Contents lists available atScienceDirect
Journal
of
Computational
Physics
www.elsevier.com/locate/jcp
A
coupled
3D–1D
numerical
monodomain
solver
for
cardiac
electrical
activation
in
the
myocardium
with
detailed
Purkinje
network
Christian Vergara
a,
∗
,
Matthias Lange
b,
Simone Palamara
a,
Toni Lassila
b,
Alejandro
F. Frangi
b,
Alfio Quarteroni
caMOX,DipartimentodiMatematica,PolitecnicodiMilano,Italy
bCISTIB,DepartmentofElectronicandElectricalEngineering,TheUniversityofSheffield,UnitedKingdom cChairofModellingandScientificComputing,ÉcolePolytechniqueFédéraledeLausanne,Switzerland
a
r
t
i
c
l
e
i
n
f
o
a
b
s
t
r
a
c
t
Articlehistory: Received10April2015
Receivedinrevisedform22November2015 Accepted7December2015
Availableonline21December2015 Keywords:
Computationalelectrocardiology Purkinjefibers
Monodomainequation Pullandpusheffect
Wepresentamodelfortheelectrophysiologyinthehearttohandletheelectrical propa-gationthroughthePurkinjesystemandinthemyocardium,withtwo-waycouplingatthe Purkinje–musclejunctions.Inboththesubproblemsthemonodomain modelisconsidered, whereasatthejunctionsaresistorelementisincludedthatinducesanorthodromic prop-agationdelayfromthePurkinjenetworktowardstheheartmuscle.Weproveasufficient conditionforconvergenceofafixed-pointiterativealgorithmtothenumericalsolutionof thecoupledproblem.Numericalcomparisonofactivationpatternsis madewithtwo dif-ferentcombinationsofmodelsforthecoupledPurkinje network/myocardiumsystem,the eikonal/eikonalandthemonodomain/monodomainmodels.Testcasesareinvestigatedfor bothphysiologicalandpathologicalactivationofamodelleftventricle.Finally,weprove thereliabilityofthemonodomain/monodomaincouplingonarealisticscenario.Ourresults underlietheimportanceofusingphysiologicallyrealisticPurkinje-treeswithpropagation solvedusingthemonodomainmodelforsimulatingcardiacactivation.
©2015ElsevierInc.All rights reserved.
1. Introduction
ThePurkinjefibers areadensenetworkofspecializedcellslocatedundertheinnersurfaceoftheheart(theendocardium) and are responsible for the fast conduction of the activation signal from the atrioventricular node to the heart muscle (myocardium). Theinclusion ofthePurkinjefibers incomputationalmodels ofelectrocardiologyhasbeeninrecentyears recognized asfundamental toaccuratelydescribing theelectricalactivationintheleft ventricle
[1–7]
.Thesefibers forma network,whichrepresentstheperipheralpartoftheconductionsystem.The electrical propagation in the Purkinje fibers has been treated with different mathematical models derived from those commonly usedfor theelectrical propagation inthe myocardium. We citefor examplethe eikonal model [4–6,8], the monodomain model [9,10], and the bidomain model [11,12]. In normal electrical propagation, the electrical signal, originating fromtheatrioventricular (AV) node, travelsalong thisnetwork andenters theventricular muscle through the
*
Correspondenceto:PiazzaLeonardodaVinci32,20133,Milan,Italy.Tel.:+390223994610.E-mailaddresses:christian.vergara@polimi.it(C. Vergara),m.lange@sheffield.ac.uk(M. Lange),simone.palamara@polimi.it(S. Palamara), t.lassila@sheffield.ac.uk(T. Lassila),a.frangi@sheffield.ac.uk(A.F. Frangi),alfio.quarteroni@epfl.ch(A. Quarteroni).
http://dx.doi.org/10.1016/j.jcp.2015.12.016 0021-9991/©2015ElsevierInc.All rights reserved.
Purkinje–musclejunctions (PMJ).Inpathologicalsituations,suchastheWolff–Parkinson–White (WPW)syndrome,thesignal mayenterthemyocardiumfromdifferentregionssothattwofrontspropagateatthesametime,onefromthenetwork to-wardsthemyocardiumandanotheroneintheoppositedirection.Capturingthecouplednatureofpropagationarisingfrom theinteractionbetweenthePurkinjenetworkandthemyocardiumisafundamentalmodeling issue.Inthisregard,different coupledmodelshavebeenconsideredintheliterature.Wecitethecoupledeikonal/eikonalmodel[13,5,6,8](thefirstmodel referstotheoneusedforthenetwork,whereasthesecondtotheoneusedforthemyocardium),theeikonal/monodomain model[4],the monodomain/bidomain model[9],the monodomain/monodomainmodel [10], andthebidomain/bidomain model[11,12].Whenmonodomainorbidomain modelsare considered,theissueofthecouplingbetweencardiac muscle andPurkinje network shouldbe properly addressed. Indeed,dueto the parabolicnature ofthese models,an explicit al-gorithm based onthe successivesolution ofthe propagationin the networkandin the muscleonly once per time step could notreproduceaccurate resultswhen multiplefrontsare propagating(ashappensforWPW).Theworkscitedabove consideredexplicitcouplingstrategieswhichgiveaccurateresultsonlyforanormalpropagation.
Inthiswork,westartfromthemonodomainmodelproposedforthePurkinjenetworkin
[9]
andconsiderthecoupling withthe monodomain model in themyocardium, obtaining a monodomain/monodomain coupled problem(Sect. 2). We observethatwiththemonodomainformulationwearenotabletostudysomespecificfeatureofpathologicalpropagations, suchastheheartfibrillation.However,thismodelisabletocapturemanycharacteristicfeaturesoftheelectricalpropagation in the heart, in particular it is suited in view of the electro-mechanical coupling. Since it allows to highly reduce the computationaltime withrespecttothebidomain model,inthisworkwechose thissimplifiedmodel.Thisallowedusto perform a well-posednessanalysis ofthe coupledproblem(difficultly applicable to the bidomain context). Moreover, we introduce a semi-implicit time discretization andan iterative algorithm forthe solution of the coupled problem arising ateach time step(Sect.3). We observethatthe proposed algorithmallows ustotreat implicitlythecouplingconditions betweenPurkinjenetworkandcardiacmuscle,thussolvingatrulycoupledproblemwhichisabletodescribealsosituations wheremultiplefrontspropagate.Finally,wepresentseveralnumericalresultswiththeaimofassessingtheeffectivenessoftheproposedalgorithmand comparing the solutions with the onesobtained withthe eikonal/eikonal model.In particular, we discussthe choice of the conduction velocities in the eikonal andmonodomain models required to obtain comparable results (Sect. 4.1). We alsoperformacomparisonforabenchmarktestbetweentheeikonal/eikonalandthemonodomain/monodomainstrategies, highlightingthe“pullandpush”effect(Sect.4.2)andthenconsiderbothnormalandpathological(WPWsyndrome) prop-agationsinan idealellipsoidalmodeloftheventricle (Sect.4.3).Interestingly,withouralgorithmwe areabletorecover, withoutanyaprioriimposition,someofthemoreinterestingfeaturesoftheelectricalpropagationintheheart,suchasthe pullandpusheffect,andthedelayatthePMJ.Finallywe applythemonodomain/monodomainmethodologytoarealistic case(Sect.4.4).
