Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
1
EVALUATIONOFTHEIMPACTOFSTRAIN-CORRECTIONONTHEORIENTATIONOFCARDIACDIFFUSIONTENSORSWITHIN-VIVOANDEX-VIVOPORCINEHEARTSPedroFFerreira1*,SoniaNielles-Vallespin2,AndrewDScott1,RanildeSilva1,PhilipJKilner1,DanielBEnnis3,
DanielAAuger4,JonathanDSuever5,XiaodongZhong6,BruceSSpottiswoode7,DudleyJPennell1,Andrew
EArai2,DavidNFirmin1
1-CardiovascularBRU,RoyalBromptonHospital,London,UnitedKingdom
2-NHLBI,NationalInstitutesofHealth,MD,UnitedStates
3-DepartmentofRadiologicalSciences,UniversityofCalifornia,LosAngeles,CA,UnitedStates
4-BiomedicalEngineering,UniversityofVirginia,VA,UnitedStates
5-GeisingerMedicalCenter,PA,UnitedStates
6-SiemensHealthcare,GA,UnitedStates
7-SiemensHealthcare,TN,UnitedStates
WordCount:5157
*Addressforcorrespondence:
Dr.PedroFerreira
CardiovascularBRU,RoyalBromptonHospital,
London,SW36NP,UK
Email:[email protected]
2
ABSTRACTPurpose: To evaluate the importance of strain-correcting STEAM-EPI cardiac diffusion tensor imaging
(cDTI).
Methods:Healthypigs(N=11)weresuccessfullyscannedwitha3DcineDENSEandaSTEAM-EPIDTIse-
quenceat3Tduringdiastasis,peaksystole,andstrainsweet-spotsinamid-ventricularshort-axisslice.The
sameDTIsequencewasrepeatedex-vivoafterarrestingthehearts ineithera relaxed (KCl-induced)or
contracted(BaCl2-induced)state.
TheDENSEdatawereusedtostrain-correctthein-vivocDTIindiastoleandsystole.Theorientationofthe
primary(HA)andsecondary(E2A)diffusioneigenvectorswascomparedwithandwithoutstrain-correction
andtothestrain-freeex-vivodata.
Results:Strain-correctionreducessystolicE2Asignificantlywhencomparedtowithoutstrain-correction
andex-vivo[medianabsoluteE2A=34.3ovsE2A=57.1o(P=0.01),E2A=60.5o(P=0.006)respectively].
ThesystolicdistributionofE2Awithoutstrain-correction isclosertothecontractedex-vivodistribution
thanwithstrain-correction,rootmeansquaredeviationof0.027vs0.038.
Conclusion:Thecurrentstrain-correctionmodelamplifiesthecontributionofmicroscopicstraintodiffu-
sionresulting inanover-correctionofE2A.Resultsshowthatanewmodelthatconsiderscellularrear-
rangementisrequired.
Keywords:cardiac;diffusiontensorimaging;strain;microstructure;cardiomyocyte;sheetlets.
3
INTRODUCTIONMyocardialmicro-architectureisremarkablycomplex.Cardiomyocytesexhibitatransmuralhelicalorgan-
izationwithhelixangles(HA)rotatingsmoothlyfrompositiveintheendocardiumtonegativeintheepi-
cardium[1–3].Additionally,cardiomyocyteshavealaminarorganization,wheresheetsofcardiomyocytes
are surrounded by collagen-lined shear layers, as found histologically in many explanted mammalian
hearts[4–7].Thesemyolaminaeplayamajorroleintheradialthickeningandlongitudinalshorteningduring
theheartcycle[8–12].Thecollectiverotationofsheetletsaccommodatedbyshearlayerslippage,driven
bycardiomyocytecontraction,isthedominantcontributortotheradialthickeningandlongitudinalshort-
eningobservedin-vivo.Muchsmallercontributionsarisefromsarcomeredrivenchangesincardiomyocyte
shape[13].Becausemyolaminaearediscreteandexistonlyverylocally,Halesetal.referredtothemas
sheetlets[14].
CardiacDiffusionTensorImaging(cDTI)isanMRimagingtechniquethatencodestheself-diffusionofwater
moleculesinthreedimensions.Adiffusiontensoriscalculatedforeachvoxelandseveraldiffusionrelated
parameterscanbeextracted.Theorientationofthetensoriscloselylinkedtothevoxels’microstructure.
Ithasbeenshownthattheprincipaldirectionofdiffusionalignswiththecardiomyocytelong-axis,while
thesecondaryandtertiarydirectionsofdiffusionprovidecross-cardiomyocytesheetlet-planeandsheetlet-
normaldirectionsrespectively[15–20].Morerecentex-vivocDTIstudieshavebeenperformedwithhearts
inrelaxedandcontractedstates,thusprobingmicrostructuralreorganizationduringcontraction[14,21].
