Content uploaded by Robert G Grossman
Author content
All content in this area was uploaded by Robert G Grossman
Content may be subject to copyright.
Tetrahedraland PolyhedralMesh EvaluationforCerebral
HemodynamicSimulation-aComparison
MartinSpiegel,ThomasRedel,Y.JonathanZhang,TobiasStruffert,JoachimHornegger,
RobertG.Grossman,Arnd Doerflerand ChristofKarmonik
Abstract—Computationalfluid dynamic (CFD)basedon
patient-specificmedicalimagingdatahasfound widespread use
forvisualizing and quantifyinghemodynamicsin cerebrovas-
culardiseasesuchascerebralaneurysmsorstenotic vessels.
This paper focusesonoptimizingmesh parametersforCFD
simulationofcerebralaneurysms.Valid blood flowsimulations
strongly depend onthemesh quality.Mesheswithacoarse
spatialresolutionmay leadto aninaccurateflowpattern.
Mesheswithalargenumber ofelementswill result in un-
necessarily highcomputationtimewhichis undesirable should
CFDbeusedforplanningin theinterventionalsetting.Most
CFDsimulationsreportedforthesevascularpathologieshave
usedtetrahedralmeshes.Weillustratetheuseofpolyhedral
volume elementsin comparisontotetrahedralmeshing ontwo
differentgeometries,asidewall aneurysmof theinternalcarotid
artery and abasilarbifurcationaneurysm.Thespatialmesh
resolutionrangesbetween5,119 and 228,118 volume elements.
The evaluationof thedifferentmesheswasbasedonthewall
shearstress previously identifiedasa onepossible parameter for
assessing aneurysmgrowth.Polyhedralmeshes showed better
accuracy,lower memorydemand,shorter computationalspeed
and faster convergence behavior(onaverage369 iterationsless).
I.INTRODUCTION
Advancesin hardware and softwarehave enabledthe
application ofCFDin variousclinicalfields,e.g.cardiology
[1]orneuroradiology [2].Differentaspectshavebeenana-
lyzedlikepatient-specifichemodynamicsimulation [3],[4],
correlation ofwall shearstress (WSS)patternwiththerisk
of ruptureofcerebralaneurysms[5],[6]orhemodynamic
characteristicspertaining tothe angleofthe aneurysmbulb
relativetotheparentartery[7].OtherCFDstudiesconsid-
eredthegeometricfactorsofaneurysms[8],[9],[10],e.g.
lesion size oraspectratiotoassess therisk ofananeurysm
M.Spiegel iswithFriedrich-AlexanderUniversityErlangen-Nuremberg
(FAU),DepartmentofComputerScience,ChairofPatternRecogni-
tion,Germany and FAU,DepartmentofNeuroradiology and Siemens
AG HealthcareSector,Forchheim, Germany and theErlangenGrad-
uateSchool inAdvancedOpticalTechnologies(SAOT),Germany
martin.spiegel@informatik.uni-erlangen.de
T.Redel iswithSiemensAG HealthcareSector,Forchheim, Germany
J.Zhang isaninterventionalneuroradiologistand vascularsurgeon with
TheMethodistHospital,Houston,TX,USA
T.Struffert isaninterventionalneuroradiologistwithFAU,Department
ofNeuroradiology,Erlangen,Germany
J.HorneggeriswithFAU,DepartmentofComputerScience,Chairof
PatternRecognition (Head)and SAOT,Erlangen,Germany
RG GrossmanisChairman oftheDepartmentofNeurosurgeryand
DirectoroftheNeurologicalInsitute,TheMethodistHospital,Houston,
TX USA
A.DoerfleriswithFAU,DepartmentofNeuroradiology (Head),Erlan-
gen,Germany
C.KarmonikisaResearchScientistwithTheMethodistHospital
ResearchInstitute,Houston,TX,USAckarmonik@tmhs.org
rupture.In[11],different inflowconditionswereinvestigated
toassess itsdependence totheWSSdistribution inthe
vicinity ofthree basilartipaneurysms.WSSpatternsplaya
major roleforassessing pathologicalvesselsystemswithin
theCFDblood flowcommunity.