2. Mathematicalmodelsfortheelectricalactivation
Inthissectionweprovidethemathematicalmodelsconsideredinthisworkforthedescriptionoftheelectricalactivation inthemyocardiumandinthePurkinjenetwork,andthecorrespondingcoupledproblem.Wewillusethesubscriptsm and p tocharacterizethequantitiesrelatedtothemyocardiumandtothePurkinjenetwork,respectively.
2.1. Activationinthemyocardium
2.1.1. Monodomainmodelinthemyocardium
Thebidomain model, whichaccountsforthepropagationoftheextra- andintra-cellular potentials(see,e.g.,
[14–16]
), isthe mostcommonlyusedmodel todescribe theelectrical activation inthe myocardium.To reducethe high computa-tionalcostsassociatedtousingthebidomainmodel,thesimplermonodomain model,whichdescribestheevolutionofthe transmembranepotential Vm inthemyocardiumdomainm,isoftenused.Itreadsasfollows:
GivenVm,0andwm,0,findVm
:
m× (
0,T]
→ R
andwm:
m× (
0,T]
→ R
dm,suchthat⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩
χ
mC
m∂
Vm∂
t+
I m ion(
Vm,
wm)
− ∇ · (∇
Vm)
=
I inm
× (
0,
T),
dwm dt=
fm(
Vm,
wm)
inm
× (
0,
T),
(
∇
Vm)
n=
0 on∂
m× (
0,
T),
Vm(
x,
0)
=
Vm,0(
x),
wm(
x,
0)
=
wm,0(
x)
inm
,
(1)where
istheconductivitytensorgivenby
(
x)
=
σ
tI+
σ
f−
σ
taf
(
x)
af(
x)
T,
σ
t andσ
f aretheconductivitiesintheorthogonalandlongitudinaldirectionswithrespecttothefibers,andaf istheunitvectoralignedwiththefibers.
χ
misthesurface-to-volumeratioofthecellmembrane,C
misthemembranecapacitance,Iionmrepresentstheioniccurrents(more precisely,currentdensitiespersurface unit), wm istheunknownvectorthat includes
thegatingandionconcentrationvariablesoftheODEsystemrepresentingasuitablecellmodel,andthevectorialfunction fmisanon-lineartermwhichdeterminestheevolutionofwm.Wehaveconsideredtheno-fluxboundarycondition
(1)
3on the ventricle[1]
.The forcingterm I,representingan externalcurrent(moreprecisely,a currentdensitypervolumeunit), willbespecifiedoncewecouplethissystemwiththe1Dmonodomainone,seeSection2.3.1.Forthesakeofexposition,inwhatfollowswewillcompactlywriteproblem
(1)
asfollowsPm
(
Vm,
wm,
I)
=
0.
The monodomain model is based on the assumption of equal anisotropy ratio between the intra- and extra-cellular domains. If thereis no injectionof currentintothe extracellular domain,this modelis indeeda good approximation of themorecomplexbidomainone
[17,18]
.Noticethat(1)
isacoupledproblem,sincethetransmembranepotentialandthe gating/ionconcentrationvariablesappearinboththedifferentialproblemsthroughthecouplingtermsIionm and fm.
2.1.2. Eikonalmodelinthemyocardium
If one is interested only in the ventricular activation times, defined as the time at which the potential reaches the intermediate value between the maximum and the resting potential [19,20], then a further simplified model could be considered, namelytheeikonalmodel, thatprovides ateachpoint theactivation time.Thismodeldiscards allthe cellular kineticsanddescribesonlythemacroscopicspreadingoftheexcitationwavefronts.Assuch,itdoesnotrequireafinespatial resolution,makingitpossibletosimulatetheactivationoflargevolumesofcardiactissueatlowcomputationalcosts.Itis indeedagoodapproximationofthebidomainmodel
[21]
forthecomputationofactivationtimes,whereas itisunsuitable todescribere-entrantphenomenasuchasarrhythmias.Inthisworkweconsidertheanisotropic eikonalequation,whichreads:
Givenum,0,findtheactivationtimesum
:
m→ R
suchthat Cf(
∇
um)
T D∇
um=
1 x∈
m,
um(
x)
=
um,0(
x)
x∈
m,
(2) wherem isthe set of boundary points generatingthe front, D
(
x)
models the anisotropic tensor that accountsfor thepresenceofthemuscularfibers,andCf
(
x)
representsthevelocityofthedepolarizationwavealongthefiberdirection.Weusethefollowingexpression
[20]
D
(
x)
=
k2I+ (
1−
k2)
af(
x)
af(
x)
T,
(3)where k isthe ratiobetweenthe conductionvelocities inthe orthogonalandlongitudinal directionswithrespect to the fibers.
Note thatsince we didnot consideranydiffusivetermin theeikonal problem,our modeldoesnottake into account theeffectsofwavefrontcurvatureortheinteractionbetweenawavefrontwitheitherthedomainboundariesorwithother fronts. This is justified by observing that in our case the myocardial activation is regulated by the Purkinje fibers, and becauseoftheirhighdensity,thediffusiontermgivesasmallcontributionwithrespecttotheadvectionone.
Problem
(2)
canbesolvedveryefficientlybythefastmarchingmethod[22]
andhasbeensuccessfullyusedforclinical applications,see[13,5,6]
.2.2. ActivationinthePurkinjenetwork 2.2.1. Monodomainmodelinthenetwork
Both thecardiomyocytes in themyocardium and theones inthe Purkinje networkare electrically connected by gap-junctions, intercellular channelsproviding a low resistancepathwayfor thespreading ofthe actionpotential
[23]
.Unlike what is usually donein themyocardium where theeffectof thegap-junctions,asa consequenceof thehomogenization process, is hidden in the conductivitytensor D, in [9] the authors proposed to explicitlymodel a gap-junctions in the Purkinje networkasaresistor placedbetweentwoPurkinje cells.Thisallows usto easilywrite theKirchhofflaws atthe bifurcation points ofthe network,sincethepotential andthecurrentinthePurkinjecell/gap junctionunitare treatedas independentvariables.Moreover,ashighlightedin[9]
,this“discrete”approach,inantithesistoahomogenizedone,allows onetodescribethesawtootheffect,see[9]
formoredetails.Fig. 1. Schematic representation of a gap-junction linking two Purkinje cells.