NosubstantialchangesofHAweremeasured,butthesheetletplaneswerefoundtorotatefromalmost
perpendiculartothelocalmyocardialwallinsystoletomorewallparallelindiastole,inagreementwith
currentmodelsofsheetletrotationduringcardiaccontraction.
In-vivocDTIhasattractedrenewedinterestaidedbynewtechnicaladvancementsinbothhardwareand
pulsesequencedesign[22–26].Douetal. in2002reportedin-vivorotationofsheetletstowardsamore
radialconformationduringsystoliccontractioninonehealthyvolunteer[27].Morerecentstudiesconfirm
asimilarpatternofsheetletrotationinacohortofhealthyvolunteersandabnormalsheetletdynamicsin
cardiomyopathypatients[28–30].
Twocommonapproachesexistforencodingdiffusionin-vivo:amotioncompensatedspin-echowiththe
encodinglastingforonlyafewtensofmilliseconds[24,26];andamonopolarstimulated(STEAM)approach
wherediffusionisencodedoveranentirecardiaccycle[31].Thespin-echoapproachencodesdiffusionin
4
asingleheartbeat,althoughthediffusionencodingtimeisaroundtwoordersofmagnitudeshorterwhen
comparedwiththeSTEAMapproach,whichresultsinprobingsmallermicrostructural lengthscalesand
concomitantlyhighermeasureddiffusivities[32–34].
ASTEAMapproachtakesfulladvantageofcardiacmotionbeingperiodic,withthediffusionencoding/de-
codinggradientsappliedwhentheheartisatthesamepositioninsuccessivecardiaccycles.Thistechnique
requiresbreath-holdingordedicatedrespiratorytriggering.However,awell-knowndisadvantageofthe
STEAMapproachisregularlydebated:cardiacstrainsensitivity.Eventhoughtheheartreturnstoitsorigi-
nal position,measuring diffusionwith a STEAM sequence requires consideration of themyocardium’s
strainhistoryduringthediffusionencodingperiod[35–37].Recentlywehavesuccessfullyacquiredin-vivo
cDTIdataonbothhealthyandcardiomyopathypatientswithamonopolarSTEAMecho-planar imaging
(EPI)sequencewithboth intraand inter-centerreproducibility[28,30,38,39],butquestionsaboutstrain
effectsremainedunanswered,inparticulartheeffectsonthemeasurementofsheetletrotationbetween
systoleanddiastole.
Myocardialstrainthroughoutthecardiaccyclecanbemeasuredandsubsequentlyapplyacorrectionto
thediffusion tensordata acquired in systoleordiastole[29,36].However, the current strain-correction
modelignoressheetletrotationasamajorcontributortowallthickening,andthereforelikelyto“correct”
thediffusiondataforanexaggeratedstraineffect.
Theobjectiveofthisstudywastoevaluatetheeffectofstrain-correctingcDTIdatabydirectlycomparing
in-vivocDTIdatawithandwithoutstrain-correctionaccordingtotheestablishedmodelproposedbyReese
etal.[36],withstrain-freeex-vivoDTIdataofthesameporcineheartsarrestedinadiastolic-likeorsystolic-
likeconformations.
5
METHODSTheporcinecDTIdataanalyzedherearepartofalargerdatasetpreviouslyusedtocomparedin-vivo,in-
situ,andex-vivoDTIscanswithhistologywhereinnostrainanalysisorcorrectionwasapplied[30].Here,
wereportdatafromasubsetinwhich3Dstrainwasacquiredsuccessfully.Ashortenedversionofthiswork
hasbeenpresentedatSCMR2017[40].
EXPERIMENTALDESIGNAnimalprocedureswereapprovedbytheNationalHeart,Lung,andBloodInstituteAnimalCareandUse
Committee.SixteenYorkshirepigs(weight30to35kg)werestudied.Anesthesiawasinducedwiththeuse
ofatropine,butorphanol,ketamine,andxylazineandmaintainedwiththeuseofinhaledsevofluraneand
supplementalIVpropofol,togetherwithmechanicalventilation.
Allimagingwasperformedat3T(MAGNETOMSkyra,Siemens,Germany)usinginvestigationalprototype
sequences,aphasedarraySiemenstorsocoilandspinearraycombination.Themeancardiaccyclelength
was737±82ms.
Followinginitialscans,whichincludedbreath-holdretro-gatedcines,thepigswerescannedwithamulti-
slice cine-DENSE sequencewith3Ddisplacementencoding[41] covering theentire left-ventricle (LV) in
ordertoobtain3Dstraintensorsthroughouttheentirecardiaccycle.ThiswasfollowedbycDTIacquisitions
inonemid-ventricularsliceatmultipletimepointsthroughoutthecardiaccycle includingdiastasisand
peaksystoleasdeterminedfromashort-axiscineinthesameplane.
