a virtual glucose homeostasis model for verification...
TRANSCRIPT
A Virtual Glucose Homeostasis Model for Verification,Simulation and Clinical Trials
Neeraj Kumar Singh
INPT-ENSEEIHT/IRITUniversity of Toulouse, France
September 14, 2016
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 1 / 25
Outline
1 Critical Systems
2 Context and Problems
3 Glucose Homeostasis Models
4 Formalization of GH
5 GH Simulator Framework
6 Hardware Implementation of GH
7 Conclusion & Future Work
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 2 / 25
Critical Systems
Definition
A system whose failure may result in injury, loss of life, economical loss orserious environmental damage. These systems require interaction betweencomputational and physical elements.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 3 / 25
System Failures
System Failures
Therac-25 (1985-1987): six people overexposed to radiation.
Pacemaker and ICD (1990-2002): 17,323 pacemakers and ICDs wereexplanted that includes 61 deaths.
Insulin Infusion Pump (IIP) (2010): 5000 adverse events that includes30 deaths.
Missing Malaysian Plane MH370 (8 March, 2014): Unknown.
Satellite Failure:+150: http://www.sat-nd.com/failures/.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 4 / 25
Context and Problems
Context
Development of virtual model for verification, simulation and clinical trialsthat can be used in the development process of medical devices.
Current Challenges
Increasing complexity of the critical medical systems.
Lack of biological simulation/environment for medical devices.
Better techniques for requirement analysis.
Needs some sound techniques to meet regulators, and certificationstandards.
Modelling critical medical systems using human-in-loop architecture.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 5 / 25
Insulin Infusion Pump
Insulin Infusion Pump
An insulin pump is a small, complex, software-intensive medical device thatallows controllable, continuous subcutaneous infusion of insulin to patients.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 6 / 25
IIP Requirements
REQ1: The device must undergo a power-on-self-test (POST) whenever devicepower is turned on.REQ2: The device must suspend all active basal delivery or bolus deliver duringpump refilling and in the case of system failure.REQ3: The device shall allow the user to manage system functionalities relatedto: stopping insulin delivery; validating basal profiles parameters; remindermanagement; and validating bolus preset parameters.REQ4: The device shall allow the user to define a basal profile that consists of anordered set of basal rates, ordered over a 24 hour day, as well as a temporarybasal, that consists of a basal rate for a specified duration of time within a 24hour day.REQ5: The device can contain several basal profiles, but only one basal profilecan be active at any single point in time.REQ6: The device must allow the user to override an active basal profile with atemporary basal, without changing the existing basal profile.REQ7: The device shall resume the active basal profile after the temporary basalterminates.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 7 / 25
IIP Requirements
REQ8: The device shall enforce a maximum dosage for the normal bolus orextended bolus.REQ9: The user shall be able to stop the active normal or extended bolus.REQ10: The device must maintain an electronic log of every operationassociated with an user alert, such as an audio alarm.REQ11: The device shall maintain a history of basal and bolus dosages over thepast n days. n always differs among brands, though most store up to 90 days ofdata.REQ12: The device shall enable the user to create a food database that can beused to store food or meal descriptions and the carbs associated with them.REQ13: The device shall allow to the user to change parameter settings of basalprofile, bolus preset, and temporary basal.REQ14: The device shall provide feedback to the user regarding system anddelivery status.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 8 / 25
Objectives
To identify gaps or inconsistencies in the IIP requirements.
Developing a closed-loop model for verification and validation of IIP.
Analysing system interaction between the GH model and IIP.
Developing a simulation model from the formal virtual GH model.
Developing a test bench using the virtual GH model and simulationfor clinical trials of IIPs.
Generating test cases to test the functional correctness of IIPsoftware.
Developing patient specific model at various level of systemdevelopment.
The virtual environment model can be used by medical industriesduring the product development.
The virtual GH environment model can be used by regulators forvalidating and certifying the medical devices.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 9 / 25
Glucose Homeostasis Models
Clinical Models
Models for
diagnostic tests
control
progression
complications
Non-Clinical Models
Models for
insulin-glucose, hepatic glucose, glucagon, and insulin receptordynamics
beta-cell insulin release
brain glucose homeostasis
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 10 / 25
Glucose Homeostasis Models
Use of the Models
To present a simulation of the glucose homeostasis system tounderstand the actual behavior.