Mostofthesesimulationsutilizedtetrahedralmeshes.More
advancedmeshing techniquesarenowavailableinturn-key
commercialCFDsoftwarepackageswhichmayleadtoare-
duction incomputing timeby reducing theoverall numberof
volume elementswhilemaintaining accuracy.While accurate
meshing isessential to obtainmesh-independentresults,ade-
ficiencyinthemeshmay notbeobviousand mayleadto non-
validresultsasotherincorrectsimulation parameters,such
asimproperinletconditionsintermsofmass orblood flow
speed,varying blood densityand viscosity,blood modeledas
Newtonianfluid.Mesh deficienciesinclude cellswith high
skewness factorand a coarsespatialresolution thatmaylead
toimprecision computation oflocalvelocity orWSSpattern.
While assessmentofthemesh qualityistime-consuming,it is
nevertheless averyimportant task.Two distinctapproaches
formeshsuitability verification havebeenintroduced[12]:
1)Comparison ofthenumericalsimulation resultswith
experimentaldata and 2)generation ofasetofmeshes
withincreasing numberofcontrolelementsalsocalledmesh
independence analysis.Phase-contrastMRImeasurements
deliversuitableresultstocompareMRvelocity valuesorpri-
mary blood flowpatternswithCFDsimulation results[13].
Meshindependence analysisisconsideredthe established
method forverifying meshaccuracytosimulate arterialflow.
Itrequiresasetup ofdifferentmeshespervesselgeometry
fromlowto high resolution.
Thegoalofthis study wastoexploreintheframework of
meshindependence analysiswhetherpolyhedralmesheswith
fewervolume elementsthancomparabletetrahedralmeshes
will result in betteraccuracywithshortercomputation time
using patient-specificimaging data.Twocerebralaneurysm
geometrieswereinvestigated:oneinternalcarotidand one
basilartipaneurysm,respectively(see Fig.1).Asetof
varying meshresolutionswas systematicallycreatedasa
benchmarkforspatialmeshresolutionsneededto obtain
accurate,mesh-independentWSSpatternsorblood flow
velocity.
II.METHODS
3-Ddigitalsubtraction angiography (DSA)imagedata
[14]ofthetwocerebralaneurysms(see Fig.1)were ac-
quired during endovascularinterventionsusing abiplane
2787
31st Annual International Conference of the IEEE EMBS
Minneapolis, Minnesota, USA, September 2-6, 2009
978-1-4244-3296-7/09/$25.00 ©2009 IEEE
TABLE I
SET OFMESHES-CASE1AND CASE2.TE T.AND POLY.ABBREVIATE TETRAH EDRALAND POLYHEDRALRES PECTIVELY.PA AND ITE R.DENOT E
PASCALAND ITERATION S.
#Tet.CellsAngleGrowthMax./Min.TriangleTet.#Iter.#Poly.CellsPoly.#Iter.Tet.WSS(Pa)Poly.WSS(Pa)
Case1 13841 25 1.2 6 /0.001 153 5119 63 9.52 10.33
26702 14 1.2 3.6/0.001 180 7283 81 11.66 11.88
52374 14 1.1 1.8/0.001 296 12467 108 13.37 13.15
76493 6 1.2 1.2/0.001 306 17695 122 14.53 13.80
86621 12 1.1 0.4/0.001 1633 18681 142 14.43 13.92
106010 12 1.1 0.36 /0.001 1291 22297 185 14.76 13.99
115014 10 1.1 0.3/0.001 451 25540 154 14.71 14.32
228118 10 1.1 0.25 /0.001 -43487 583 -15.58
Case2 31696 17 1.2 0.42 /0.001 242 8814 111 5.91 5.26
43063 17 1.17 0.37 /0.001 378 10843 176 7.18 6.20
56176 15 1.12 0.34 /0.001 365 13280 185 7.06 6.51
75469 13 1.10 0.32 /0.001 437 16795 202 7.33 6.84
98399 10 1.10 0.3/0.001 462 21108 211 7.43 7.15
114227 10 1.10 0.28 /0.001 496 23846 230 7.47 7.23
148625 8 1.09 0.25 /0.001 714 30167 269 7.62 7.45
174225 6 1.10 0.24 /0.001 626 35665 254 7.48 7.57
(a) (b)
Fig.1.Three dimensionalreconstruction ofthe evaluatedaneurysms:(a)
Case1 depictsainternalcarotidaneurysm(b)Case2illustratesabasilartip
aneurysm
flat-panelsystemSiemensC-ArmSystem(dBA,Siemens
AG HealthcareSector) inErlangen(Germany)and Houston
(USA).3-Dimagereconstructionsofbothaneurysmsledto
animagevolumeforCase1 of138.24×138.24×57.72mm
withavoxelspacing of0.27×0.27×0.13mm and forCase2
97.28×97.26×66.43mm (voxelspacing 0.19×0.19×0.1mm)
respectively.Formesh generation and smoothing,theMarch-
ing Cubes[15]and aLaplacian-basedsmoothing algorithm
(VTK KitwareInc.)wasappliedasanaddtionalstep.The
imagedatawas storedasastereolithographicfile(STL)pro-
viding theinputdataforthe commercialmeshing software
GAMBIT(ANSYSInc.).