Alltheanimalsweresubsequentlyeuthanizedunderanesthesiaineithera“diastoliclike”statewithan
intravenousadministrationofKCl(4meq/Kg)(N=8),ora“systoliclike”statebyintravenousadministration
ofBaCl2(40meq/Kg)(N=8).Theheartswereexcisedandrinsedinnormalsaline.Preparationfollowed
theprotocolsdescribedbyKungetal.[18].Thecoronaryarterieswereflushed,andthecardiacchambers
filledwithvinylpolysiloxane(MicrosonicInc.USA).Theheartswereplacedinacylindricalcontainerfilled
withasusceptibilitymatchingfluid(FomblinY-LVAC6-06,SolvaySolexis,USA)andweresupportedusing
open-cellfoam.cDTIwasthenrepeatedinthestaticexcisedheartsforamid-ventricularsliceusingthe
samesequenceasused in-vivo.This slicewasvisuallypositionedtomatch the in-vivosliceasmuchas
possible.
6
STRAINIMAGINGAfreebreathingnavigator-gated2Dmulti-slicespiralcineDENSEimagingwith3Ddisplacementencoding
wasperformedwithcontiguousslicescoveringtheentireLV.Sequenceparametersincluded:inplaneFOV
320mmx320mmwithavoxelsizeof2.5x2.5mm2,slice-thickness8mm,flip-angle10degrees,timeof
repetition14ms,echotime1.16ms,temporalresolution32ms,displacementencodingfrequencyat0.10
cycles/mm,6spiralinterleaves.
WhileDENSEdatawasacquiredtocovertheentireLVonly5sliceswereusedfor3Dstraincalculation:two
slicesbothaboveandbelowthechosenmid-ventricularsliceof interest, theremainingslicesweredis-
carded.TheDENSEdatapost-processingmatrixwastherefore128x128x5pixels.
3D strain and right stretch tensorswere calculated from the DENSE data for themid-ventricular slice
throughoutthecardiaccyclewithMATLAB(Mathworks,Massachusetts,USA)basedsoftwaredeveloped
attheUniversityofVirginia[41,42].Theinitialmyocardialsegmentationwasperformedsemi-automati-
callyusingtheopen-sourcesoftwareDENSEanalysis[43,44].
Thestraindatawereusedfortwopurposes: firstly, to identifythetwosweet-spottimes inthecardiac
cycle(Figure1A);andsecondly,toderiveandregisterstretchtensorstothediffusiondatatobecorrected
(Figure1B)(seeSupportingInformation).
Thecomponentsofstrainalongthelocalcoordinatesofthediffusioneigensystemwascalculatedatpeak
systole.Torsionwasalsocalculatedbetweenthemostbasalandthemostapicalsliceaccordingtoequation
2fromRüsseletal.[45].
DIFFUSIONTENSORIMAGINGBreath-holdin-vivoDTIdatawasacquiredinonemid-ventricularsliceat6to9timepointsinthecardiac
cycle.ASTEAM-EPIsequencewithamonopolardiffusionencodingschemewasusedwiththefollowingb-
values:bref=50s/mm2(20averages)inasingledirection;b=500s/mm2(20averages)in6diffusionen-
codingdirections;b=150 s/mm2andb=350 s/mm2 in6directions (2averages). Zonalexcitation; fat
saturation;diffusionweightingtime1RRinterval;echotimeTE=30ms;BW=1838Hz/pixel;SENSEparallel
imagingaccelerationfactorof2;FOV=320mmx120mm;matrix=160x60pixels;EPIechotrainlength
=30readouts;echospacing=0.65ms;EPIreadoutduration=19.5ms;spatialresolution=2x2x8mm3
7
interpolatedto1x1x8mm3byzerofillingk-space.Ex-vivoDTIscanswereperformedwithanidenticalpro-
tocol,witharepetitiontimesetto1400ms(diffusiontime700ms).
DTIdatawereanalyzedfromonlyfourcardiacphases:diastasis,peaksystole,andthetwoacquisitions
closesttothesweet-spottimesgivenbythecircumferentialstraincurve.
TheDTI datawere post-processedwith software developed in-house (MATLAB). Themyocardiumwas
manuallysegmentedexcludingtherightventricle,bloodpoolandpapillarymuscle.Threeparametersre-
latedtothediffusiontensororientationwereextracted:twoanglesconcerningtheprimaryeigenvector:
helix-angle(HA),andtransverseangle(TA);andananglerelatedtothesecondaryeigenvectorE2A[18,28]
(Figure2).IncontrasttoHA,TAandE2Ademonstratenoobvioustransmuralorganization,thereforefor
simplicity,themedianabsoluteangleisrepresentedwithoutthepolarityoftheangle.However,forcom-
pletenesstheanglepolarityisincludedintheDTIparametermapsandhistograms.Themeandiffusivity
(MD)ofthediffusiontensorwasalsocalculated.