To assist the medical experts for simulation purpose of the glucosedynamics.
Disadvantages
Existing models are based on complex mathematics, therefore thesemodels
are difficult and make simulation very time-consuming.
are not suitable for presenting an abstract behavior of GH.
are not suitable for verification purpose.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 11 / 25
Glucose Homeostasis Models
Use of the Models
To present a simulation of the glucose homeostasis system tounderstand the actual behavior.
To assist the medical experts for simulation purpose of the glucosedynamics.
Disadvantages
Existing models are based on complex mathematics, therefore thesemodels
are difficult and make simulation very time-consuming.
are not suitable for presenting an abstract behavior of GH.
are not suitable for verification purpose.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 11 / 25
Glucose Homeostasis System
HighCPlasmaCGlucoseCLevel
LowCPlasmaCGlucoseCLevel
NormalCGlucoseCLevelCMaintainedC
α-cellsβ-cells
GlucagonRelease
InsulinRelease
GlucoseCUtilizationCandCStorageCinCLiverC
asCGlycogen
LiverCConvertsCGlycogenCtoC
Glucose
GlucoseCLevelCDrops
GlucoseCLevelCRises
Pancreas
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 12 / 25
Glucose Homeostasis Automata
Glucose Level Drops
Glucose Level Rises
Hi No Lo
AcBc
Li
St Tr
Insulin Release
GlucagonRelease
Figure: The Glucose Homeostasis AutomataNeeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 13 / 25
The Glucose Homeostasis Definition
Definition 1 (The Glucose Homeostasis System)
Given a set of nodes N, a transition T is a pair (i, j), with i, j ∈ N . A transition is denoted by i j. The glucose homeostasis system is a tuple GHS = (N, T, N0) where:
• N = { Hi, No, Lo, Ac, Bc, Li, St, Tr } is a finite set of landmark nodes in the glucosehomeostasis network;
• T ⊆ N × N = {No 7→ Hi, Hi 7→ No, No 7→ Lo, Lo 7→ No, Hi 7→ Hi, No 7→ No, Lo 7→ Lo, Hi7→ Bc, Lo 7→ Ac, Bc 7→ Li, Ac 7→ Li, Li 7→ St, Li 7→ Tr, St 7→ No, Tr 7→ No, St 7→ Hi, Tr 7→ Lo,Tr 7→ Hi} is a set of transitions to present data flow between two landmark nodes;
• N0 = No is the initial landmark node (normal glucose level);
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 14 / 25
Abnormal Glucose Homeostasis System
HighbPlasmabGlucosebLevel
LowbPlasmabGlucosebLevel
NormalbGlucosebLevelbMaintainedb
Alpha-CellsbDefect;Abnormalb
GlucagonbRelease
InsulinRelease
Beta-CellsbDefect;InsufficientborbnoInsulinbSecretion
InsufficientborbnobGlucagonb
SecretionPancreas
InsulinbResistanceinbCellsb
ExcessbInsulinborbExtreambExercise
ExcessbGlucagonbSecretion
PersistentHighbPlasmab
GlucosebLevel
Hyperglyoemia-inducedDiabetesbComplications
PersistentLowbPlasmab
GlucosebLevel
PersistentHighbPlasmab
GlucosebLevel
β-cells α-cells
Figure: Abnormal Glucose Homeostasis System
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 15 / 25
Blood Glucose Level
Property 1 (Blood Glucose Level)The blood glucose level defines different stages, such as hyper- glycemia, hypoglycemia andnormal. The glucose level is low (hypoglycemia) if FPG ∈ [0,70) or OGTT ∈ [0,70), and theglucose level is high (hyperglycemia) if FPG ≥ 125 or OGTT ≥ 200, and the glucose level isnormal if FPG ∈ [70,100) or OGTT ∈ [70,140). We classify pre-diabetes to be the range whereFPG ∈ [100,125) or OGTT ∈ [140,200).