TheGAMBITcurvaturesize function [16]wasusedtoeasily
generatedifferentsurface meshesexhibiting agranularity
fromcoarsetofine.Thisfunction constrainsthe anglebe-
tween outward-pointing normalsforany twoadjacentsurface
triangles.That isparticularly useful inthe caseofhighly
curvedsurfaceslike aneurysmgeometries.Hence,regions
with high curvature aremeshedinmoredetail thanthose
withlowercurvature.Fourparametersdefinethe curvature
size function i.e.anglein degree,growthin percentage,max.
and min.trianglesize.TableIcontainsadetailed overview
ofthe appliedcurvaturesize function parameters.Givena
certain geometryand certain parametervalues,themeshing
algorithmwill automaticallychoosethefinalnumberof
trianglesforthesurface mesh.
Thenumberoftetrahedralmeshelementsforthewhole
volumedependson theresolution ofthegivensurface mesh
-thelargerthenumberoftriangles,representing thesurface,
thelargerthenumberoftetrahedralmeshelements.
Mesheswereimportedintothesimulation softwareFluent
(ANSYSInc.).Foreachtetrahedralmesha corresponding
polyhedralmeshwasgenerated by a conversion algorithm
partoftheFluentCFDsolver.Thesurface ofthe aneurysm
and theparentarteryweremodeledasrigid-wallsand no
slipas shearcondition.Blood wasmodeledasanincom-
pressibleNewtonianfluidwithadensity of1050kg/m3and
aviscosity of0.004 N/m2s[17].Theboundaryconditions
forall conductedsimulationswere asfollows: theinletwas
consideredasavelocityinletand all outletsweremodeled
aspressureoutletzero.Aconstant inflowrateof0.5m/s
wasappliedinthesteady simulations.We considereda
steady-statesimulation asconverged,iftherelativeresiduals
fallsunder0.001 (i.e.the absolutevaluesoftheresiduals
werereduced by three ordersofmagnitude).Aseriesof
steady-statesimulationswereperformedaccording tothe
meshes shownintableIin ordertoeliminatetheissueof
unsuitableCFDmeshesand tocomparetheresultsgiven
frompolyhedraland tetrahedralmeshes.Figure2 gives
an overviewabout thesimulation resultsconcerning WSS
distribution.TheArea-Weighted-AverageWSSdistribution
wasusedtoanalyze theWSSdifferencesbetween polyhedral
and tetrahedralmeshesintermsofnumbers.It isdefinedas
1
AZφdA=
1
A
n
X
i=1
φi|Ai|(1)
2788
whereAdenotesthetotalarea being consideredand φi
describesthewall shearstress associatedwiththefacetarea
Ai.
III.RESULTS
Thesimulation series showstwofindings.1)theWSS
pattern ofpolyhedralmesheslook morehomogeneousthan
theonesofthetetrahedralmeshesasindicated by theyellow
circlesinfigure2.TableI (columns9and 10)depictstheval-
uesofthe averageArea-Weighted-AverageWSSdistribution
considering onlythe aneurysmarea asillustratedinfigure
2.Thedifferencesbetween bothmeshtypesarenegligible
considering thefact thatpolyhedralmeshesexhibit farless
cell elementsthantetrahedralmeshes.2)tetrahedralmeshes
exhibit afarworse convergence asthepolyhedralones(see
tableIcolumn 6 and 8).Onaverage,tetrahedralmeshes
need 369 iterationsmorethanthepolyhedralmeshes.The
highestresolvedtetrahedralmesh(case1 228118 tetrahedral
elements)did notconverge according tothe applied boundary
condition asdescribedinthemethod section.
IV.DISCUS SIO N
Acertainmeshresolution isneededtoresolvetheWSS
distribution on avalid base.TheWSSpatternis similar for
tetrahedral(whenconverged)and polyhedral.However,it
changeswithincreasing spatialresolution and convergesat
a certainlevel(see figure2and tableI).Polyhedralmeshes
can be consideredasaviable alternativetotetrahedral
meshesbecausethey do notonlyshowbettercomputational
convergence [18].Theyare alsoabletoresolvetheWSS
patternwithfarless controlvolumespermeshinamoreho-
mogeneousmannerthanthetetrahedralmeshes.Thereason
forthisliesinthewaytheWSSmagnitudeiscalculated
which dependson twomajoraspects:1)onlythose cell
elementsare consideredwhichactuallyshare a face with
thevesselboundary.Notall tetrahedralelementslocatedat
thevesselboundarysharenecessarilyanentireface withthe
boundary-somemaytouchtheboundarywithitscorner.
