an
alarm must be issued to the physicians. (Layer 4)
38
Hierarchical Layers Metrics Proposed Design Asynchronous
Two Layers Number of states 1887 12084
Checking time (s) <1 3
Three Layers Number of states 10651 603056
Checking time (s) 2 143
Four Layers Number of states 21888 Out of memory
Checking time (s) 8 Out of memory
Summary
 Reducing verification complexity by proposing:
• Hierarchical organ-system architecture
» Capture human homeostasis
» Allow distributed patient adverse events handling
• Consistent view generation and coordination protocol
» Gather concise and comprehensive physiological information
» Low complexity compared to asynchronous design.
39
REDUCING CYBER MEDICAL TREATMENT
COMPLEXITY: A TREATMENT
VALIDATION PROTOCOL
P.-L. Wu, D. Raguraman, L. Sha, R. B. Berlin, and J. M. Goldman. A treatment validation
protocol for cyber-physical-human medical systems. To appear In Software Engineering and
Advanced Applications (SEAA), 2014 EUROMICRO Conference on.
40
Introduction and Technical Challenges
 Timely and correctly perform treatments while monitoring patient
responses is critical for patient safety.
 Treatment validation is essential and consists of three aspects
• Precondition
» A treatment can be performed only if the preconditions are satisfied
• Potential side effects
» The side effects of a treatment may adversely affect other treatments or
invalidate previously satisfied preconditions
• Expected patient responses
» The effectiveness of a treatment is non-deterministic
 The purpose of this work is to assist medical staff to correctly
perform treatments rather than automatically performing treatments.
41
Like traditional cyber validation mechanisms, the proposed protocol
guarantees certain preconditions are satisfied before taking actions.
Unlike traditional cyber validation mechanisms, the proposed protocol
may not have the full control of physical components, such as patient
conditions.
Cardiac Arrest Resuscitation –
State of the Art Practice
42
Supervisory System
User Interface
Defibrillator
Airway
Breathing
satisfactory
Rhythm
Non-
Shockable
1. Activate
defibrillator
2. Check
Airway &
EKG Rhythm
Epinephrine
BloodPH
< 7.4
Urine <
12 mL/s
3. Inject
epinephrine
4-a. Check
preconditions
4-b. Request
corrective
treatments
Performing Treatments
Sodium Bicarbonate
6. Post order execution
Sodium Bicarbonate
No Calcium Chloride
Increase IV Fluid
5. Sodium
bicarbonate &
Increase IV
Airway
Breathing
is affected
BloodPH <
7.4
Sodium Bicarbonate
No Calcium Chloride
7. Side effects
adversely affect
preconditions
8. Patient
responses not
as expected
9. Exceptions 10. Alternative
Treatment
Assisted
Ventilation
43
Correctness
 Theorem 1. (Preconditions satisfaction) Under the proposed
protocol, a treatment is performed only if all preconditions of the
treatment are satisfied.
 Definition: A tree is well-formed if each unsatisfied precondition have
a tree node for correcting it.
 Theorem 2. (Tree Traversal) The root node of a well-formed tree is
reachable if the corrective treatments are effective and the
preconditions are not invalidated by the side effects.
 Theorem 3. (Dynamic adaptability) Suppose side effects of a
treatment invalidate any precondition and make the tree become non-
well-formed. The protocol updates the tree to be well-formed if the
medical staff correctly specifies the corrective treatments.
44
Verification
Medical safety
properties
P1: Defibrillator is activated only if the rhythm is shockable and airway
and breathing is under control.
P2: Epinephrine is injected only if the blood pH value is larger than 7.4
and urine flow rate is higher than 12 mL/s.
P3: If the side effect of sodium bicarbonate adversely affects the
breathing, the tree is updated with a new treatment node for assisted
ventilation.
Protocol correct
ness properties
P4: There is no deadlock in the system.
P5: A treatment is performed only if all its preconditions are satisfied.
P6: If side effect does not occur, the root node of the tree is added to
the executing list
P7: If side effects invalidate a precondition, the tree is updated to be
well-formed.
45
Summary
 In order to reduce cyber medical treatment complexity, a validation
protocol is proposed to
• Check preconditions
• Monitor potential side effects
• Check expected responses
 We use a model checking tool to verify both safety and correctness
properties.
46
OUR FUTURE PLAN
47
REDUCING MENTAL WORKLOAD
COMPLEXITY:
WORKFLOW NAVIGATION AND GUIDANCE
DISPLAY SYSTEM FOR IMPROVING
SITUATION AWARENESS
48
State of Current Research
 (Main focus) Lack of context-dependent information and lack of
adaption ability to the dynamic context changes
 Context-dependent information
• Context is a set of state variables, such as patient’s organ status and
current medical procedure.
• Context information is important for physicians and system to validate
treatments and trigger alarms.
 Our system organizes the context information and adapts to the
context changes according to the medical workflow
49
Medical Workflow
 Medical workflow intends to capture the best practice to assist
medical staff to diagnose and treat patients.
 American Heart Association (AHA) provides a family of resuscitation
workflows for different life-threatening heart rhythms, including
ventricular fibrillation, asystole and tachycardia, etc.
 The purpose of the developed system is to assist medical staff to
synchronize the team actions to the workflow in real-time by
• codifying the best practice medical workflows
• keeping track of the workflow steps
• flagging deviation from the workflow
• highlighting the real-time progress of the treatments, the status of the
treatments, and the patient’s physiological responses
50
Organ system states
• Asystole
• Tachy.
• …
• Asystole WF
• Tachy WF
• …
Codified workflows
• HR
• BP
• ECG rhythm
• …
Physiological
measurements
51
Drug
guidelines
Treatment
workflow
Workflow Manager
Medical Device
Configuration Management
Resuscitation Monitoring System
Device Interoperability
Head Nurse
52
Situation Awareness Display
Treatment validation
Workflow Configuration
Physicians
The images are from http://www.pacificu.edu/pa/
Microsoft Surface
Evaluation: Comparison
 Vital signs only display
 Vital signs + workflow
 Vital signs + workflow + treatment validation status
 Compare to different iterations of our design
• Features added in iterations
» Automatically generate medical records
» Confirmation of drug order and given
53
Evaluation: Candidate Metrics
 Response time to a simulated event
 Accuracy(# of errors) of the performed actions
 Memory demand
 NASA-TLX
 Other questionnaire
• Proper abstraction of workflows
• Usefulness of treatment validation
• Meaningfulness of the alarms
54
Deviation from the Workflow
Hard (safety-critical)
deviation
Soft (non-safety-
critical) deviation
System controllable Automatically forbid a. Stop once
b. Request for
confirmation
c. Adapt the workflow
Not system controllable Flag and disengage a. Flag the deviation
b. Adapt the workflow
55
• Adapt the workflow after the deviation is detected in collaboration
with physicians.
• Simply switch to a new workflow may not work.
• For example, patient is injected certain drug that will cause severe
adverse interactions with another drug required in the new workflow.
Source Code Verification
 In my previous work, verification is performed at model design level.
 In order to guarantee correctness and safety of the developed
system, verification must also be performed at source code level.
 We are currently re-engineering the system architecture to separate
safety-critical and non-safety-critical software components.