Blood Sugar Level Fasting Plasma Glucose (FPG) Oral Glucose Tolerance Test(mg/dL) (mg/dL)
Normal 70..99 70..139Pre-Diabetes 100..125 140..199High glucose 126 or above 200 or aboveLow glucose 0..69 0..69
Table: FPG and OGTT Test Values for Glucose Level
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 16 / 25
Abstract Model
Context
axm1 : Glucose level , {Normal}, {High}, {Low})axm2 : partition(GHS , {OK}, {KO})
NoHi Lo
Figure: Automata of an Abstract Model
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 17 / 25
Abstract Machine
inv1 : Current Glucose Level ∈ Glucose levelinv2 : Diabetic Condition ∈ GHSinv3 : Diabetic Condition = KO⇔
Current Glucose Level = High ∨ Current Glucose Level = Lowinv4 : Current Glucose Level = Normal ⇔ Diabetic Condition = OK
EVENT Normal GlucoseWHENgrd1 : Current Glucose Level = Normal∨
Current Glucose Level = Low∨Current Glucose Level = High
THENact1 : Current Glucose Level := Normalact2 : Diabetic Condition := OK
END
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 18 / 25
Abstract Machine
EVENT High GlucoseWHENgrd1 : Current Glucose Level = Normal ∨ Current Glucose Level = High
THENact1 : Current Glucose Level := Highact2 : Diabetic Condition := KO
END
EVENT Low GlucoseWHENgrd1 : Current Glucose Level = Normal ∨ Current Glucose Level = Low
THENact1 : Current Glucose Level := Lowact2 : Diabetic Condition := KO
END
N. K. Singh, Hao Wang, Mark Lawford, Thomas S. E. Maibaum, and Alan Wassyng “Formalizing the Glucose HomeostasisMechanism”, 16th International Conference on Human-Computer Interaction (HCI 2014), LNCS, Springer InternationalPublishing, pp. 460–4271, Vol-8529, 2014.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 19 / 25
Progressive Refinements
Refinement
Gradually introduce details and complexity to the specification byremoving non-determinism and by adding more events.
Refinement 1 : To introduce the α-cells and β-cells of Pancreasincluding their releasing functions and defects.
Refinement 2 : To define the liver behaviour for converting orstoring the glucose in the body.
Refinement 3 : To add the abnormal conditions of Pancreas,diabetic conditions, and diabetes complications.
Refinement 4 : To introduce the Blood Sugar Concentration forassessing the Diabetes and Pre-diabetes.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 20 / 25
GH Simulator Framework
Physiology of GH
Physiology of the Pancreas
Physiology of the Liver
Insulin-glucose Dynamics
Glucagon-glucose Dynamics
GH Abnormality
Physiology of pancreatic alpha-cells and beta-cell
Modeling Toolsand
Computation Tools
SimulationKernel
Required Parameters
User Interface and
Visualization of GH
Formal Specification
of GH
Finite elements,Finite differences,Lumped elements,etc.
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 21 / 25
Hardware Implementation of GH
User Interface and
Visualization of GH
Formal Specification
of GH
GHSimulation
Modeling and Implementation
of GH Model
HardwarePlatform
FPGA, Arduino, Snickerdoodle, etc
Matlab orLabVIEW
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 22 / 25
Usage Scenarios of Environment Model
Behavioural Requirements
To Discover Essential Safety Properties
Patient Safety in Closed-loop
To Generate Automatic Test Cases
Test Bench for Clinical Trials of IIPs
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 23 / 25
Conclusion & Future Work
Conclusion
Logic based mathematical modelling of the GH (Normal andAbnormal)
Developed a stepwise formal model of GH environment model
Applying model checking and theorem prover to prove safetyproperties.
Proposed Simulation framework
Proposed Hardware Implementation framework
To meet the V&V requirements of certification standards like FDAetc.
Future Work
Integration of GH model and IIP for developing the closed-loop model
Development of GH simulator
Test bench for IIPs through developing the hardware platform of GH
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 24 / 25
Neeraj Kumar Singh A Perspective on Environment Modelling September 14, 2016 25 / 25