While all polyhedralelementslocatedat thewall share an
entireface withtheboundaryitself.2)Thedistancesofthe
consideredtetrahedralcentersarenotequal tothevesselwall
leading toamoreinhomogeneousWSSappearance.
Froma clinicalperspective,the convergence aswell asthe
computation speed ofCFDsimulationsare crucialaspects
regarding future clinicalCFD-based diagnostic and treatment
tools.It isnotfeasibleforthetreating physicianto have
tochangesimulation parametersto optimize convergence
behavior.Ourevaluation has shownthat thepolyhedral-based
meshesaremorestable and fasterand should beusedina
futurestandardizedclinicalsimulation workflow.
Theresultsobtained during this study may be affected
duetoseveral limitations.Ourassumptionsconcerning the
conductedCFDexperimentsdiffer fromtheinvivostatein
termsof rigid vesselwalls,Newtonian-based blood fluidand
thedetermination oftheboundaryconditions,respectively.
Theoutflowsofthepatient-specificmodelsaredefinedas
pressureoutletzerowhich doesnothavetomatchwiththe
realenvironment.Theremightbenaturalresistancesat the
outflows.Aswith othercomputationalstudies,it isassumed
that theselimitationshaveonlyminoreffectson theresulting
flowpattern[19].However,futurework hastofocuson the
reduction oftheselimitation inthesenseofvalidation against
invivomeasurements.
V.CO NCLUSIO N
This study representsthefirstmeshindependencyanalysis
inthefield ofcerebralblood flowsimulation togetherwith
a comparison ofpolyhedraland tetrahedral-basedmeshes.
Here,weput thefocusfromthe actualflowpatternwithin
aneurysmsoritscorresponding WSSpatterntothe evaluation
oftheCFDmesh.Our resultsillustratetheimportance ofa
well-foundedmesh granularityevaluation beforestarting a
blood flowsimulation in orderto getreliableblood flow
simulation resultstosupport the clinicaldecision making.
Thisapproachservesasafirstkeysteptowardsafuture
clinicalCFDapplication wherethemesh generation process
hasto be automatedasmuchaspossible.Furthermore,
theresultswill allowto deviaterequirementsto geometric
modelaccuracyacquired by modernimaging methodsand
to optimize CFDcomputation timewithout loss ofaccuracy.
VI.ACKNOWLED GMENTS
The authorswouldliketothank Dr.RalfKroeger (ANSYS
Germany GmbH) forhisadvice and simulation support.
The authorsgratefullyacknowledgefunding oftheErlangen
GraduateSchool inAdvancedOpticalTechnologies(SAOT)
by theGermanNationalScience Foundation (DFG)inthe
framework ofthe excellence initiative.
REFERENCES
[1]A.Theodorakakos,M.Gavaises,A.Andriotis,A.Zifan,P.Liatsis,
I.Pantos,E.Efstathopoulos,and D.Katritsis,“Simulation ofcardiac
motion on non-newtonian,pulsating flowdevelopment inthehuman
leftanteriordescending coronaryartery,”PhysicsinMedicineand
Biology,vol.53,pp.4875–4892,2008.
[2]C.Karmonik,A.Arat,G.Benndorf,S.Akpek,R.Klucznik,
M.Mawad,and C.Strother,“Atechniqueforimproved quantitative
characterization ofintracranialaneurysms,”American JournalofNeu-
roradiology,vol.25,no.August,pp.1158–1161,2004.
[3]D.Steinman,J.Milner,C.Norley,S.Lownie,and D.Holdsworth,
“Image-basedcomputationalsimulation offlowdynamicsinagiant
intracranielaneurysm,”American JournalofNeurradiology,vol.24,
pp.559–566,April 2003.
[4]M.Castro,C.Putman,and J.Cebral,“Patient-specific computational
fluid dynamicsmodeling ofanteriorcommunicating arteryaneurysms
astudy ofthesensitivity ofintra-aneurysmalflowpatternstoflow
conditionsinthe carotidarteries.”American JournalofNeurradiology,
vol.27,pp.2061–2068,2006.