• For the safety-critical components and their interaction protocols, we will
use CBMC, which is a Bounded Model Checker for C/C++ programs, to
fully verify the safety and correctness properties.
• For the non-safety-critical components, we will develop test suites to
increase code coverage.
• The safety-critical components only use but not depend on the non-
safety-critical components.
56
Published Work and Future Plans
 Published work
• P.-L. Wu, W. Kang, A. Al-Nayeem, L. Sha, R. B. Berlin Jr, and J. M. Goldman. ‘A low
complexity coordination architecture for networked supervisory medical systems.” In:
Proc. ICCPS’13.
• P.-L. Wu, D. Raguraman, L. Sha, R. B. Berlin, and J. M. Goldman. “A treatment
validation protocol for cyber-physical-human medical systems.’ To appear in
EUMIRCRO SEAA’14.
• H. Yun, P.-L. Wu, M. Rahmaniheris, C. Kim, and L. Sha. “A reduced complexity design
pattern for distributed hierarchical command and control system.’ In: Proc. ICCPS’10.
• W. Kang, P.-L. Wu, M. Rahmaniheris, L. Sha, R. B. Berlin, and J. M. Goldman.
“Towards organ-centric compositional development of safe networked supervisory
medical systems.’ In: Proc. CBMS’13.
 Plans for publication
• “A workflow navigation and guidance system for improving situation awareness”,
targeting CHI’15
• “An adaptive protocol for handling deviations from the medical workflows”, targeting
ICCPS’15
57
Conclusion
 Reduce and control complexity in respect of
• Verification complexity
• Cyber medical treatment complexity
• Mental workload complexity
 We have developed
• interruptible RPC and consistent coordination protocol to reduce
verification complexity.
• treatment validation protocol to reduce treatment complexity
 We are developing workflow navigation and guidance system for
reducing mental workload complexity.
58
Additional Slides
59
Model-based Engineering
 Model-based engineering is a system development methodology
which exploits the reusable models in order to reduce complexity.
 However, reusing a model during system integration requires
• the assumptions or preconditions of each integrated model are fully
checked and satisfied
» For example, Ariane 5's accident was due to the violation of assumption of
maximum horizontal speed.
• the interactions of the models are thoroughly verified against potential
unsafe interactions
» For example, the surgical fire during an airway laser surgery is due to unsafe
interactions between laser scalpel and ventilator.
60
Design Patterns for Workflow
Navigation and Guidance Systems
 We identified several design patterns
• Real-time treatment: Certain treatments must be continued or separated
for a time interval
» For example: CPR must continue for at least 2 minutes and epinephrine
should be given every 3~5 minutes
• Physiological conditions: the preconditions, side effects, and patient’s
responses of the treatments can be described as physiological
conditions, such as blood pressure > 120
 In addition, we provide verifications for both model design and C
code implementation levels using UPPAAL and CBMC (Bounded
Model Checker for C/C++).
61
62
Interleaving Problem
 Model ohecking
 R: in case of no pending msg, MSMGR, SSMGR and SS must be in the same state
(track or scan)
Maude> search (init,off,off,none,none,none,none) =>+ (targeting,airscan,airscan,none,none,none,none ) .
Solution 1 (state 136)
states: 137 rewrites: 261
Maude> show path label 136 .
MM_takeoff
MM_takeoff_waypoint
MM_waypoint_patternsearch
SM_off_airscan
MM_patternsearch_abort
SS_off_airscan
SS_airscan_airtrack
SM_airscan_airtrack
SM_any_off
MM_abort_waypoint
MM_waypoint_patternsearch
MM_patternsearch_targeting
SS_any_off
SM_off_airscan
SS_off_airscan
MSMGR SSMGR SS
waypoint
abort
waypoint
pattern
targetting
airscan
airscan
airtrackairtrack
off
airscan airscan
inconsistency
scanAir
scan_air
track
track scanAir
off
stop
stop
UPPAAL Model for Pulmonary
Controller
63
UPPAAL Model for Pulmonary
Controller
64
PAE propagating
Cmd executing and unlocking
Locking and consistent view generating
phase
System Models and Definitions
 PhysiologicalCondition (PC) is defined as a tuple < Checker, PM, Operator, RV
>, where
• Checker is the entity capable of checking the, Checker ∈{System, MedicalStaff},
• PM (physiological measurement) ∈ {EKGRhyhtm, HeartRate, BloodPressure...},
• Operator ∈ {>, <, =, ≤, ≥},
• RV is the reference value, which can be threshold, trend, or pattern of the
physiological measurements.
 A treatment is defined as a tuple < Agent, Action, PS, SS, ES, L >, where
• Agent is the entity that performs the treatment, Agent ∈ {MedicalDevices,
MedicalStaff},
• Action is the set of executable instructions,
• PS is a set of preconditions that must be satisfied before preforming the Action
• SS is a set of potential side effects
• ES is a set of expected responses after performing the treatment
• L is the life cycle, which specifies the time interval between the treatment being
performed and the treatment has no further effect on the patient
65
Major Mental Workload Complexity Sources – Using
Cardiac Arrest Resuscitation as a Case Study
 Most of medical records taken during the procedures are paper-based
and not electronically integrated with the Electronic Medical Records
(EMR) in real-time.
• # of spoken words, questions raised
 Medical staff must memorize the workflow
• # of medical errors (reviewed by an expert)
• Response time to a simulated event
 No validated treatments information
• # of medical errors
 Medication orders are verbally communicated
• # of spoken words, questions raised
 Task completion time, NASA-TLX can serve as general metrics for
mental workload complexity.
66
State of Current Practices
 Single-Sensor-Single-Indicator is not enough: "State of the art"
medicine practices usually focus on one control loop (one
measurement, HR/BP, and one contributor, drug, disease, etc).
 Lack of integrated and well-organized display: Related
information are usually scattered across multiple devices and
screens. This situation significantly increases medical staff's mental
workload and the chance of preventable medical errors.
 Some research work has addressed the above problems by
providing integrated display.
67
Interactive Situation Awareness
Monitor for Physicians
 Reduce the large volume of
medical information to a
small set of concise and
comprehensive information
according to the chosen
workflow
68
Order Management Tablet
(Handheld) for Nurse in Charge
 Transform the concurrent, asynchronous
and chaotic order management environment
into an orderly, safe, and fault-tolerant one.
Ensure each issued order is
• consistent with the preconditions, the timing,
workflow, patient physiological status
• consistent with the available equipment, and
personnel (to be implemented)
• interoperable with hospital EMR (to be
implemented)
 The tablet interacts with our backend system
and the situation awareness monitor
 In the event of system or device failure, the
order management tablet has specified
safety functions.
69
Microsoft Surface
Virtual Patient Station for
Training and Evaluation
 To support pre-recorded data
mode
• The EKG signal data are
obtained from PhysioNet.
• Different heart rhythms can be
selected at anytime by the
physician who is responsible
for simulating patient
responses.
• Patient vital signs, including
heart rate, blood pressure, etc.,
can be selected as well.
• Physicians, residents, and
nurses will “treat” the virtual
patients as if they were real.
70
Situation Awareness Monitor Layout
Sensor readings
Confirm rhythm
Treatments
EKG signal
Treatment timelines
Medication
Messages
Log
Current Workflow
71