Thediastolicandsystolicdiffusion tensordatawerecalculatedwithandwithoutstrain-correction,and
comparedtothesweet-spotsandtotheex-vivodiffusiondataofthesamemid-ventricularslice.
WilcoxonranksumandKruskal-Wallis (with follow-onpairwisecomparisons) testswereusedtoassess
differences.AP-valueequalorlowerthan0.01wasusedforthefollow-onpairwisecomparisonstoreduce
theprobabilityoftypeIerrorswhenperformingmultiplecomparisons.Inaninitialanalysis,wefailedto
findanyconsistentsignificantdifferencesbetweendifferentsectorsofthewallforDTImeasuresandhave
thereforekeptouranalysisglobaltoreducemeasurementnoise.
8
RESULTSDiffusiontensordataweresuccessfullyacquiredforallpigsand3Dstraindataweresuccessfullyextracted
fromtheDENSEdatafor11oftheoriginal16pigs.TheremainingfivepigshadpoorqualityDENSEimage
dataand/orphase-unwrappingissues.Allsubsequentanalysisisthereforeshownfor11pigsonly(6with
adiastolicarrest,5withasystolicarrest).TheDTIacquisitiontriggertimeclosesttothecalculatedsweet-
spottimeswasdifferentby24±12ms.TwoDTIsweet-spotdatasets(of22)wereexcludedastheclosest
DTItriggertimedifferedbymorethan40ms.Themeanheart-ratewas83(10)bpm,andtheLVejection
fractionwas0.49(0.06)%.
STRAINMEASUREMENTSThemid-ventricularmeanradial,circumferential,andlongitudinalstrain-timecurvesareshowninfigure
3A.Themeasuredmean(standarddeviation)peakradial,circumferentialandlongitudinalstrainswere:
Err=0.21(0.06),Ecc=-0.14(0.02),Ell=-0.11(0.01).Thesweet-spotsgivenbytheEcccurvesarelocatedat
24%(7%)and74%(12%)oftheRRintervaltime.Globaltorsionwas5.3(1.5)degrees(figure3B).Thelargest
componentsofstrainarealignedwiththesheetlets(Esspositivestrain)andwiththemyocytes(Emmneg-
ativestrain).Theshearcomponents(Ems,Emn,andEsn)aresmalleratpeaksystole(figure3C).
DIFFUSIONTENSORDATAThemeasuredDTIparametersareshownintable1.Therearesignificantdifferencesbetweendiastolicand
systolicE2Avaluesforin-vivowithoutandwithstrain-correction(P<0.001andP=0.005respectively)and
forex-vivo(P=0.004).Additionallytherearesignificantdifferencesforthein-vivotransverseanglewithout
strain-correction(P=0.015),andfortheex-vivoHArange(P=0.004:P=0.03).Nosignificantdifferenceswere
foundforanymeasurebetweensweet-spot1andsweet-spot2.Meandiffusivitywasfoundstatistically
differentbetweendiastoleandsystoleforbothwithandwithoutstrain-correction(P<0.01andP=0.001
respectively)butnotforex-vivo(P=0.66).MDisapproximately40%lowerfortheex-vivoheartscompared
tothein-vivo.
Inadditiontomeasuringchangesfromdiastoletosystole,wealsocomparedthein-vivoresultswithand
withoutstrain-correctionwiththeex-vivoarrestedheartsinthesamecorrespondingstate(bottomoftable
9
1).ExamplemapsshowingthethreeDTIorientationmeasuresin-vivo(bothwithandwithoutstrain-cor-
rection)andex-vivoareshowninFigure4fortwohearts,onearrestedinarelaxeddiastolic-likeandone
inacontractedsystolic-likeconformation.VisuallyHAandTAmapsshowasimilarconfigurationbetween
diastoleandsystole,andbetweenwithout/withstrain-correctionandex-vivo.E2Amapsshowpronounced
differencesbetweendiastoleandsystoleforthedatawithoutstrain-correction,withstrain-correction,and
ex-vivo.E2Amapsindiastoleshowlowpositiveandnegativeangleswhileinsystoletheyhavehighpositive
andnegativeangles.InbothdiastoleandsystolethereisasignificantdifferenceinE2Abetweenwithand
withoutstrain-correction(P=0.003andP=0.01respectively).Ex-vivoE2Asdiffersignificantlyfromthere-
spective systolic strain-correctedE2A (P=0.006)but less so from thediastolicnon-strain-correctedE2A
(P=0.037).Additionalangulardifferenceswerefoundbetweenthestrain-correcteddataandtheex-vivo
dataforthesystolicendocardialHAmeanvalues(P<0.001)andbetweenthediastolicTAwithandwithout
strain-correction(P=0.007).
