Transcript Proposal for Grid/cloud Service for Parameter Sweep in Modelica

Calibrating models of human physiology
using scientific cloud
Tomáš Kulhánek
Institute of Pathological Physiology, First Faculty of
Medicine, Charles Univerzity in Prague, Czech
Republic
EGI Champion
Who we are

Institute of Pathological Physiology

Interdisciplinary team (~10 people)- physicians, mathematicians, computer
scientists, biomedical engineers, painters/graphical designers, …

mathematical modeling of human physiology, Software system for simulation
application, Graphical design, Educational portal www.physiome.cz/atlas
Modelica
Modelica - is an open standard, object-oriented,
declarative, multi-domain modeling language for
component-oriented modeling of complex systems.
Industry - automotive companies, such as Audi,
BMW, Daimler, Ford, Toyota, VW use Modelica to
design energy efficient vehicles and/or improved air
conditioning systems. Power plant providers, such
as ABB, EDF, Siemens use Modelica, as well as
many other companies.
Research - projects within Europe spend 75 Mil. €
in the years 2007-2015 to further improve Modelica
and Modelica related technology. This is performed
within the ITEA2 projects EUROSYSLIB,
MODELISAR, OPENPROD, and MODRIO
Tools – commercial (3DS Dymola, Wolfram
System Modeler, MAPLE, Simulation X), freeopensource (OpenModelica)
Modelica in Physiology
Combination of hydraulic, biochemical, thermofluid, osmotic domain
HumMod - Kofránek, Jiří, Mateják,Marek, Privitzer, Pavol: HumMod large scale physiological model in Modelica. 8th. International
Modelica konference 2011, Dresden, Germany
Physiolibrary – free library for modeling physiology,
www.physiolibrary.org, 1st price Modelica Free Library Award, 10th
International Modelica Conference, March 12, 2014, Lund, Sweden
What is our motivation
Tell me, I‘ll forget.
Show me and I may remember.
Involve me and I‘ll understand.
Models
Mathematical models of physiological
subsystems integrated into one unit
respiration
Acidbase,
ionic,
volume
and
osmotic
balance
circulation
Model
of
aircraft
Blood gases
Haematopoiesis
Digestion
Neural and
hormonal control
… and other
susbsystems Energy metabolism
Influence of drugs
Models of medical
devices
Model of human
physiology
Motivation for modeling
Experiment,
measurement of
experimental data
Model, simplified
mathematical
description of reality
Verification,
Validation, compare
model simulation
with real data
Example – model of cardiovascular system
Elastic baloon
Hydraulic resistor
Hydraulic valve
Pulmonary
circulation
Heart
Systemic circulation
Example – model of cardiovascular system
Example – model of cardiovascular system
Example – model of cardiovascular system
RRAIN – resistance of vena cava=0.003 mmHg*s/ml
EITHV – elastance of vena cava = 0.0182 mmHg/ml
Example – model of cardiovascular system
Example – model of cardiovascular system
Ventricula Elastance EMAX = 4 mmHg/ml
Example – model of cardiovascular system
Ventricula Elastance EMAX = 4 mmHg/ml
RRAIN – resistance of vena cava=0.003 mmHg*s/ml
EITHV – elastance of vena cava = 0.0182 mmHg/ml
…
Parameters of the
model
Example – model of cardiovascular system
Ventricula Elastance EMAX = ?
RRAIN – resistance of vena cava=?
EITHV – elastance of vena cava = ?
…
What are the values of the
parameters, which fits to
concrete human?
Example – model of cardiovascular system
Ventricula Elastance EMAX = ?
RRAIN – resistance of vena cava=?
EITHV – elastance of vena cava = ?
…
What
are the values of
the parameters, which
fits to concrete human?
- Hard(invasive) to measure directly
- Measure other model variable
- compute parameter, fit simulation
with measured data
„parameter identification“