More Related Content

What's hot

London Ambulance Problems
London Ambulance ProblemsLondon Ambulance Problems
London Ambulance Problems
KennyBHS
 

What's hot (6)

London Ambulance Problems
London Ambulance ProblemsLondon Ambulance Problems
London Ambulance Problems
 
Chapter13 -- ensuring integrity and availability
Chapter13  -- ensuring integrity and availabilityChapter13  -- ensuring integrity and availability
Chapter13 -- ensuring integrity and availability
 
Las Failure
Las FailureLas Failure
Las Failure
 
Siegel - keynote presentation, 18 may 2013
Siegel  - keynote presentation, 18 may 2013Siegel  - keynote presentation, 18 may 2013
Siegel - keynote presentation, 18 may 2013
 
Dynamic RWX ACM Model Optimizing the Risk on Real Time Unix File System
Dynamic RWX ACM Model Optimizing the Risk on Real Time Unix File SystemDynamic RWX ACM Model Optimizing the Risk on Real Time Unix File System
Dynamic RWX ACM Model Optimizing the Risk on Real Time Unix File System
 
Ieeepro techno solutions 2013 ieee embedded project an integrated design fr...
Ieeepro techno solutions   2013 ieee embedded project an integrated design fr...Ieeepro techno solutions   2013 ieee embedded project an integrated design fr...
Ieeepro techno solutions 2013 ieee embedded project an integrated design fr...
 

Viewers also liked (6)

Understanding Complexity of Organizational and System Dynamics
Understanding Complexity of Organizational and System DynamicsUnderstanding Complexity of Organizational and System Dynamics
Understanding Complexity of Organizational and System Dynamics
 
Complexity of organizational design and its effect scaling agility
Complexity of organizational design and its effect scaling agilityComplexity of organizational design and its effect scaling agility
Complexity of organizational design and its effect scaling agility
 
Motivation and theories of motivation
Motivation and theories of    motivationMotivation and theories of    motivation
Motivation and theories of motivation
 
Motivation and Learning
Motivation and LearningMotivation and Learning
Motivation and Learning
 
Motivation
MotivationMotivation
Motivation
 
Motivation ppt
Motivation pptMotivation ppt
Motivation ppt
 

Similar to Complexity Reduction for Cyber-Physical-Human Medical Systems

Published Research Paper
Published Research PaperPublished Research Paper
Published Research Paper
Yuxuan Huang
 
A survey of online failure prediction methods
A survey of online failure prediction methodsA survey of online failure prediction methods
A survey of online failure prediction methods
unyil96
 
Hospital Management System
Hospital Management SystemHospital Management System
Hospital Management System
idowume
 
Ensuring the feasibility of a $31 million OR expansion project: Capacity plan...
Ensuring the feasibility of a $31 million OR expansion project: Capacity plan...Ensuring the feasibility of a $31 million OR expansion project: Capacity plan...
Ensuring the feasibility of a $31 million OR expansion project: Capacity plan...
SIMUL8 Corporation
 
Composite Intrusion Detection in Process Control Networks
Composite Intrusion Detection in Process Control NetworksComposite Intrusion Detection in Process Control Networks
Composite Intrusion Detection in Process Control Networks
guest8fdee6
 

Similar to Complexity Reduction for Cyber-Physical-Human Medical Systems (20)

Low Complexity System Designs for Medical Cyber Physical Human Systems
Low Complexity System Designs for Medical Cyber Physical Human SystemsLow Complexity System Designs for Medical Cyber Physical Human Systems
Low Complexity System Designs for Medical Cyber Physical Human Systems
 
computers in clinical development
 computers in clinical development computers in clinical development
computers in clinical development
 
111 118
111 118111 118
111 118
 
Published Research Paper
Published Research PaperPublished Research Paper
Published Research Paper
 
Cost and accuracy aware scientific workflow composition for service oriented ...
Cost and accuracy aware scientific workflow composition for service oriented ...Cost and accuracy aware scientific workflow composition for service oriented ...
Cost and accuracy aware scientific workflow composition for service oriented ...
 
Developing bottom up devices and advanced manufacturing processes - nov 29 20...
Developing bottom up devices and advanced manufacturing processes - nov 29 20...Developing bottom up devices and advanced manufacturing processes - nov 29 20...
Developing bottom up devices and advanced manufacturing processes - nov 29 20...
 
Mc calley pserc_final_report_s35_special_protection_schemes_dec_2010_nm_nsrc
Mc calley pserc_final_report_s35_special_protection_schemes_dec_2010_nm_nsrcMc calley pserc_final_report_s35_special_protection_schemes_dec_2010_nm_nsrc
Mc calley pserc_final_report_s35_special_protection_schemes_dec_2010_nm_nsrc
 
A survey of online failure prediction methods
A survey of online failure prediction methodsA survey of online failure prediction methods
A survey of online failure prediction methods
 
Resilience and recovery
Resilience and recoveryResilience and recovery
Resilience and recovery
 
Cybernetics in supply chain management
Cybernetics in supply chain managementCybernetics in supply chain management
Cybernetics in supply chain management
 
Hospital Management System
Hospital Management SystemHospital Management System
Hospital Management System
 
Human Error & Risk Factor Affecting Reliability & Safety
Human Error & Risk Factor Affecting Reliability & SafetyHuman Error & Risk Factor Affecting Reliability & Safety
Human Error & Risk Factor Affecting Reliability & Safety
 
Detection of Wastewater Pollution Through Natural Language Generation With a ...
Detection of Wastewater Pollution Through Natural Language Generation With a ...Detection of Wastewater Pollution Through Natural Language Generation With a ...
Detection of Wastewater Pollution Through Natural Language Generation With a ...
 
Comp10 unit6c lecture_slides
Comp10 unit6c lecture_slidesComp10 unit6c lecture_slides
Comp10 unit6c lecture_slides
 
Profile tulasi v1.1
Profile tulasi v1.1Profile tulasi v1.1
Profile tulasi v1.1
 
Smart Traffic Lights that Learn ! Multi-Agent Reinforcement Learning Integ...
Smart Traffic Lights that Learn !    Multi-Agent Reinforcement Learning Integ...Smart Traffic Lights that Learn !    Multi-Agent Reinforcement Learning Integ...
Smart Traffic Lights that Learn ! Multi-Agent Reinforcement Learning Integ...
 