TheabsolutemedianE2AvaluesfromTable1arealsodisplayedinascatterplot(Figure5).Themobilityof
the individual myocardial median E2A values between diastole and systole for the in-vivo data with-
out/withstrain-correction,betweenthetwosweet-spottimes,andtherespectivearrestedex-vivovalues
areshown.AlargerotationofE2isseeninthenon-strain-correcteddatafromamorewallparallelconfor-
mationindiastole(lowangles),toamorewall-perpendicularconformationinsystole(highangles).Asim-
ilareffectisseeninthestrain-correcteddata,althoughtherotationissmaller.Thebiggestdifferencebe-
tweenin-vivodatawithandwithoutstrain-correctionisinthesystolicphasewherethestrain-corrected
E2Aissignificantlylowerthanthenon-strain-correctedandtheex-vivocontractedsystolic-likearrests[me-
dianabsoluteE2A=34.3ovsE2A=57.1o(P=0.01),E2A=60.5o(P=0.006)respectively].AsinTable1,E2Ais
similarbetweenthetwosweet-spots.
Figure6andFigure7provideadditionalanalysisofHA,TAandE2A.Figure6plotsthemeantransmuralHA
lineprofiles,andhistogramsofthemyocardialdistributionofHAvalues.Thereisalineartransmuralpro-
gressionofHAvaluesinallcases,andaslightlywiderdistributionofHAvaluesintheex-vivodata.Noclear
transmuralorganizationwasfoundforTAandE2A.Figure7showsthedistributionofTAandE2Avalues.
ThereisnoapparentdifferencebetweentheTAwithandwithoutstrain-correctioninanyoftheplots,in
contrasttothelargedifferencesinE2A.AsobservedinFigure5,thesystolicdistributionsofE2Aangles
withoutstrain-correctionappearmoresimilartotheex-vivorelaxedandcontracteddistributionsthanaf-
terstrain-correction.Thein-vivosystolicdistributionappearstobethemostdifferentbetweenwithand
10
withoutstrain-correction;whilethedistributionpeaksat±90degreeswithoutstrain-correction,thepeak
occursatamuch lowerabsoluteangleof25degreeswithstrain-correction.Thecombinedsystolicand
diastolicrootmeansquaredeviationofthein-vivohistogramswithoutandwithstrain-correctiontothe
ex-vivohistogramsis0.027and0.038respectively.Thein-vivohistogramswithoutstrain-correctionare
thereforeclosertotheex-vivo.
11
DISCUSSIONInpreviousworkweshowedthatinhealthyhumanheartsthereisasignificantrotationofE2Aastheheart
contractsfromdiastoletosystole,whileHAremainsrelativelyunchanged;however,theconfoundingcon-
tributionofstraintotheseresultswasunclear[28].Subsequentpre-clinicalstudieshavedemonstrateda
similarpatternofE2Arotationinaporcinemodelimagedin-vivo,andin-situandex-vivointheabsenceof
straineffects.Thesimilarityofthein-vivo,in-situandex-vivoresultsenabledustoinferthatthatpotential
straineffectsmustbesmall,althoughnostrain-correctionwasperformedinourpreviousreports.Inthis
work,weassessedtheperformanceofthestandardstrain-correctionmodel[35]bydirectlycomparingin-
vivoDTIdatabeforeandafterstrain-correctionwithex-vivostrain-freedata.Webelievethistobethefirst
studycomparingin-vivodiffusiondatawithandwithoutstrain-correctionwiththesameheartsafterarrest.
Apreviousstudypresentedstrain-correctedSTEAM-EPIdataacquiredinpeak-systoleanddiastole,demon-
stratingtheeffectsofstrain-correction[29].However,thesestudieslackedaground-truthwhichwehave
includedintheformofex-vivoDTIinthesamehearts.Wewerethereforeabletoquantifytheeffectof
strainonin-vivocDTIandassesstheperformanceofconventionalstrain-correction.