[5]L.Jou,D.Lee,H.Morsi,and M.Mawad,“Wall shearstress on
rupturedand unrupturedintracranialaneurysms at theinternalcarotid
artery,”American JournalofNeurradiology,vol.29,pp.1761–1767,
October2008.
[6]J.Cebral,M.Castro,J.Burgess,R.Pergolizzi,M.Sheridan,and
C.Putman,“Characterization ofcerebralaneurysms forassessing
risk of ruptureby using patient-specific computationalhemodynamics
models.”American JournalofNeurradiology,vol.26,pp.2550–2559,
2005.
[7]M.Ford,S.Lee,S.Lownie,D.Holdsworth,and D.Steinman,“On
the effectofparent-aneurysmangleon flowpatternsin basilartip
aneurysms:Towardsasurrogategeometricmarkerofintra-aneurismal
hemodynamics,”JournalofBiomechanics,vol.41,pp.241–248,2008.
2789
Case1
Polyhedral Tetrahedral
5119
18681
25540
13841
86621
115014
Case2
Polyhedral Tetrahedral
8814 31696
16795 75469
35665 174225
0.0 Pascal 58.6 Pascal
0.0 Pascal 121.3 Pascal
Fig.2.WSSdistribution forpolyhedraland tetrahedralmeshes.TheWSSpatternscomputedfrompolyhedralmeshesappearmorehomogeneousthan
tetrahedral-basedWSSpatterns(see yellowcircles).Polyhedralmesheswere alsoabletorepresentsignificantWSSpatternwithfewercellsthantetrahedral
onesasmarked by theredcircles.Thenumbersbelowtheindividualfiguresdenotethenumberofcellsofthemesh.
[8]H.Ujiie,Y.Tamano,K.Sasaki,and T.Hori,“Isthe aspectratioa
reliableindexforpredicting theruptureofasaccularaneurysm?”
Neurosurgery,vol.48,p.495502,2001.
[9]A.Nader-Sepahi,M.Casimiro,J.Sen,and N.Kitchen,“Isaspectratio
areliablepredictorofintracranialaneurysmrupture?” Neurosurgery,
vol.54,p.13431347,2004.
[10]M.Raghavan,B.Ma,and R.Harbaugh,“Quantifiedaneurysmshape
and rupturerisk,”JournalofNeurosurgery,vol.102,p.355362,2005.
[11]C.Karmonik,G.Benndorf,R.Klucznik,H.Haykal,and C.Strother,
“Wall shearstress variationsin basilartipaneurysms investigated
withcomputationalfluid dynamics,”inEngineering inMedicineand
Biology,2006.
[12]S.Prakashand C.R.Ethier,“Requirementsformeshresolution in 3d
computationalhemodynamics,”JournalofBiomechanicalEngineer-
ing,vol.123,pp.134–144,April 2001.
[13]C.Karmonik,R.Klucznik,and G.Benndorf,“Blood flowincerebral
aneurysms:comparison ofphase contrastmagneticresonance and
computationalfluid dynamics–preliminaryexperience,”vol.180,no.3,
pp.209–215,2008.
[14]N.Heran,J.Song,K.Namba,W.Smith,Y.Niimi,and A.Berenstein,
“Theutility ofdynact in neuroendovascularprocedures,”American
JournalofNeurradiology,vol.27,pp.330–332,February 2006.
[15]H.Lorensen,W.E.Cline,“Marching cubes:Ahigh resolution 3d
surface construction algorithm,”ComputerGraphics,vol.21,no.4,
pp.163–169,1987.
[16]F.Cooperation,Gambit 2.4User’sGuide.E.I.du PontdeNemours
and Co., 2007.
[17]T.Hassan,M.Ezura,E.Timofeev,T.Tominaga,T.Saito,A.Takahashi,
K.Takayama,and T.Yoshimoto,“Computationalsimulation ofthera-
peuticparentartery occlusion totreatgiantvertebrobasilaraneurysm,”
American JournalofNeurradiology,vol.25,pp.63–68,January 2004.
[18]M.Peric,“Flowsimulation using controlvolumesofarbitrary poly-
hedralshape,”inERCOFTACBulletin,no.62,September2004.
[19]J.Cebral,M.Castro,S.Appanaboyina,C.Putman,D.Millan,and
A.Frangi,“Efficientpipelineforimage-based patient-specific analysis
ofcerebralaneurysmhemodynamics: technique and sensitivity,”IEEE
Transactionson MedicalImaging,vol.24,pp.457–467,2005.
2790