Ensuring the feasibility of a $31 million OR expansion project: Capacity plan...
Ensuring the feasibility of a $31 million OR expansion project: Capacity plan...Ensuring the feasibility of a $31 million OR expansion project: Capacity plan...
Ensuring the feasibility of a $31 million OR expansion project: Capacity plan...
 
Cps security bitsworkshopdec15.2012 (1)
Cps security bitsworkshopdec15.2012 (1)Cps security bitsworkshopdec15.2012 (1)
Cps security bitsworkshopdec15.2012 (1)
 
CPSSecurityBITSWorkshopDec15.2012 (1).pptx
CPSSecurityBITSWorkshopDec15.2012 (1).pptxCPSSecurityBITSWorkshopDec15.2012 (1).pptx
CPSSecurityBITSWorkshopDec15.2012 (1).pptx
 
Composite Intrusion Detection in Process Control Networks
Composite Intrusion Detection in Process Control NetworksComposite Intrusion Detection in Process Control Networks
Composite Intrusion Detection in Process Control Networks
 

Recently uploaded

IAC 2024 - IA Fast Track to Search Focused AI Solutions
IAC 2024 - IA Fast Track to Search Focused AI SolutionsIAC 2024 - IA Fast Track to Search Focused AI Solutions
IAC 2024 - IA Fast Track to Search Focused AI Solutions
Enterprise Knowledge
 
CNv6 Instructor Chapter 6 Quality of Service
CNv6 Instructor Chapter 6 Quality of ServiceCNv6 Instructor Chapter 6 Quality of Service
CNv6 Instructor Chapter 6 Quality of Service
giselly40
 
Artificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and MythsArtificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and Myths
Joaquim Jorge
 

Recently uploaded (20)

[2024]Digital Global Overview Report 2024 Meltwater.pdf
[2024]Digital Global Overview Report 2024 Meltwater.pdf[2024]Digital Global Overview Report 2024 Meltwater.pdf
[2024]Digital Global Overview Report 2024 Meltwater.pdf
 
IAC 2024 - IA Fast Track to Search Focused AI Solutions
IAC 2024 - IA Fast Track to Search Focused AI SolutionsIAC 2024 - IA Fast Track to Search Focused AI Solutions
IAC 2024 - IA Fast Track to Search Focused AI Solutions
 
Automating Google Workspace (GWS) & more with Apps Script
Automating Google Workspace (GWS) & more with Apps ScriptAutomating Google Workspace (GWS) & more with Apps Script
Automating Google Workspace (GWS) & more with Apps Script
 
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
 
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
 
Handwritten Text Recognition for manuscripts and early printed texts
Handwritten Text Recognition for manuscripts and early printed textsHandwritten Text Recognition for manuscripts and early printed texts
Handwritten Text Recognition for manuscripts and early printed texts
 
Boost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfBoost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdf
 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected Worker
 
Finology Group – Insurtech Innovation Award 2024
Finology Group – Insurtech Innovation Award 2024Finology Group – Insurtech Innovation Award 2024
Finology Group – Insurtech Innovation Award 2024
 
GenAI Risks & Security Meetup 01052024.pdf
GenAI Risks & Security Meetup 01052024.pdfGenAI Risks & Security Meetup 01052024.pdf
GenAI Risks & Security Meetup 01052024.pdf
 
Scaling API-first – The story of a global engineering organization
Scaling API-first – The story of a global engineering organizationScaling API-first – The story of a global engineering organization
Scaling API-first – The story of a global engineering organization
 
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot TakeoffStrategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
 
A Domino Admins Adventures (Engage 2024)
A Domino Admins Adventures (Engage 2024)A Domino Admins Adventures (Engage 2024)
A Domino Admins Adventures (Engage 2024)
 
Boost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivityBoost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivity
 
CNv6 Instructor Chapter 6 Quality of Service
CNv6 Instructor Chapter 6 Quality of ServiceCNv6 Instructor Chapter 6 Quality of Service
CNv6 Instructor Chapter 6 Quality of Service
 
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationFrom Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
 
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
 
Artificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and MythsArtificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and Myths
 
TrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
TrustArc Webinar - Stay Ahead of US State Data Privacy Law DevelopmentsTrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
TrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
 
08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking Men08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking Men
 