Inthisworkweusethediscretemodelproposedin
[9]
.Asaconsequence,thegap-junctionresistanceneedstobe com-patiblewiththehomogenized conductiontensorofasingle Purkinjecell/gapjunctionunit.Tothisaim,letσ
p∗ denotetheconductivityinthecells,andRg theresistanceoverthegap-junction.Bothtogether determinetheequivalent conductivity
σ
p= (
σ
p∗l)/(
l+
σ
p∗Rgπρ
2)
ofasinglecell/gap-junctionunit,wherel isthelengthofthecellandρ
itsradius.Itisimportanttonotethat,incontrastto
[9]
,thischoiceoftheequivalentconductivitydependsonthephysicalpropertiesofthecelland notonthenumericalparameters(spacediscretizationstep).Wethereforehaveasequenceofelementaryunitscomposedby twoPurkinjecellsconnectedbyagapjunction,which are characterized by the same spatial coordinates (see Fig. 1). Each of these units is characterized by the extra-cellular andintra-cellularpotentialsandby thecurrentsrelatedto thecellsattheleft andattheright(identifiedwiththeindex
−
and+
, respectively) and to the gap junction (identified with the index g). We assume here that the extra-cellular potentialφ
e is constant for each unit, so that we can consider the transmembrane potential as the effective potentialunknown.Thus,foreachunittheunknownsoftheproblemarethetransmembranepotentialsVg
,
V+p,
V−p andthecurrentsIg
,
I+p,
I−p.We assume that thebifurcation andintersection points ofthe networkare located incorrespondenceof some ofthe gap-junctions.Kirchhofflawsatthegenericbifurcationorintersectionyield
p
j=1
Ig,j
=
0,
Vg,1= . . . =
Vg,p,
(4)where p is thenumberof branchesissuingfromthebifurcation orintersection andwe adopttheusual convention that enteringcurrentsarepositiveandexitingcurrentsarenegative.
FromOhm’slawatthegap-junctions,wealsohave
Ig
= ±
Vg
−
V±pRg
/
2.
(5)TheintracellularcurrentI±p thatflowsinthePurkinjecellcanbewrittenas
I±p
= −
πρ
2σ
p∂
Vp±∂
l,
where
ρ
istheradiusofthePurkinjecell,andσ
p istheequivalentintracellularconductivity[9]
.Thankstotheconservationofcurrentsatthegap-junction, Ig
=
I+=
I−,wehaveIg
= −
πρ
2σ
p∂
V+p∂
l=
πρ
2σ
p∂
V−p∂
l.
(6)Tosummarize,themonodomainmodelwithgap-junctionsinthePurkinjenetworkisgivenbythemonodomainequation writtenin eachsegment ofthenetwork,together withtherelations atthe gap-junctions
(5)
–(6) andwiththe continuity relationsatthebifurcation/intersectionpoints(4)
:GivenVp,0
,
wp,0andhA V,findV±p,i:
Si× (
0,T]
→ R,
Vg,i:
Si× (
0,T]
→ R
andwp,i:
Si× (
0,T]
→ R
dp,
i=
1,. . . ,
P ,such⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩
χ
pC
p∂
V±p,i∂
t+
I ion p(
Vp±,i,
w±p,i)
−
∂
∂
lσ
p∂
Vp±,i∂
l=
0 in Si× (
0,
T],
i=
1, . . . ,
P,
∂
w±p,i∂
t+
fp(
V ± p,i,
w±p,i)
=
0 in Si× (
0,
T],
i=
1, . . . ,
P,
Vg,i=
V+p,i+
Ig,2iRg=
Vp−,i−
Ig,2iRg in Si× (
0,
T],
i=
1, . . . ,
P,
Ig,i= −
πρ
2σ
p ∂V+p,i ∂l=
πρ
2σ
p ∂V−p,i ∂l in Si× (
0,
T],
i=
1, . . . ,
P,
ik pk i=ik1Ig,i=
0 at bk,
k=
1, . . . ,
P,
t∈ (
0,
T],
Vg,ik 1= . . . =
Vg,ikpk at bk,
k=
1, . . . ,
P,
t∈ (
0,
T],
−
π σ
pρ
2∂
Vp±∂
l(
s0)
=
hA V t∈ (
0,
T],
−
π σ
pρ
2∂
Vp±∂
l(
sj)
=
Nj j=
1, . . . ,
N,
t∈ (
0,
T],
V±p=
Vp,0(
x)
inp
,
w±p=
wp,0(
x)
inp
,
(7)where Siarethesegmentsofthenetworksuchthat
iP=1Si=
p,p beingthePurkinjenetworkdomain,
χ
p thesurface-to-volumeratioofthecellmembrane,Iion
p theioniccurrents(orcurrentdensitiespersurfaceunit),
C
p(
x)
isthemembranecapacitance,l isthecurvilinearcoordinatealongthenetwork,s0thecoordinateoftheatrioventricularnode,sj
,
j=
1,. . . ,
N,thecoordinatesofthePMJ,bk thecoordinatesofthebifurcationandintersectionpoints,andik1
,
. . . ,
ikpk arethe pk indicesrelatedtothe potentialsandcurrentsinvolvedatthebifurcation/intersection pointbk.Equations
(7)
8 representNeumann boundary conditionsatthePMJ, which areeitherinlets oroutletsforthe system.We leave forthemomentthe data Njunspecified:theywillbeprovidedbythecouplingwiththemyocardialactivation,seeSect.2.3.1. Forthesakeofexposition,inwhatfollowswewillcompactlywriteproblem
(7)
asfollowsPp
(
V+p,
Vp−,
Vg,
Ig,
w+p,
w−p,
N)
=
0,
wheretheunknownsaredefinedgloballyinallthenetworkstartingfromtheirvalueoneachsegment Si.
Acomputational convergenceanalysisofthenumericalsolutiontowards theexactoneforproblem
(7)
inthePurkinje network hasbeen performed by us in [24]. This is the first attempt to validate the fullydiscrete representation ofthe network given by gap-junction/Purkinje cell units. The results showed convergence of the solution both for steady and pulsatiletestcases.2.2.2. Eikonalmodelinthenetwork
Inthecaseofanetworkofone-dimensionallinesegmentsrepresentingthePurkinjefibers,we canconsideragainthe eikonalmodelwithoutdiffusion:
Givenup,0,findtheactivationtimesup
:
p→ R
suchthat⎧
⎨
⎩
Cp∂
up∂
l=
1 x∈
p,
up(
x)
=
up,0(
x)
x∈
p,
(8) wherep isthesetofpointsgeneratingthefrontinthenetwork(forexample,inanormalpropagation,theAVnode)and
Cp theconductionvelocity(5–10timesgreaterthanthemuscularone
[25]
).Again,weneglectthediffusiontermsincethehighadvectiontermVp dominatesanydiffusionprocess.