Severalimportantresultswerefoundregardingtensororientation.ChangesinHAandTAthroughoutthe
cardiaccycleweresmallandtheseparameterswere largelyunaffectedbystrain-correction.Ex-vivoHA
andTAvaluesanddistributionweresimilartoin-vivo,althoughtheex-vivoHAdistributionwasflatterthan
thein-vivocounterparts.Itisnoteworthythatourdatashowedverylowtransverseanglesatallcardiac
phasesinterrogated,indicatingahighlycoherentmyocyteorientation.IncontrastE2Avariesastheheart
contracts.Figure5andFigure7(lowerpanel)showthatin-vivoE2Achangedfrompredominantlylowab-
soluteanglesindiastoletopredominantlyhighabsoluteanglesinsystolewithoutstrain-correction.The
samedatawithstrain-correctionshowedamuchsmallershiftbetweensystoleanddiastole,withthelarg-
estdifferencesobservedinsystolewhenpre-andpost-strain-correctiondatawerecompared.Asexpected,
thedistributionofE2Avaluesatthetwosweet-spotswassimilar,becausetheheartwasinasimilarcon-
figuration.Whencomparingwithoutandwithstrain-correctiondatawiththecorrespondingex-vivore-
sults,wherenotissuestrainhistoryispresent,thewithoutstrain-correctiondatamorecloselyresembles
theex-vivodistributionsofE2A.Theseresultsareconsistentwithasubstantial rotationofE2between
diastoleandsystole, inkeepingwithcurrentmodelsofcardiaccontractionduetorotationofsheetlets
12
withshearlayerslippage.Thecurrentstrain-correctionmodelartefactuallyreducesmeasuredE2Amobil-
ity.Astheheartspendsalongerproportionofthecardiaccycleinarelaxedratherthancontractedstate,
theeffectofstrain-correctionismostmarkedondataacquiredatpeaksystole.
ThechangeindistributionofE2Aanglesfromdiastoletosystolewasmuchmorepronouncedthanthat
measuredinaLangendorffperfusedratheartmodelfromHalesetal.[14].Moreworkisneededtoascer-
tain ifthisdisparity isduetoprotocoldifferences, inparticulartheshorterdiffusionmixingtimeofthe
spin-echo sequenceusedbyHalesetal.,oranatomicalandphysiologicaldifferencesbetween the two
species.
Thein-vivoresultspresentedhereareconsistentwiththoseofStoecketal.inhumanhearts[29]andalso
withnumericalsimulations(seesupportingfigureS1andsupportingtableS1).Strain-correctingthediffu-
siondatasignificantlyreducedE2Amobility,andbroughtsystolicE2Avaluesclosetothesweet-spotvalues
wherestraineffectsareminimal.TheobservationthatE2Amobilityisreducedbystrain-correctionisnot
unexpected.MacroscopictissuestrainmeasuredwithMRIisintimatelyconnectedtocellularrearrange-
mentandsheetletshear,andthereforedirectlycorrelatedwithE2AmobilityasmeasuredbycDTI.Asthese
measuresareinextricablyinterconnected,correctingformacroscopicstraininevitablyleadstoalteredE2A
mobility.
Theeffectsofstrain-correctioncanbecategorizedaccordingtothreecardiacorthogonalcoordinates.The
diffusion tensororientationmeasuresareonlyaffectedby strain thathascomponentsalong thesame
directionsastherespectiveeigenvectors.Forexample,theprimaryeigenvectorofdiffusion,whichdefines
HA,iscommonlyalignedclosetothelongitudinal-circumferentialplane,withonlysmallcomponentsinthe
radialdirection.HAisthereforeminimallyaffectedbyradialstrain,commonlythedirectionwiththegreat-
estmeasuredstrain.Additionally,itwasfoundthattheeffectsoflongitudinalandcircumferentialstrain,
whicharetypicallyofsimilaramplitude,haveopposingeffectsonHA(seesupportingtableS2).Thiswork
thereforesuggeststhattheglobaleffectofstrainandstrain-correctiononHAissmall.Bycontrast,E2is
mainlyaffectedbythecumulativeeffectofradialandlongitudinalstrains,resultinginE2beingaffected
morebystrain-correction.
Thestrain-correctionmodelestablishedbyReeseetal.[35]assumesthehearttobeformedofagelatin
likematerialwhichisstretchedandcompressedatdifferentstagesofthecardiaccycle.Ifwearemeasuring
13
diffusionatpeaksystole,thenwhilethediffusionisbeingencodedmyocardialtissueinitiallyshortensra-
diallybeforecontractingtoitsoriginalposition.Thediffusionencodingintheradialdirectionistherefore
enhancedbythehistoryofmaterialstrain(Figure8).Converselyifmeasuringdiffusioninadiastolicstate,
thenthematerialwillbestretchedintheradialdirectionbeforecomingbacktoitsoriginalposition,re-
sultinginareductionofthemeasureddiffusionintheradialdirection.Analogousstraineffectswillhappen
inthelongitudinalandcircumferentialdirection.Thisisthebasisofthepreviouslyproposedstrain-correc-
tionmodel:fromthe3Dstretchtensorhistorythroughoutthecardiaccycle,onecancorrectthemeasured
3D diffusion tensor to cancel these strain effects. However, this model assumes a uniform diffusivity
throughoutthecardiaccycle,withnobarriersonthescaleswemeasurediffusionon;additionally,itas-
sumesthatmacroscopiccardiacdeformationisaresultofelasticmaterialstretchonly.Itisnowgenerally
recognizedthatthemajorityofwallthickening,andlongitudinalshortening,isnotduetocardiomyocyte
stretchbutduetocellularreorganizationandtransmuralshearmoderatedbyitslaminarorganization.For
thisreason,itisthoughtthatthepreviouslyproposedmodelmaysignificantlyoverestimatetheeffectsof
microscopiccardiomyocytestrainbasedonthemeasuredmacroscopictissuedeformation.