Complexity Reduction for Cyber-Physical-Human Medical Systems

  • 1. Low Complexity System Designs for Medical Cyber-Physical-Human Systems (CPHS) Po-Liang Wu University of Illinois at Urbana-Champaign Ph.D. Preliminary Examination, May 2014 1
  • 2. 2  The size and complexity of cyber-physical systems have increased significantly.  “These advances have not only made it possible to reach the frontier faster; they have also increased by orders of magnitude the levels of complexity open to exploration and experimentation. Understanding complexity and learning how best to harness these new capabilities are both a challenge and a responsibility.” NSF Strategic Plan, September 2006 Control & Harnessing Complexity of Cyber-physical Systems
  • 3. Introduction  Medical systems are both safety-critical and highly complex and our goal is to have a formally verifiable safe system in spite of the complexity, using the approach recommended by National Academic Science: • “One key to achieving dependability at reasonable cost is a serious and sustained commitment to simplicity, including simplicity of critical functions and simplicity in system interactions. This commitment is often the mark of true expertise”  Thus, we must understand the sources of complexity and their interactions in medical systems. 3 Goal: reduce and control complexity Complexity control is to control the interactions between the system components, for example, preventing the non-safety-critical components propagates the faults to the safety-critical components and cause failure of the whole system.
  • 4. Motivation – Complexity and Interactions 4 High mental workload increases the likelihood of medical errors The medical errors may cause severe complications. The degradation of patient conditions further increases the cyber-physical complexity Physical complexity: more medical devices must be added to handle the complications, more treatments must be performed Cyber complexity: more exception handling routines must be performed, more device interactions must be checked, etc Further increases the human mental workload.
  • 5. Dimensions and Measurements of Complexity  From a computer science perspective, complexity reflects the workload of performing tasks by measuring number of steps or elapsed time.  Computational complexity • Worst-case time requirements as a function of the size of its input (most usefully, in terms of the O-notation)  Verification complexity • Measures the state space to verify the system properties.  Cyber medical treatment complexity • Measures the number of steps to perform a treatment at run-time.  Mental workload complexity • Measures human memory and mental computation demand for performing tasks. • It can be measured by task completion time, accuracy, NASA-TLX, etc.  Other complexity 5
  • 6. Major Sources of Complexity  Sources of verification complexity • Message interleaving due to asynchronous communication • Inconsistency of distributed system states due to race conditions Summary: The combinatorial possible execution sequences and message interleaving must be checked. 6
  • 7. Communication Patterns (Remote Procedure Call) Supervisor Child device C1 Async C2 R1 Supervisor Child device C1 Sync C2 R1 R2 7 • Maximum flexibility • Easily cause state space explosion • Low complexity • Too restrictive
  • 8. State Space Explosion 8 msg queue size 1 2 3 # states 2336 122394 N/A (out of memory)  Model checking state space  Count total number of states Node A Node B msg3 msg2 msg1 Channel objectFSM object FSM object Size = 3 msgA msgB msgC Channel object The number of states is the Cartesian product of the states of node A, node B, and the communication channels. In our experiment, the state space quickly explodes as the more message interleaving is allowed.
  • 9. Major Sources of Complexity  Sources of cyber medical treatment complexity • Medical staff need to validate preconditions, monitor side effects, and check patient responses when performing a treatment • Incorrectly performing a treatment may cause severe medical complications. 9 ActivateDefibrillator Rhythm == Shockable InjectEPI BloodPH > 7.4 UrineFlow > 12 mL/s Inject- SodiumBicarbonate Airway & Breathing Treatment Preconditions
  • 10. Major Sources of Complexity  Sources of cyber medical treatment complexity • Medical staff need to validate preconditions, monitor side effects, and check patient responses when performing a treatment • Incorrectly performing a treatment may cause severe medical complications. 10 ActivateDefibrillator Rhythm == Shockable InjectEPI BloodPH > 7.4 UrineFlow > 12 mL/s Inject- SodiumBicarbonate Airway & Breathing AssistedVentilation The side effect of bicarbonate may adversely affect patient’s breathing. Urine flow rate may become lower than 12 due to kidney function degradation. IncreaseIVFluid
  • 11. Major Sources of Complexity  Sources of mental workload complexity • Physicians and nurses are unable to mentally keep track of physiological measurements minute by minute. • Most physicians can mentally correlate a small number of variables at a time but unable to correlate many variables in depth. • Unorganized and lack of context-dependent information » Context is a set of state variables, such as patient’s organ status and current medical procedure. » Context information is important for physicians and system to validate treatments and trigger alarms. Summary: Physicians must memorize and correlate tremendous information, which significantly increases their mental workload. State of the current medical practice 11
  • 12. Background – Existing Complexity Reduction Methodologies and Mechanisms  Verification complexity • Model checking techniques » Abstraction » Partial-order reduction • System design » Atomic Transaction/Serializability » Physically Asynchronous Logically Synchronous (PALS) architecture » Simplex Architecture  Cyber medical treatment complexity • Machine learning, expert systems • Closed-loop supervisory control  Mental workload complexity • Integrated display systems to improve situation awareness • GOMS(Goals, Operators, Methods, and Selection rules) model 12
  • 13. Technical Challenges  Concurrency related complexity • C1. In CPHS, asynchronous communication is commonly used due to interrupts and exceptions. » Ideas: Communication protocol to limit the possible message interleaving but still provide adequate flexibility. » Preliminary result: Interruptible Remote Procedure Call protocol • C2. Concurrent control of distributed devices and controllers result in complicated race conditions. » Ideas: – Architecture design to separate the concerns of controllers – Communication protocol to prevent conflict race conditions » Preliminary result: Hierarchical organ-based architecture, Consistency protocol • Summary: Proposing run-time control protocols to reduce unnecessary or inconsistent interleaving due to concurrency. 13
  • 14. Technical Challenges  C3. Medical complications due to incorrectly performed treatments. • Ideas: Validate treatments in respect of preconditions, side effects, and expected responses • Preliminary result: Treatment Validation Protocol  C4. Situation awareness improved display and assistance system. • Ideas: » Integrated display of physiological measurements, treatment status, and medical records. » Workflow-driven context information. • Work in progress: Workflow navigation and guidance system 14
  • 15. Technical Challenges  C5. Deviation from the best practice guidelines • Ideas: • Work in progress: » A protocol to adapt the workflows when deviation is detected in collaboration with medical staff. » Source code verification of the developed protocol using CBMC, which is a Bounded Model Checker for C/C++ programs. 15 Hard (safe-critical) deviation Soft (non-safety-critical) deviation System controllable Automatically forbid a. Stop once b. Request for confirmation c. Adapt the workflow Not system controllable Flag the deviation and disengage a. Flag the deviation b. Adapt the workflow
  • 16. REDUCING VERIFICATION COMPLEXITY: INTERRUPTIBLE RPC ARCHITECTURAL PATTERN 16 H. Yun, P.-L. Wu, M. Rahmaniheris, C. Kim, and L. Sha. A reduced complexity design pattern for distributed hierarchical command and control system. In Proceedings of the 1st ACM/IEEE ICCPS, pages 42-49. ACM, 2010.