2.3. Coupledproblems
ThePurkinjefibersformasubendocardialnetworkcharacterizedbyahighconductionvelocityandareisolatedfromthe muscle,exceptattheirendpoints,thePMJ,whicharelocatedontheendocardium.ThroughthePMJ,thesignalcouldeither entertheventriclefromthenetwork,asinanormalpropagation(orthodromicpropagation), orenterthenetworkfromthe myocardium, as happensforsome pathologicalconditions (antidromicpropagation), see,e.g.
[6]
. Inboth cases a delay at thePMJisobserved,inparticularanorthodromicdelaydo ofabout5–15 msandanantidromicdelayda ofabout2–3 ms [26,27].Thus,wehaveacoupledproblembetweentheelectricalpropagationinthe1Dnetworkandinthe3Dmyocardium wherethecouplingpointsarethePMJ.In what follows, we describe two possible coupled strategies, namely the monodomain/monodomain (MM) and the eikonal/eikonal(EE)ones.
Fig. 2. Schematicrepresentationofagenericmyocardialdomainmandofagenericnetworkp.Thenodes0representstheAVnode,whereasthenode
s1ands2arethePMJ,whichactassourcetermsforthemyocardiumthroughthespheresofradiusr centeredinthePMJ.
2.3.1. Monodomain/monodomaincoupling
The MM strategy hasbeen introduced in
[9]
,and isbased on using(1) forthe myocardium and(7)for thePurkinje network. However, in that work the authors considered an explicit coupling between the two subproblems which was basedontheirsequentialsolution(networkfirstandthenmyocardium)onlyoncepertimestep.Thisdidnot allowedthe authorstotreatcaseswhere,besidesthefrontpropagatingfromtheAVnode,otherfrontsoriginatefromthemuscleasin pathologicalconditions.Oneofthemajornoveltiesofthepresentworkistoconsideranimplicitcouplingbetweenthetwo subproblems,asdetailedinwhatfollows.Towritethecoupledsystem,we needtointroduceamodeldescribingthepropagationoftheelectrical signalthrough thePMJ.Fromhistologicalinspection,PMJsappeartobecomposedbytransitionalcellsconnectingtogether thedistalpart ofthe Purkinje fibers andthesurrounding myocardialcells [28]. Adetailedmodel ofthe PMJis presentedin [29], with the aimofstudyingthe conductiondelay atthe PMJ. However, inthiswork we considera simplermodel,based onthe introduction ofa PMJ resistance
[11,9]
,which provides a good approximation ofthe real behavior ofthe PMJ asshown in[30].Theinfluenceofthe PMJonthetwo subdomains(the myocardiumandthePurkinje network)hasbeen modeled intermsofexchange ofcurrents.Ononehand,thePMJs actassources forthemyocardiumthrough regions ofinfluence modeled asspheresofradiusr centeredinthePMJforasuitable r (seeFig. 2
)[11]
.Ontheotherhand,thePMJsprovide thecurrenttothenetworkthroughtheprescriptionofNeumannboundaryconditionsforproblem(7)
(rememberrelation(7)4).
Asdiscussed,thePMJhasbeenmodeledasaresistanceelement,sothatthecurrent
γ
j atthe j-thPMJcanbewrittenthankstotheOhm’slawasfollows
γ
j=
V+p(sj)+V−p(sj) 2−
1 Ar Br(sj) Vmdx RPMJ j=
1, . . . ,
N,
t∈ (
0,
T],
(9)where
B
r(
sj)
istheballofradiusr centeredatthepointsj, Ar thevolumeofthisballandRPMJtheresistanceofthePMJ(supposedto bethe sameforallthe PMJ).The potentialappearing atthenumeratorofthe previousequation isnothing butthe jump betweenthe Purkinjenetwork potential and themyocardial potential atthe PMJ. Notice that the value of the potential fromthePurkinje network side havebeen chosen astheaverage ofthe two potentials V+p and Vp− atthe terminalnodeofthenetwork(thePMJ),sincealsoherewehaveusedagap-junctionmodel.Instead,thepotentialfromthe myocardiumsidehasbeencomputedastheaverageofthemyocardialpotential Vm overtheballinvolvedintheexchange
ofthecurrent.
Summarizing,byusingthenotationintroducedintheprevioussubsections,thecoupledMMproblemreadsasfollows: Findforeacht,V+p
,
V−p,
Vg,
Vm,
Ig,
wp+,
w−p,
wm+wm− andγ
j,
j=
1,. . . ,
N,suchthat⎧
⎪
⎪
⎨
⎪
⎪
⎩
Pm Vm,
wm,
N j=1A1rI
Br(sj)γ
j+
I ext=
0,
Pp V+p,
V−p,
Vg,
Ig,
w+p,
w−p,
γ
=
0,
PPMJ V+p,
V−p,
Vm,
γ
=
0,
(10)where PPMJ
=
0 representsrelations(9)
,I
Y isthecharacteristicfunctionrelatedtotheregionY⊂
m,andIextanexternalcurrent.
2.3.2. Eikonal/eikonalcoupling
Adifferentstrategyconsistsincouplingtheeikonalproblems
(2)
and(8)
(EEstrategy).Againthecouplingisprovidedat thePMJ,sothatthesetm in
(2)
andp in
(8)
couldcontainalsosomeofthePMJ.UnliketheMMstrategy, inthiscaseitwasnecessarytoidentifytheorthodromicPMJs,that istheonesthat bringthe signalfromthenetworktothemyocardium,andtheantidromicPMJsthatbringthesignalfromthemyocardiumtothe net-work.Indeed,thesolutionsoftheeikonalproblemsrepresentfrontspropagatingfromtheirsourcepoints.Then,inourcase wehadingeneraltwofronts,onecomingfromtheAVnodeandanotheronegeneratedinthemyocardiumdueto patholog-icalconditions(suchastheWPWsyndromeortheleftbundlebranchblock).Wereferthereaderto
[6,8]
forfurtherdetails.3. Numericalsolutionofthemonodomain/monodomaincoupledproblem
In this section we propose an algorithm for the numerical solution of the MM coupled problem (10). In particular, in Section 3.1we firstintroduce thetime discretization followedby afixed point algorithm,whose convergenceanalysis is carried out inSection 3.2. Finally,in Section 3.3we provide details aboutthe numericalsolution ofthe monodomain subproblemsarisingateachiterationofthefixedpointalgorithm.