„Model calibration“
Example – model of cardiovascular system
Measure variable from real patient
Estimate parameters and simulate and compare
with the measured data (curve fitting)
Repeat until simulation gives similar result as
the measured data
What we have done
System for parameter identification
-
Modelica model => FMI Executable
(commercial Dymola tool)
-
Web portal and web services => .NET,
REST API (ServiceStack,SignalR)
-
Non-linear mathematical models =>
Global optimization methods => Genetic
algorithm
-
Each iteration produces independent
tasks => we tried Desktop grid (BOINC),
local cluster, EGI cloud (CESNET NGI)
-
Kulhánek T., Identification of model
parameters in cloud deployed simulation
service, IEEE EMBS 2013 – Osaka,
Japan, EGI TF 2013 - Madrid
What we have done
Model of saturation O2 in Hemoglobin
-
Data Imai (1972), Roughton(1967), …
-
Theory:
-
Find K1 … K4 to fit the data
K1=K2 = 0.1301628020369
K3 = 0.744154273954473
K4 = 32.4804162428542
200 000 simulations in local cluster in 20
minutes
What we have done
Complex model of human physiology
(HumMod)
-
Origin www.hummod.org
-
Modelica implementation
Single simulation – 10-60 s
Calibrating model for test parameter took 7
days on local cluster vs. 18 hours on cloud:
What we have done
Model of hemodynamics of
cardiovascular system
Meurs et al. (2006-2011), Burkhoff et
al.(1997-2013) 7), Fernandez de Canete
et al. (2013), …
-
Find factors of vena cava elastance,
5 hours
in local
cluster elastance,… to fit
resistence,
ventricular
30pacient
minutesdata
in EGI
cloud (30
(pressure
on CPU)
different body
positions)
heart.heart.leftHeart.ventricularElastance.factor0.276715208845232
pulmonaryCirculation.EPA.factor 2.42977968242538
pulmonaryCirculation.RPP.factor 0.326200900501185
pulmonaryCirculation.RLAIN.factor
7.03415309534463
heart.heart.leftHeart.RxAOutflow.factor 0.0579461745835701
heart.heart.leftHeart.RxVOutflow.factor 0.115592156812037
heart.heart.rightHeart.ventricularElastance.factor
0.00138825200390164
heart.heart.rightHeart.RxVOutflow.factor 0.0339999571366457
systemicCirculation.systemicVeins.RETHV.factor
21.5099238722829
systemicCirculation.systemicVeins.EITHV.factor
0.183594380817957
systemicCirculation.systemicVeins.VITHVU.factor
0.00115139326808882
systemicCirculation.systemicVeins.RRAIN.factor
29.4603401420539
systemicCirculation.systemicPeripheralVessels.RTA.factor 2.58765141868037
systemicCirculation.systemicPeripheralVessels.RSP.factor 0.744428239928714
systemicCirculation.systemicArteries.EETHA.factor 0.318031878463708
systemicCirculation.systemicArteries.RETHA.factor 0.0792556174368193
systemicCirculation.systemicArteries.EITHA.factor
0.416618612811922
Summary
Calibrating models of human physiology
Motivation
Education – train students using simulators
Research – construct, validate models
System
Web portal, .NET web services, REST API, …
local cluster, EGI cloud (CESNET NGI)
Models
Simple models – suitable to calibrate locally
Middle models – ?
Complex models – suitable in HPC cloud
What next?
Calibration of model of farmacodynamics & farmacokinetics
– use case in healthcare
Model calibration – iteration steps can‘t be parallelized
Parameter sweep – iterations can be parallelized, Monte
Carlo simulation
integrate model simulation to grid middleware, or desktop
grid middleware
Repository of valid values of physiological models
- repository of validated models
- results of model calibration
- validated values from journal articles
Motto
Tell me, I‘ll forget.
Show me and I may remember.
Involve me and I‘ll understand.
Thank you for your attention
Acknowledgment: EGI, CESNET NGI
Supported by: FR CESNET 431/2011
[email protected]
www.physiovalues.org