  • 17. Communication Patterns Supervisor Child device C1 Async C2 R1 Supervisor Child device C1 Sync C2 R1 R2 Supervisor C1 stop Interruptible R1partial Child device C2 stop R2 17 • Maximum flexibility • Easily cause state explosion • Maximum complexity reduction • Too restrictive • Control asynchronous interactions in terms of mesage passings. • Allow bounded asynchronous interrupt and make the system back to synchronous • The increase of state space is also bounded
  • 18. Init B1 Next cmd! F_Result? Ready Exe cmd?F_Result! stop? B2 stop! stop? Supervisor Child Device FSM Definition of Interruptible RPC Pattern (Normal Case) 18 P_Result!cmd P_R stop
  • 19. Init B1 Next cmd! F_Result? Ready Exe cmd?F_Result! stop? B2 stop! stop? Supervisor Child Device FSM Definition of Interruptible RPC Pattern (Interleaving Handling) 19 P_Result! F_R stop
  • 20. Evaluation  Simplified distributed command & control system • One manager and one or more child devices • Modeling using Maude, a model checking tool  Evaluation Strategy • Compared Maude models following different patterns: synchronous, interruptible, asynchronous RPC • Varied number of child devices in the model • For asynchronous RPC, we also varied maximum queue size. 20
  • 21. Evaluation #child devices sync Interrup tible async(2) async(3 ) async(4) async(5 ) 1 5 29 70 121 195 275 2 11 133 1410 4426 11816 23732 3 22 701 28691 184418 782417 N/A(*) 21 State space (# states) #child devices sync Interrup tible async(2) async(3 ) async(4) async(5 ) 1 0 1 4 7 12 275 2 1 13 160 549 1571 23732 3 1 120 6586 49307 1096480 N/A(*) Model checking time (msec) (*) not finished in 12 hours
  • 22. Summary  Distributed command & control system model • Interleaving problem • State space explosion problem  Interruptible RPC design pattern • Allow asynchronous interruption • Reduces state space for model checking  However, this work does not address exception handling issues. • Child devices may actively send exceptions due to unexpected events, hardware failure, etc. 22
  • 23. REDUCING VERIFICATION COMPLEXITY: A LOW COMPLEXITY COORDINATION ARCHITECTURE 23 P.-L. Wu, W. Kang, A. Al-Nayeem, L. Sha, R. B. Berlin Jr, and J. M. Goldman. A low complexity coordination architecture for networked supervisory medical systems. In Proceedings of the 4th ACM/IEEE ICCPS, pages 89-98. ACM, 2013. W. Kang, P.-L. Wu, M. Rahmaniheris, L. Sha, R. B. Berlin, and J. M. Goldman. Towards organ- centric compositional development of safe networked supervisory medical systems. In Computer-Based Medical Systems (CBMS), 2013 IEEE 26th International Symposium on, pages 143-148. IEEE, 2013.
  • 24. State of the Art Medical Systems  Integrated Clinical Environment (ICE) infrastructure is logically centralized control system. • The medical devices are coordinated by a single ICE supervisor controller. 24 http://www.mdpnp.org/
  • 25. Motivating Example – A Laparoscopic Surgery  A camera is inserted into the patient’s abdomen and the CO2 is pumped into the abdomen to create a space for surgery.  The ventilator can be programmed to compensate the SpO2 level within the limits set by physicians • Such as tidal volume, respiratory rate, FiO2  Excessive CO2 pressure may inhibit the breathing volume and result in low SpO2 level. 25Extreme pressure would collapse the vena cava/hepatic vein and cause the shock. CO2 pump Monitor scalpel http://images.google.com
  • 26. Inconsistent Views and Actions May Cause Safety Hazard  SpO2 sensor raises a patient adverse events (PAE), caused by SpO2 level lower than the threshold set by the physicians  ICE supervisory controller requests to decrease CO2 pump pressure without knowing the status of the surgical process.  Decreasing CO2 pressure might cause insufficient operating space and a safety hazard.  Multiple devices coordination is essential to avoid safety hazards 26 Laparoscopic Surgery ICE Supervisory Controller CO2 Pump SpO2 Sensor Oxygen Saturation Low Decrease Pressure Video Camera Operating Space Safety Hazard
  • 27. Complexity of Asynchronous Interaction Hierarchical Layers Asynchronous Design State space Time (s) Two Layers 12084 3 Three Layers 603056 143 Four Layers Out of memory Out of memory 27  Complex interaction in medical systems can easily occur due to patient adverse events (PAE) and cause state space explosion.
  • 28. Hierarchical Organ-system Architecture  Medical Devices Layer • Medical devices report the physiological measurements and raise PAEs  Organ System Control Layer • It is responsible for stabilizing an organ-system by coordinating the devices. • If a PAE can be locally compensated, the organ- system controller will handle it • Otherwise, the PAE must be propagated to the high-layer controller. 28 SpO2 low Cmd
  • 29. Hierarchical Organ-system Architecture  Surgery and Anesthesia Control Layer • Surgical and anesthetic controller are responsible for inter-organ system coordination. • The anesthetic controller is the primary controller for anesthesia related events. • The surgical controller focuses on handling surgery related events. 29 SpO2 low Cmd
  • 30. Hierarchical Organ-system Architecture  Benefits • It naturally fits into the ICE infrastructure and enables localized PAE handling. If PAEs can be handled by different subtrees, the higher level controllers are not interfered. • Capture human homeostasis. (An organ system can compensate moderate abnormalities without interfering other organ systems)  Engineering challenges • Localized PAE handling may introduce message races to the controllers of different layers. • A communication protocol is required to guarantee consistency. 30
  • 31. Consistent View Generation and Coordination (CVGC) Protocol • Design goals: » The protocol ensures each controller has all its child nodes’ status. » Avoid conflict race conditions • The device setting is locked during the consistent view generation to avoid inconsistency. » Each device and controller has a lock to protect its setting. • A controller can send commands to change the device settings if and only if it holds all its child devices’ locks. 31 Like traditional locking mechanisms, such as tree locking, the proposed protocol aims to prevent race conditions that may cause safety hazards. Unlike traditional locking mechanisms, the system needs to handle both top-down and bottom-up message races.
  • 32. Consistent View Generation and Coordination (CVGC) Protocol  The CVGC protocol has three phases • PAE propagation phase • Locking and consistent view generating phase • Command execution and unlocking phase 32 Ca Cb PAE1&CV& LockSet
  • 33. Consistent View Generation and Coordination (CVGC) Protocol  The CVGC protocol has three phases • PAE propagation phase • Locking and consistent view generating phase • Command execution and unlocking phase 33 Ca Cb PAE1&CV& LockSet Cc lockAndReport lockAndReport lockAndReport message is used for locking a subtree and gathering the consistent view. A controller can send a command to change device settings if and only if the lockAndReport is processes and the controller receives the locks and consistent view.
  • 34. Consistent View Generation and Coordination (CVGC) Protocol  The CVGC protocol has three phases • PAE propagation phase • Locking and consistent view generating phase • Command execution and unlocking phase 34 Ca Cb PAE1&CV& LockSet Cc CV& LockSet CV& LockSet After processing the lockAndReport message and gathering consistent view and locks, Cb can send a command to change its child devices’ settings.
  • 35. Race Condition Handling  A race condition occurs when lockAndReport message is propagated downward while another PAE2 is propagated upward.  When Cc receives PAE2, its child devices’ settings have been locked. Therefore, Cc does not need to further propagate the lockAndReport message.  Cc sends PAE2&CV&LockSet to Cb.  Cb has the consistent view and all the locks.  How to handle multiple PAEs is pre- specified by the physicians. 35 Ca Cb PAE1&CV& LockSet Cc PAE2&CV& LockSet lockAndReport