ForthenumericalsolutionoftheEEcoupledproblemweadoptherethestrategyproposedin
[8]
foranormal propaga-tion,andextendedtotreatalsopathologicalconditionsin[6]
.Wereferthereadertotheseworksforfurtherdetails. 3.1. NumericalalgorithmForthe3Dproblem
(1)
1weproposeasemi-implicittimediscretization,withthediffusivetermtreatedimplicitlythrough thebackwardEulermethod,andthecouplingtermImiontreatedexplicitly.Theequation(1)
2isdiscretizedwiththeforward Eulermethod:⎧
⎨
⎩
χ
mC
mt Vm
− ∇ · (∇
Vm)
=
χ
mC
mt V n m
−
χ
mImion(
Vmn,
wnm)
+
I inm
,
wm=
wnm−
t fm(
Vmn,
wnm)
inm
,
(11) wherewehavedroppedthecurrentindexn+
1 intheunknownsonthelefthandsideforthesakeofsimplicity.Thesameapproachwasconsideredforthetimediscretizationofthe1Dproblems
(7)
1 and(7)
2:⎧
⎪
⎪
⎨
⎪
⎪
⎩
χ
pC
pt V ± p,i
−
∂
∂
lσ
p∂
V±p,i∂
l=
χ
pC
pt Vp±,i n
−
χ
pIionp V±p,i n,
w±pn in Si
,
i=
1, . . . ,
P,
w±p,i=
w±p,in−
t fpV±p,in,
w±p,in in Si,
i=
1, . . . ,
P.
(12)Withthisinmind,wecanintroducesuitableoperators
PmandPp andcompactlywritethediscretized-in-timeproblems (11)and(12)
asPm(
Vm,
I)
=
0 andPpV+p
,
V−p,
Vg,
N=
0,respectively(N isagaintheNeumanndataprescribedatthe PMJ).Notice thatwedidnotexplicitlyindicatethedependenceoftheprevious operatorson wm,
w+p,
w−p sincethey arenot involved directlyin thecoupling, andthat thedependence of
Pp andPm on thequantities atprevious time step isunderstood.Thisallowsustowritethediscretized-in-timeversionoftheMMproblem
(10)
asfollows:Findforeachn,Vp+
,
V−p,
Vg,
Vm,
Igandγ
j,
j=
1,. . . ,
N,suchthat⎧
⎪
⎪
⎨
⎪
⎪
⎩
Pm Vm,
N j=1A1rI
Br(sj)γ
j+
I ext=
0,
Pp Vp+,
V−p,
Vg,
γ
=
0,
PPMJ V+p,
V−p,
Vm,
γ
=
0.
(13)Forthesolutionofthediscretized-in-timeMMcoupledproblem
(13)
we proposeafixedpoint strategy,whereateach iteration thecurrentsγ
j computedatthepreviousiteration areusedto solvethe3D andthe1D problems,andthen thevaluesofthepotentialsareusedtoupdatethePMJcurrents.Thisideaissummarizedin
Algorithm 1
reportedbelow. 3.2. AnalysisIn this section we provide a convergence analysisof Algorithm 1in the particular caseof
p composed by a single
branch,thuswithnobifurcations/intersections.Thankstothetimediscretizationused,ateachtimestepweobtainalinear problem, hencewe can restrict ourselves to analyzethe convergence towards the nullsolution in the caseof vanishing forcing terms. We can therefore set Iext
=
0,hA V=
0, and nullinitial conditions, and set to zero the quantities at theprevious time step. Moreover, we notice that since we do not have anybifurcation/intersection points, there are no Vg
and Ig inthis caseandwehave only Vp to describe thecell potential, insteadof V+p and Vp−. Finally,we cantranslate
the solution Vp and Vm into Vp
=
0 and Vm=
0 correspondingtotherestingpotential conditions.Moreover, weassumethattheioniccurrentsarezerowhenthetransmembranepotentialequalstherestingpotential.Inview oftheanalysis,we introducetheweakformulationsofthemonodomainproblems.Thus,ourfixedpointstrategycanberewrittenasreported in
Algorithm 2
.Algorithm1Solutionofthediscretized-in-timeMMcoupledproblem.
Let k be the iteration index within each time step. Set k=0 and γj(0) =γ0,j :=
V+p n (sj)+ V−p n (sj) 2 − 1 Ar Br(sj) Vn mdx RPMJ , j=1,. . . ,N, with V+p n ,V−p n ,Vn
mtheconvergedsolutionattheprevioustimestep,andchooseatoleranceε>0;thenatiterationk+1
whileγ(k)−γ(k−1) >ε
1. Solvethediscretized-in-timemonodomainproblem(1)inthemyocardiumwithappliedcurrentsgivenbyγ(k),thatis
Pm ⎛ ⎝Vm(k+1), N j=1 1 ArIBr(sj) γ(jk)+Iext ⎞ ⎠ =0;
2. Solvethediscretized-in-timemonodomainproblem(7)inthePurkinjenetworkwithNeumannboundaryconditionsatthePMJgivenbyγ(k),thatis
Pp (V+p)(k+1), (V−p)(k+1),V(gk+1),γ(k) =0; 3. Compute γ(k+1) j = V+p (k+1)(s j)+ V−p (k+1)(s j) 2 − 1 Ar Br(sj) Vm(k+1)dx RPMJ , j=1, . . . ,N; (14) 4. Setk=k+1. end
Algorithm2SolutionofthereducedMMdiscretized-in-timecoupledproblem. Letk betheiterationindexwithineachtimestep.Setk=0 andγ(0)=γ
0,andchooseatoleranceε>0;thenatiterationk+1 whileγ(k)−γ(k−1) >ε
1. Solvethefollowingdiscretized-in-timemonodomainprobleminthemyocardiumwithappliedcurrentsgivenbyγ(k):
FindVm∈H1(m)suchthat
m χmCm Vm(k+1) t Wmdx+ m D∇Vm(k+1)· ∇Wmdx= N j=1 1 Ar Br(sj) γ(k) j Wmdx, (15) forallWm∈H1(m);
2. Solvethefollowingdiscretized-in-timemonodomainprobleminthePurkinjenetworkwithNeumannboundaryconditionsatthePMJgivenbyγ(k):
FindVp∈H1(p)suchthat
p χpCp V(pk+1) t Wpdl+ p σp ∂V(pk+1) ∂l ∂Wp ∂l dl= − 1 πρ2 N j=1 γj(k)Wp(sj), (16) forallWp∈H1(p);
3. Computethevalueofγ(jk+1)with(14);
4. Setk=k+1. end
Thecoupledproblemin
Algorithm 2
canberewrittenasfollows⎧
⎪
⎪
⎨
⎪
⎪
⎩
Vm(k+1)=
Fm(
γ
(k))
inm
,
V(pk+1)=
Fp(
γ
(k))
inp
,
γ
(k+1)=
F PMJ V(pk+1),
Vm(k+1),
where Fm
: R
N→
H1(
m),
Fp: R
N→
H1(
p),
and FPMJ:
H1(
m)
×
H1(
p)
→ R
N providetheexplicitexpressions oftheunknowns obtainedfrom (15), (16) and(14). Algorithm 2 can be written in compact formas the following fixed point iteration
where
F
: R
N→ R
Ns.t.F
(
γ
)
=
FPMJ
(
Fp(
γ
),
Fm(
γ
)).