TheworkbySonnenblicketal.measuredareductionof15%inthelengthofthecardiomyocytesarcomere
inthecanineheart[13],whichresultsinapproximately8%increaseinradiusforincompressiblecardiomy-
ocytes.Thein-vivomedianwallthickeninginthisstudywasapproximately28%.Assumingwehavesimilar
cardiomyocytesarcomerelengthandradiuschangesintheporcineheart,thenapproximately30%ofthe
macroscopicwallthickeningcouldbeduetomicroscopiccardiomyocytestrain.Microstructuralstrainand
reorganizationofthetissuewillthereforeimpactonthemeasuredtensor,butthemodelrequiredtode-
scribethisishighlycomplexandyettobeelucidated.Inthispaper,wemainlyanalyzetensororientation,
butthereisanimportantrotationalinvariantmeasurethatweshoulddiscuss:meandiffusivity.MDpro-
videsameasureofthemeansquaredistancediffusedbywatermoleculesinagiventime.WiththeSTEAM
sequence,diffusionisencodedoveranentireheartcycle.Diffusivitymeasurementswilldependonmyo-
cyteshape,strainandperfusioneffects;allthreevarythroughoutthecardiaccycle.Withoutstrain-correc-
tion,MDdecreasesbetweendiastoleandsystole(1.11to0.99x10-3mm2s-1,p=0.01),andtheopposite
happenswithstrain-correction(1.00to1.14x10-3mm2s-1,p=0.001).AsimilarresultwasfoundbyStoeck
etal. inthehumanheart[29].Themeasuredex-vivodiffusivitiesareapproximately40%lower,whichis
expected when diffusion measurements change from body temperature to room temperature. Im-
portantly,thereisnosignificantchangeinMDex-vivofromdiastoletosystole(P=0.66),wherenostrainor
perfusioneffectsarepresent.
14
WehavesimplifiedtheTAandE2Aanalysisbycalculatingthemedianvaluefromtheabsolutedistribution,
ignoringanglepolarity.WebelievethistobeavalidsimplificationbecausethesignedTAandE2Adistribu-
tionsfromFigure7appearsymmetric.ThemedianvalueofabsoluteE2Aisthereforeameasureofsheetlet
tiltrelativetothelocalheartwall.
Inthiswork,weassumethattheex-vivoarrestedrelaxedandcontractedconformationsapproximateto
diastoleandsystolein-vivo.Duetothelackofbloodpressureloading,thearrestedcontractedheartwall
thicknesswasfoundtobe140%ofthein-vivosystolicwallthickness.Histologyoftheseex-vivopighearts
afterfixationwasperformedforanotherstudyandsheetletandshearlayerorientationswerefoundto
correlatetodiffusionE2orientationmeasuredwithDTIinbothex-vivoandin-vivodata[30].Wetherefore
believethattheex-vivocontractedsystolic-likestateapproximatesthein-vivoconformationofthemicro-
structure.
Therearetwoimportantlimitationstothiswork:thelownumberofanimalsstudied,andtheavailable
spatial-resolutionofcurrentin-vivoMRmeasurementsofstrainanddiffusion.Diffusiontensorestimation
fromavoxelcontainingalargenumberofcardiomyocytesandmyolaminaecanonlybemeaningfulifthere
iscoherenceofthemicrostructure.Histologystudieshaveobservedtwopopulationsofcountersloping
sheetletsandshearlayers[18];moreadvanceddiffusiontechniqueswillberequiredtointerrogatethese
multiplepopulations[46].Theaccuracyofradialandlongitudinalstrainisalsoaffectedbythelowspatial
resolutionused inDENSE imaging.Porcinestrainvaluesmeasured inthisworkwerefoundtobe lower
than typical valuesmeasured in humans, therefore the effects of strain-correction are expected to be
greaterinhumans.DENSEcineimagingusesprospectiveECGgating,andthereforethelateststagesofthe
cardiaccyclemaynotbeimaged.Someoftheindividualstraincurvesexhibitalatesystolicpeak(more
than60%oftheRR-interval)whichisnotphysiologicalandlikelyduetounderestimationoftheR-Rinterval
anderrorsintheDENSEanalysis.Microvascularperfusionwill influenceourdiffusionmeasurementsin-
vivo,inparticularmeandiffusivity[47],althoughitdoesnotseemtosubstantiallyimpactprincipaleigen-
vectororientationsignificantly[48].