  • 36. Consistent View Generation and Coordination (CVGC) Protocol  The CVGC protocol has three phases • PAE propagation phase • Locking and consistent view generating phase • Command execution and unlocking phase 36 Ca Cb Cc Cmd CmdCmd Cmd If Cb can handle PAE, it sends the command to change the device settings and release the locks.
  • 37. Correctness of the CVGC Protocol  Model checked architectural properties • CVGC protocol is deadlock free. • A controller can send a command iff it receives the consistent view and the locks. • The locking mechanism prevents multiple controllers command the same device simultaneously.  Timing analysis • µ: message transmission delay • WCET: worst case execution time 37
  • 38. Complexity Evaluation  Verified safety properties • The pulmonary controller will increase the respiratory rate of the ventilator when the SpO2_low PAE is received if the respiratory rate is below the maximum rate. (Layer 2) • If the surgery is at a critical point, the supervisory controller should not allow a reduction of the laparoscopic CO2 pump pressure. Instead, an alarm must be issued to the physicians. (Layer 4) 38 Hierarchical Layers Metrics Proposed Design Asynchronous Two Layers Number of states 1887 12084 Checking time (s) <1 3 Three Layers Number of states 10651 603056 Checking time (s) 2 143 Four Layers Number of states 21888 Out of memory Checking time (s) 8 Out of memory
  • 39. Summary  Reducing verification complexity by proposing: • Hierarchical organ-system architecture » Capture human homeostasis » Allow distributed patient adverse events handling • Consistent view generation and coordination protocol » Gather concise and comprehensive physiological information » Low complexity compared to asynchronous design. 39
  • 40. REDUCING CYBER MEDICAL TREATMENT COMPLEXITY: A TREATMENT VALIDATION PROTOCOL P.-L. Wu, D. Raguraman, L. Sha, R. B. Berlin, and J. M. Goldman. A treatment validation protocol for cyber-physical-human medical systems. To appear In Software Engineering and Advanced Applications (SEAA), 2014 EUROMICRO Conference on. 40
  • 41. Introduction and Technical Challenges  Timely and correctly perform treatments while monitoring patient responses is critical for patient safety.  Treatment validation is essential and consists of three aspects • Precondition » A treatment can be performed only if the preconditions are satisfied • Potential side effects » The side effects of a treatment may adversely affect other treatments or invalidate previously satisfied preconditions • Expected patient responses » The effectiveness of a treatment is non-deterministic  The purpose of this work is to assist medical staff to correctly perform treatments rather than automatically performing treatments. 41 Like traditional cyber validation mechanisms, the proposed protocol guarantees certain preconditions are satisfied before taking actions. Unlike traditional cyber validation mechanisms, the proposed protocol may not have the full control of physical components, such as patient conditions.
  • 42. Cardiac Arrest Resuscitation – State of the Art Practice 42
  • 43. Supervisory System User Interface Defibrillator Airway Breathing satisfactory Rhythm Non- Shockable 1. Activate defibrillator 2. Check Airway & EKG Rhythm Epinephrine BloodPH < 7.4 Urine < 12 mL/s 3. Inject epinephrine 4-a. Check preconditions 4-b. Request corrective treatments Performing Treatments Sodium Bicarbonate 6. Post order execution Sodium Bicarbonate No Calcium Chloride Increase IV Fluid 5. Sodium bicarbonate & Increase IV Airway Breathing is affected BloodPH < 7.4 Sodium Bicarbonate No Calcium Chloride 7. Side effects adversely affect preconditions 8. Patient responses not as expected 9. Exceptions 10. Alternative Treatment Assisted Ventilation 43
  • 44. Correctness  Theorem 1. (Preconditions satisfaction) Under the proposed protocol, a treatment is performed only if all preconditions of the treatment are satisfied.  Definition: A tree is well-formed if each unsatisfied precondition have a tree node for correcting it.  Theorem 2. (Tree Traversal) The root node of a well-formed tree is reachable if the corrective treatments are effective and the preconditions are not invalidated by the side effects.  Theorem 3. (Dynamic adaptability) Suppose side effects of a treatment invalidate any precondition and make the tree become non- well-formed. The protocol updates the tree to be well-formed if the medical staff correctly specifies the corrective treatments. 44
  • 45. Verification Medical safety properties P1: Defibrillator is activated only if the rhythm is shockable and airway and breathing is under control. P2: Epinephrine is injected only if the blood pH value is larger than 7.4 and urine flow rate is higher than 12 mL/s. P3: If the side effect of sodium bicarbonate adversely affects the breathing, the tree is updated with a new treatment node for assisted ventilation. Protocol correct ness properties P4: There is no deadlock in the system. P5: A treatment is performed only if all its preconditions are satisfied. P6: If side effect does not occur, the root node of the tree is added to the executing list P7: If side effects invalidate a precondition, the tree is updated to be well-formed. 45
  • 46. Summary  In order to reduce cyber medical treatment complexity, a validation protocol is proposed to • Check preconditions • Monitor potential side effects • Check expected responses  We use a model checking tool to verify both safety and correctness properties. 46
  • 48. REDUCING MENTAL WORKLOAD COMPLEXITY: WORKFLOW NAVIGATION AND GUIDANCE DISPLAY SYSTEM FOR IMPROVING SITUATION AWARENESS 48
  • 49. State of Current Research  (Main focus) Lack of context-dependent information and lack of adaption ability to the dynamic context changes  Context-dependent information • Context is a set of state variables, such as patient’s organ status and current medical procedure. • Context information is important for physicians and system to validate treatments and trigger alarms.  Our system organizes the context information and adapts to the context changes according to the medical workflow 49
  • 50. Medical Workflow  Medical workflow intends to capture the best practice to assist medical staff to diagnose and treat patients.  American Heart Association (AHA) provides a family of resuscitation workflows for different life-threatening heart rhythms, including ventricular fibrillation, asystole and tachycardia, etc.  The purpose of the developed system is to assist medical staff to synchronize the team actions to the workflow in real-time by • codifying the best practice medical workflows • keeping track of the workflow steps • flagging deviation from the workflow • highlighting the real-time progress of the treatments, the status of the treatments, and the patient’s physiological responses 50 Organ system states • Asystole • Tachy. • … • Asystole WF • Tachy WF • … Codified workflows • HR • BP • ECG rhythm • … Physiological measurements
  • 52. Workflow Manager Medical Device Configuration Management Resuscitation Monitoring System Device Interoperability Head Nurse 52 Situation Awareness Display Treatment validation Workflow Configuration Physicians The images are from http://www.pacificu.edu/pa/ Microsoft Surface
  • 53. Evaluation: Comparison  Vital signs only display  Vital signs + workflow  Vital signs + workflow + treatment validation status  Compare to different iterations of our design • Features added in iterations » Automatically generate medical records » Confirmation of drug order and given 53
  • 54. Evaluation: Candidate Metrics  Response time to a simulated event  Accuracy(# of errors) of the performed actions  Memory demand  NASA-TLX  Other questionnaire • Proper abstraction of workflows • Usefulness of treatment validation • Meaningfulness of the alarms 54
  • 55. Deviation from the Workflow Hard (safety-critical) deviation Soft (non-safety- critical) deviation System controllable Automatically forbid a. Stop once b. Request for confirmation c. Adapt the workflow Not system controllable Flag and disengage a. Flag the deviation b. Adapt the workflow 55 • Adapt the workflow after the deviation is detected in collaboration with physicians. • Simply switch to a new workflow may not work. • For example, patient is injected certain drug that will cause severe adverse interactions with another drug required in the new workflow.