Toprovetheconvergenceofthepreviousiterations,weneedtoshowthatthereexistsaconstantC
∈ [
0,1)suchthatF
(
γ
(k))
≤
Cγ
(k)∀
k,
(17)foreach
γ
(0),where·
istheusualEuclideannorm.Thisiswhatisprovedinthefollowingresult.Proposition1.Underthefollowingassumptions: – Thereexisttwoconstants0
<
b<
B suchthatb
ξ ≤ ξ
tD(
x)ξ
≤
Bξ
2,
∀ξ ∈ R
2,
(18)fora.e.x
∈
m;– Theparameters
σ
pandb satisfyσ
p≥
4N3/2C2 Tπρ
2R2 PMJ,
b≥
4N 3/2 A3r/2R2PMJ,
(19)whereCTisthetraceconstantfortheSobolevspaceH1
(
p)
;– Thetimestep
t
>
0 ischosensuchthatt
≤
minχ
pC
pσ
p;
χ
mC
m b;
(20)then,thereexistsaconstantC
∈ [
0,1)suchthat(17)
issatisfied.Proof. Fromthedefinitionof
F
wecanwriteF
(
γ
(k))
2
=
F PMJ(
Vm(k+1),
V(pk+1))
2=
N j=1 V(pk+1)(
sj)
−
A1r Br(sj)V (k+1) m dx RPMJ 2≤
2 R2PMJ N j=1⎛
⎜
⎝
V(pk+1)(
sj)
2+
⎛
⎜
⎝
A1 r Br(sj) Vm(k+1)dx⎞
⎟
⎠
2⎞
⎟
⎠ ,
(21)whereweusedtheinequality
(
a+
b)
2≤
2a2+
2b2 a,
b∈ R.
Regarding thefirst termatthe righthandsideof
(21)
,we canapply thetrace theorem(see [31]). Wenoticethat inour casetheboundaryoftheprobleminthenetworkisgivenbythePMJsj andbytheAVnode s0,sowehaveN
j=1 Vp(k+1)(
sj)
2≤
N j=1 Vp(k+1)(
sj)
2+
V(pk+1)(
s0)
2≤
CTV(pk+1)2H1( p).
(22)Regardingthesecondtermattherighthandsideof
(21)
,weusethefollowinginequalityholdingforeveryboundeddomainand0
≤
p≤
q≤ ∞
: zLp()≤ ||
1
p−1q
zLq()
,
providedthatz
∈
Lq()
andwhere||
isthesizeofthedomain.Inourcaseweset=
B
r(
sj),
p=
1,q=
2,soweobtain Vm(k+1)L1(Br(s j))=
Br(sj)|
Vm(k+1)|
dx≤
ArVm(k+1)L2(Br(s j)).
(23)Therefore,wehavethefollowingestimateforthesecondtermattherighthandsideof
(21)
N j=1⎛
⎜
⎝
A1 r Br(sj) Vm(k+1)dx⎞
⎟
⎠
2=
1 Ar2 N j=1 Vm(k+1)2L1(B r(sj))≤
Ar A2 r N j=1 Vm(k+1)2L2(Br(s j))≤
1 Ar N j=1 Vm(k+1)2L2( m)=
N Ar Vm(k+1)2L2( m)≤
C1Vm(k+1)2H1(m),
(24)withC1
=
ANr.Then,owingto(22)
and(24)
,(21)
readsF
(
γ
(k))
2≤
2 R2PMJ CTV(pk+1)2H1( p)+
C1V (k+1) m 2H1( m).
(25)Now,wehavetofindsuitableestimatesfortherighthandsideof
(25)
intermsofγ
(k).Tothisaim,wetake W p
=
V(pk+1)asatestfunctionin
(16)
obtainingχ
pC
pt Vp(k+1)
2 L2( p)
+
σ
p∂
V(pk+1)∂
l 2 L2(p)= −
1πρ
2 N j=1γ
j(k)Vp(k+1)(
sj).
(26) Thus,wehave C2V(pk+1) 2 H1( p)≤
NCTπρ
2 V(pk+1)H1( p) N j=1
|
γ
j(k)|,
withC2=
min{
χpCtp;
σ
p}
,andthen V(pk+1) H1( p)≤
C3γ
(k),
(27) withC3=
N 3/2C T C2πρ2.We proceed now by considering the equation inthe myocardium
(15)
, andwe take Wm=
Vm(k+1) asa test function,obtainingfrom
(18)
theestimateχ
mC
mt V (k+1) m 2L2(m)
+
b∇
V (k+1) m 2L2(m)≤
N j=1 1 Arγ
j(k) Br(sj) Vm(k+1)dx.
(28)Then,owingto
(23)
,wehaveN
j=1 1 Arγ
j(k) Br(sj) Vm(k+1)dx=
1 Ar N j=1|
γ
j(k)|
Vm(k+1)L1(B r(sj))≤
√
Ar Ar N j=1|
γ
(jk)|
Vm(k+1)L2(Br(s j))≤
1 Ar N j=1|
γ
j(k)|
Vm(k+1)L2( m)=
N Arγ
(k)V(k+1) m H1( m)
.
Thepreviousinequalitytogetherwith
(28)
gives Vm(k+1) H1( m)≤
C4γ
(k),
(29) withC4=
!
N Ar 1 min{χm Cm t ;b} .Thus,puttingtogether
(25)
,(27)
and(29)
,weobtain(17)
withC
=
2 R2 PMJ(
CTC3+
C1C4)
=
2 R2 PMJN3/2C2T min
{
χpCp t;
σ
p}
πρ
2+
N Ar 3/2 1 min{
χmCm t;
b}
.
Dueto(20)
,weobtain C=
2 R2PMJN3/2C2 T
σ
pπρ
2+
1 b N Ar 3/2,
whichislessthanonebecauseof
(19)
.2
Remark1. Wenotice that theassumptions on the parameters
σ
p andb given by(19)
depend on thevalue of thetraceconstant CT,which isnotcomputableforgeneraldomains.Thereforewe cannotdetermineexplicitlythevalue of
σ
p andb that guarantee that
F
is a contraction. Nevertheless, in all the numerical experiments reported in what follows, we experienced thattheproposed algorithmnot onlyconverges,butitdoesso(withinmachineaccuracy) inafinitenumber ofiterations.Remark2.Therestrictionon
t givenby
(19)
shouldbematchedwiththeonerequiredforstability oftheforwardEuler methodsfortheODEsystems(1)
2 and(7)
2.Thus,theeffectivet isthesmallerofthesetwo.
3.3. Solutionofthestand-alonesubproblems
In thissection we detail thenumerical strategiesused to solvethe 3D andthe 1D monodomainsubproblems arising at each iteration of
Algorithm 1
.For the solutionof the3D subproblem,we consider Lagrangian finite elements andan implicit/explicit method,see [1]. Forthe solutionof the1D subproblem we followthe methodology presentedin [9]. In particular,we assumetohaveasystemofgap-junction/Purkinjecellsforeachnode ofthemesh.Foreachsegmentofthe network Si,weknowthevaluesofVng,i andIng,iattheprevioustimesteptn.Then,thenumericalschemetocompute Vg,iandIg,i foreachsegment Siattimetn+1 canbedividedintofoursteps:
1. Recoveringthetransmembranepotential
V±p,in.By considering (7)3, we can recover thevalue ofthe transmembrane potentialasfollows: V±p,i n=
Vng,i∓
I n g,iRg 2;
2. Operatorsplitting–firstpart.Wecomputetheintermediatepotentials
Vp±,i n+1/2 asfollows:C
p V±p,i n+1/2−
Vp±,i nt
= −
I ion p V±p,i n,
w±p,i n;
(30)3. Updateof Vg and Ig.We computetheintermediatevaluesVgn+1,i /2 andIng,+1i /2 withthefollowingexpressionsobtained
bymanipulatingthetwoequationsin
(7)
3:Ing+,i1/2
=
V+p,i n+1/2−
V−p,i n+1/2 Rg,
Vng,+i1/2=
V+p,i n+1/2+
Vp−,i n+1/2 2;
(31)4. Operatorsplitting–secondpart.Thesecondpartoftheoperatorsplittingshouldbegivenby
χ
pC
p Vp±,i−
V±p,i n+1/2t
−
∂
∂
lσ
p∂
V±p,i∂
l=
0.
Table 1
Parametersusedinthenumericalexperiments,suitablereferences,andphysiologicalranges.
E–E M–M Ref. Range
χm(cm−1) – 1400 [35] – χp(cm−1) – 1467 [36] – Rg(kOhm) – 500 [9] – r(cm) – 0.06 [11,37] [0.01–0.1] ρ(cm) – 0.0017 [36] – RPMJ(kOhm) – 11000 [9,38] [1000–25 000] ε – 10−5 – – do(ms) 5.0 – [26,27] [5–15] da(ms) 2.0 – [26,27] [2–3] σp(kOhm−1cm−1) – 35.0 [25] – σf (kOhm−1cm−1) – 1.334 [35] – σt(kOhm−1cm−1) – 0.176 [35] –
Now,byaddingthesetwoequationsandbydividingby2,weobtainthanksto
(31)
χ
pC
p Vg,i−
Vng+,i1/2t
−
∂
∂
lσ
p∂
Vg,i∂
l=
0.
(32)As finiteelement basisto solvethe previous problemwe usetheone-dimensional cubicHermitebasis,so thatwe can directlyrecoveralsothederivative ofthepotential,whichisrelatedtothecurrent(recalling
(7)
4).Hermitefiniteelements aresuitableforsuchapurposeastheyarebaseduponsolvingthepotentialanditsderivativeateachnode.Finally,oncewehavedetailedhowtocomputethe valuesof Vg,i andIg,i foreachsingle segmentofthenetwork,we
needtoenforcetheKirchhofflaws
(7)
5–6tocomputetheglobalVgandIg.Tothisaim,wemodifytheglobalfiniteelementmatrixassociatedtothecollectionof
(32)
bysubstituting1’sor0’sintherowsrelatedtobifurcationorintersectionpoints accordinglyto(7)
5–6.4. Numericalexperiments
In thissection we presentseveralnumerical results withthe aim ofassessing the reliability of Algorithm 1to solve the MM coupled problem andcomparing the results withthose obtainedwith the EE coupled problem. First of all, in Section4.1wediscusshowtoestimateaconstantconductionvelocityfromthecoupledmonodomainproblemstobeused in the eikonal ones in view of the forthcoming comparison. After this preliminary step, in Section 4.2 we consider an academictestcasewithsimplifiedgeometriestocomparetheresultsobtainedwiththetwodifferentstrategies,whereas, inSection4.3weapplythesestrategiestosimulatebothanormalandapathologicalpropagationinanellipsoidalidealized leftventricle.Finally,inSection4.4,weapply
Algorithm 1
toarealisticgeometry.AllthenumericalresultsrelatedtotheMMproblemhavebeenobtainedwiththeparallel FiniteElementlibraryLifeV, developed atMOX – Politecnico di Milano, REO/ESTIME – INRIA,CMCS – EPFL, andE(CM)2 – Emory University.For the 3D monodomainproblemweconsidered
P
1 Lagrangianfiniteelements, whereas forthe1D problemcubic Hermitefinite elements.Forboththemonodomainproblems,wechoseatimestept
=
0.01 ms.Theionicmodelsusedinournumerical experimentsweretheDiFrancesco–Noble model[32]
forthePurkinjecells,andtheLuo–Rudy-I model[33]
forthemyocardial cells.ThenumericalschemesforsolvingthecoupledEEproblemhavebeenimplementedinastand-aloneandserialcode based on the VTK 5.8.0 library [6,8]. For the solution of the single eikonal problems, we considered the fast marching method (FMM) proposed in[22] for the1D problem andthe modified version ofthe FMM proposed in [34] forthe 3D problem.Ifnototherwise specified,inallthe numericalexperimentsweused thedatacollected inTable 1,wherewe reported alsosuitablereferencesand,forsome,therangesofthevaluesreportedtherein.
Notice thatwe didnot needtoprescribe explicitlythedelay atthePMJinthe MMmodel,since inthiscasethePMJ resistancemodelitselfwasabletointroducesuitabledelays.
4.1. Assessingtheconductivitiesinviewofthecomparisons
Intheset-upoftheforthcomingnumericaltests,wefaced twocriticalpoints:(i) thechoiceofproperquantitiestobe comparedinviewofadiscussionoftheresults,and(ii)theuseofcomparableconductionvelocitiesinboththeMMand EEcoupledproblems.
The first issue is crucial because the output of the monodomain problem is the transmembrane potential, whereas the one of the eikonal problem is the local activation time. Then, in view of the comparisons, we computed from the transmembranepotentialstheactivationtimesprovidedbythemonodomainproblems,definedagainasthetimeatwhich thepotentialreachesthemeanvaluebetweentherestingpotentialandtheplateaupotential. Thisallowed ustocompare thesevalueswiththeonesprovidedbytheEEproblem.Tothisaim,wedenotedwithuM
Fig. 3. Myocardial domain. Test for the estimation of the conduction velocity.