  • 56. Source Code Verification  In my previous work, verification is performed at model design level.  In order to guarantee correctness and safety of the developed system, verification must also be performed at source code level.  We are currently re-engineering the system architecture to separate safety-critical and non-safety-critical software components. • For the safety-critical components and their interaction protocols, we will use CBMC, which is a Bounded Model Checker for C/C++ programs, to fully verify the safety and correctness properties. • For the non-safety-critical components, we will develop test suites to increase code coverage. • The safety-critical components only use but not depend on the non- safety-critical components. 56
  • 57. Published Work and Future Plans  Published work • P.-L. Wu, W. Kang, A. Al-Nayeem, L. Sha, R. B. Berlin Jr, and J. M. Goldman. ‘A low complexity coordination architecture for networked supervisory medical systems.” In: Proc. ICCPS’13. • P.-L. Wu, D. Raguraman, L. Sha, R. B. Berlin, and J. M. Goldman. “A treatment validation protocol for cyber-physical-human medical systems.’ To appear in EUMIRCRO SEAA’14. • H. Yun, P.-L. Wu, M. Rahmaniheris, C. Kim, and L. Sha. “A reduced complexity design pattern for distributed hierarchical command and control system.’ In: Proc. ICCPS’10. • W. Kang, P.-L. Wu, M. Rahmaniheris, L. Sha, R. B. Berlin, and J. M. Goldman. “Towards organ-centric compositional development of safe networked supervisory medical systems.’ In: Proc. CBMS’13.  Plans for publication • “A workflow navigation and guidance system for improving situation awareness”, targeting CHI’15 • “An adaptive protocol for handling deviations from the medical workflows”, targeting ICCPS’15 57
  • 58. Conclusion  Reduce and control complexity in respect of • Verification complexity • Cyber medical treatment complexity • Mental workload complexity  We have developed • interruptible RPC and consistent coordination protocol to reduce verification complexity. • treatment validation protocol to reduce treatment complexity  We are developing workflow navigation and guidance system for reducing mental workload complexity. 58
  • 60. Model-based Engineering  Model-based engineering is a system development methodology which exploits the reusable models in order to reduce complexity.  However, reusing a model during system integration requires • the assumptions or preconditions of each integrated model are fully checked and satisfied » For example, Ariane 5's accident was due to the violation of assumption of maximum horizontal speed. • the interactions of the models are thoroughly verified against potential unsafe interactions » For example, the surgical fire during an airway laser surgery is due to unsafe interactions between laser scalpel and ventilator. 60
  • 61. Design Patterns for Workflow Navigation and Guidance Systems  We identified several design patterns • Real-time treatment: Certain treatments must be continued or separated for a time interval » For example: CPR must continue for at least 2 minutes and epinephrine should be given every 3~5 minutes • Physiological conditions: the preconditions, side effects, and patient’s responses of the treatments can be described as physiological conditions, such as blood pressure > 120  In addition, we provide verifications for both model design and C code implementation levels using UPPAAL and CBMC (Bounded Model Checker for C/C++). 61
  • 62. 62 Interleaving Problem  Model ohecking  R: in case of no pending msg, MSMGR, SSMGR and SS must be in the same state (track or scan) Maude> search (init,off,off,none,none,none,none) =>+ (targeting,airscan,airscan,none,none,none,none ) . Solution 1 (state 136) states: 137 rewrites: 261 Maude> show path label 136 . MM_takeoff MM_takeoff_waypoint MM_waypoint_patternsearch SM_off_airscan MM_patternsearch_abort SS_off_airscan SS_airscan_airtrack SM_airscan_airtrack SM_any_off MM_abort_waypoint MM_waypoint_patternsearch MM_patternsearch_targeting SS_any_off SM_off_airscan SS_off_airscan MSMGR SSMGR SS waypoint abort waypoint pattern targetting airscan airscan airtrackairtrack off airscan airscan inconsistency scanAir scan_air track track scanAir off stop stop
  • 63. UPPAAL Model for Pulmonary Controller 63
  • 64. UPPAAL Model for Pulmonary Controller 64 PAE propagating Cmd executing and unlocking Locking and consistent view generating phase
  • 65. System Models and Definitions  PhysiologicalCondition (PC) is defined as a tuple < Checker, PM, Operator, RV >, where • Checker is the entity capable of checking the, Checker ∈{System, MedicalStaff}, • PM (physiological measurement) ∈ {EKGRhyhtm, HeartRate, BloodPressure...}, • Operator ∈ {>, <, =, ≤, ≥}, • RV is the reference value, which can be threshold, trend, or pattern of the physiological measurements.  A treatment is defined as a tuple < Agent, Action, PS, SS, ES, L >, where • Agent is the entity that performs the treatment, Agent ∈ {MedicalDevices, MedicalStaff}, • Action is the set of executable instructions, • PS is a set of preconditions that must be satisfied before preforming the Action • SS is a set of potential side effects • ES is a set of expected responses after performing the treatment • L is the life cycle, which specifies the time interval between the treatment being performed and the treatment has no further effect on the patient 65
  • 66. Major Mental Workload Complexity Sources – Using Cardiac Arrest Resuscitation as a Case Study  Most of medical records taken during the procedures are paper-based and not electronically integrated with the Electronic Medical Records (EMR) in real-time. • # of spoken words, questions raised  Medical staff must memorize the workflow • # of medical errors (reviewed by an expert) • Response time to a simulated event  No validated treatments information • # of medical errors  Medication orders are verbally communicated • # of spoken words, questions raised  Task completion time, NASA-TLX can serve as general metrics for mental workload complexity. 66
  • 67. State of Current Practices  Single-Sensor-Single-Indicator is not enough: "State of the art" medicine practices usually focus on one control loop (one measurement, HR/BP, and one contributor, drug, disease, etc).  Lack of integrated and well-organized display: Related information are usually scattered across multiple devices and screens. This situation significantly increases medical staff's mental workload and the chance of preventable medical errors.  Some research work has addressed the above problems by providing integrated display. 67
  • 68. Interactive Situation Awareness Monitor for Physicians  Reduce the large volume of medical information to a small set of concise and comprehensive information according to the chosen workflow 68
  • 69. Order Management Tablet (Handheld) for Nurse in Charge  Transform the concurrent, asynchronous and chaotic order management environment into an orderly, safe, and fault-tolerant one. Ensure each issued order is • consistent with the preconditions, the timing, workflow, patient physiological status • consistent with the available equipment, and personnel (to be implemented) • interoperable with hospital EMR (to be implemented)  The tablet interacts with our backend system and the situation awareness monitor  In the event of system or device failure, the order management tablet has specified safety functions. 69 Microsoft Surface
  • 70. Virtual Patient Station for Training and Evaluation  To support pre-recorded data mode • The EKG signal data are obtained from PhysioNet. • Different heart rhythms can be selected at anytime by the physician who is responsible for simulating patient responses. • Patient vital signs, including heart rate, blood pressure, etc., can be selected as well. • Physicians, residents, and nurses will “treat” the virtual patients as if they were real. 70
  • 71. Situation Awareness Monitor Layout Sensor readings Confirm rhythm Treatments EKG signal Treatment timelines Medication Messages Log Current Workflow 71