Fig. 4. Conductionvelocitiesinthemyocardiumasafunctionofthelocalcoordinatel estimatedfromthesolutionofamonodomainproblem:Cx t on A B
(left),CtyonAC (center),andCf onA D (right).Testfortheestimationoftheconductionvelocity.
inthenetworkandinthemyocardium,respectively,obtainedbysolvingtheMM problem,andwithuEp
(
x)
anduEm(
x)
the activationtimesinthenetworkandinthemyocardium,respectively,providedbytheEEstrategy.For what concerns point (ii) above, we neededto use comparable parameters in order to obtain meaningful results. In particular, we remark that the propagationvelocities have a differentnature in the monodomain model than in the eikonal one.Indeed,inthefirstcasetheconductionvelocityoftheelectricalsignalisnotconstant intime andspaceand dependsonthesolution.Forexample,thepropagationvelocitychangeswhentwowavefrontscollideorwhenthewavefront interactswiththeboundaryofthedomain.Onthecontrary,intheeikonalproblems,theconductionvelocityisaprescribed parameterofthemodel,andthereforeitdoesnotdependonthesolutionoftheproblem.However,weobservethatinthe caseofa single wavefront,in themonodomaincasesthe conductionvelocityisalmost constant farfromtheboundaries. Thissuggestsastrategytoestimateareferenceconstantconductionvelocityfromthemonodomainmodel,tobethenused intheeikonalmodel.ThiswasdoneforboththePurkinjenetworkandthemyocardium.
To this aim, we considered two reference scenarios, one for the myocardium given by the cuboid with dimensions 0.3
×
0.7×
2.0 cm,seeFig. 3
,andoneforthenetworkgivenbyasingle Purkinjefiber.Forthemyocardium,weestimated twoconductionvelocities,oneinthedirectionofthefibers(Cf)whichisparallelto A D,andtheotheroneinthedirectiontransversetothefiber(Ct).Todothis,wesolvedthemonodomainprobleminthecuboidwithasourcecurrentappliedin
theinternalcornerofsize0.15 cm withoneofthecornerscoincidingwith A andsidesparalleltotheonesofthecuboid, see
Fig. 3
.ThisallowedtoobtaintheactivationtimeuMm
(
x)
andtodefinethefollowingvelocities Cf(
x)
:=
1 ∂umM ∂z,
Cit(
x)
:=
1 ∂uM m ∂i i=
x,
y.
Then, weevaluatedthesequantitiesalongthethreesegments A B, AC and A D,see
Fig. 4
.Thus,we tookasanestimation oftheconductionvelocitiesCf,mandCt,mprovidedbythemonodomainproblemthemeanvalueofthesequantities,Cf,m
=
1 Nr xi∈A D Cf(
xi),
andFig. 5. ConductionvelocityCp inthePurkinjefibersasafunctionofthelocalcoordinatel estimatedfromthesolutionofthemonodomainproblem.Test
fortheestimationoftheconductionvelocity.
Ct,m
=
1 2⎛
⎝
1 Nx xi∈AC Ctx(
xi)
+
1 Ny xi∈A B Cty(
xi)
⎞
⎠ ,
whereNr,NxandNy arethenumbersofpointsinA D,AC and A B,respectively.Theconductionvelocitiesfoundwiththese
estimateswerethenusedintheeikonalequation
(2)
–(3).Inparticular,weusedCf=
Cf,mandk=
Ct,m/
Cf,m.Referringtothedatareportedin
Table 1
,wefoundCf=
0.067 cm/ms andk=
0.43.ForthePurkinjenetworkweproceededinasimilarway.Inparticular,weconsideredthepropagationofawavefrontina singlePurkinjefiberoflength5 cmbysolvingtheone-dimensionalmonodomainproblem.Weappliedattheleftboundary acurrentstrongenoughtotriggertheexcitationofaPurkinjecell,whereasontherightboundaryahomogeneousNeumann condition.ThisallowedtoobtaintheactivationtimeuM
p
(
x).
Sinceinthiscasewehadonlyonedirectionofpropagation,weestimatedtheconductionvelocityinthesinglePurkinjefiberasfollows
Cp,m
=
1 Np Np i=1 Cp(
xi),
whereNp isthenumberofnodesofthemeshdiscretizingthePurkinjefiberand
Cp isgivenby Cp(
x)
=
1 ∂uMp ∂s.
(33)In
Fig. 5
(right)we depicttheevolutionofCp inthePurkinjefiber.Inparticular,we usedavalueσ
p (seeTable 1
)whichallowed ustoobtain aphysiological valueof theconductionvelocity [25]. Weobservethat, farfromthe boundaries,the conductionvelocitywas almostconstantandequalto0.3 cm/ms,whereas,neartotheboundaries, thewavefront interac-tionsresultedinanon-constant conductionvelocity.Wethususedtheestimatedvalue Cp,m asconductionvelocity Cp in
the1Deikonalproblem
(8)
.For the computations we have used as discretization steps hm
=
0.001 cm for the cuboid, leading to 200 000 nodesand1.1milliontetrahedralelements, andhp
=
0.0165 cm forthePurkinje fiber.Thevalue ofhp waschosen equaltothecharacteristiclengthofaPurkinjecell
[36]
.4.2. Numericaltestinacuboid
InthissectionwereporttheresultsofatestinacuboidforacomparisonoftheEEandMMcouplingstrategiesinthe caseoforthodromic propagation. The myocardialgeometrywas thesame considered intheprevious section (see Fig. 3), whereasforthePurkinjefibersweconsideredasimplenetworkcharacterizedbythreesegmentsandonebifurcationpoint. Thisnetworkliedononesideofthecuboiddomain,similartophysiologicalsituationwherethePurkinjefibersarelocated justbeneaththe endocardium,see
Fig. 6
. Thesignal enters fromtheAV node,represented by s0 inFig. 6
, left,andthen reachesthePMJs1ands2.ThroughthesetwoPMJs,thesignalentersthemyocardium.Forthecomparison, we firstcomputedtheactivation maps,whicharerepresented inFig. 6,right. We noticethatthe MMandEEstrategiesdescribeasimilaractivationpatterninboththePurkinjenetworkandinthemyocardium.Toexamine furtherindetailtheactivationpattern, wealsocomputedthecumulativepercentageofactivatedtissue,whichisdepicted in
Fig. 7
.Notethattheslopeofthisquantitygivesususefulinformationaboutthepropagationvelocityofthewavefrontin bothdomains.WebeginbyanalyzingthePurkinjenetwork.InFig. 7
(left)thepercentageofactivatedtissueinS1andthen S2isrepresented.Inparticular,inthePurkinjenetworktheEEmodelresultsinaconstantconductionvelocitythroughtheFig. 6. RepresentationofthemyocardiumandPurkinjenetworkdomains(left),andactivationmapsinthecaseofMM(top,right)andEE(bottom,right) strategies.Testinthecuboid.
Fig. 7. ComparisonofthepercentageofactivatedtissueinthePurkinjenetwork(left)andinthemyocardium(right)forthetwodifferentcouplingstrategies. Testinthecuboid.
junction,whereastheMMmodelfeaturesadistinctdiscontinuity.Tobetterinvestigatethisphenomenon,wecomputethe followingquantity 1
∂uMp ∂l(
x)
,
(34)whichisanestimateoftheconductionvelocityinthePurkinjenetworkprovidedbythemonodomainproblem.Noticethat thisquantityisdifferentingeneralfromthevalue
(33)
,sincethelatterhasbeencomputedinthecaseofasinglewavefront propagatinginasinglefiber.Werepresentin