15
CONCLUSIONSThecurrentlyacceptedmodelofstrain-correctionfordiffusiondataobtainedwithaSTEAM-EPIsequence
doesnotsignificantlyaltermeasuresrelatingtotheprimaryeigenvector(HAandTA).However,ithasa
considerable impacton theorientationof thesecondaryeigenvector (E2A). In-vivoE2Avalueswithout
strain-correctionapproximatemorecloselytothestrain-freeex-vivovaluesinarelaxeddiastolic-likeand
contractedsystolic-likestates,thanafterstrain-correction.Amorecompletemodelofdynamicmyocardial
microstructureisneededtoenableaccuratestrain-correctiontakingintoaccountcellularrearrangement
andsheetletshear.
16
FIGURECAPTIONS
Figure1–A:diagramshowinghowthesweet-spottimesarecalculated.Thesweet-spottimesaredefinedwhenthemeasuredstrainequalsitsmeanvalue.ThemeanLVcircumferentialstrain-timecurvewasused,asthehighernumberofpixelsalongthecircumferentialdirectionarelikelytomakeitmorerobustthanthe radialor longitudinal strain curves.B: flowchart showinghow thestrain-correction isapplied to thediffusiontensors.Dobsisthediffusiontensorbeforestrain-correction;U(t)isthestretchtensorthroughoutanentirecardiaccycleD;Disthediffusiontensorafterstrain-correction.
Figure2–DiagramshowingthedefinitionsofHA,TAandE2A.TocalculateHA,E1isprojectedinthelocallongitudinal-circumferential plane shown in blue (E1proj). The anglebetweenE1 and E1proj defines TA.The angleof E1proj to the circumferential axis definesHA.HA is positive for a right-handedhelixwhenviewedfromtheapex.TAispositiveifitrotatesclockwisefromthelongitudinal-circumferentialplaneandnegativeifitrotatescounterclockwise,whenlookingalongthecircumferentialaxisdirection.Theprojec-tionof E1 is thenused todefine theplaneperpendicular to it, knownas the cross-cardiomyocyteplaneshowninred.Thisplanecontainstwoorthogonaldirections:theradialdirectionandthewalltangentdi-rection.ThesecondaryeigenvectorE2 isprojected intothisplaneandtheanglebetweenE2projandthewall-tangentdirectiondefinesE2A.E2A ispositive ifpointing towards theapexandnegative ifpointingtowardsthebase.
Figure3– (A)Radial (Err),circumferential (Ecc),and longitudinal (Ell) straincurvesnormalizedtotheRRintervalfortheanterior,lateral,inferiorandseptumsectors.Thethinlinesrepresentthesubjects’meanstrain curves averagedover the LV sliceof interest, and the thick lines represent the intersubjectmeanstraincurve.(B)globaltorsioncurves.(C)Straintensorcomponentsalongthediffusiontensorcoordinatesystematpeaksystole(mm–alongmyocytes,ss–alongsheetlets,nn–alongsheetlet-normal;shearcom-ponents:ms–myocytesheetlet,mn–myocytesheetlet-normal,sn–sheetletsheetlet-normal).
Figure 4 – ExampleDTImapsof tensororientationmeasurements:HA, E2A, and TA for twohearts, onearrestedindiastole(left)andonearrestedinsystole(right).
Figure5–MedianabsoluteE2Avalues.Thein-vivovaluesshowthemobilityofE2Afromadiastolictoasystolic phasewith andwithout strain-correction and at the two-time points closest to the sweet-spottimes.Therespectivecolor-codedex-vivoheartsarealsoshownforeitheradiastolicorsystolicarrest.
Figure6–Top:meantransmuralHAlineprofileswithintersubjectinterquartilerangeattheendocardium;mesocardium; epicardium. Line profiles are shown for diastole and systole with and without strain-correction,atthetwosweet-spottimes,andforthearrestedex-vivohearts.TheR2valuesaretheadjusted
17
R2 for a linear fit. Bottom:HA histograms of allmyocardial voxels at diastole (with andwithout strain-correction),systole(withandwithoutstrain-correction),thetwosweet-spots,andfortheexplantedheartsarrested in a diastolic/systolic like state. Thehistogram curves show the intersubjectmedian and inter-quartilerange;binsizeis10degrees.
Figure 7 – TA (Top) and E2A (Bottom)histogramsof allmyocardial voxels at diastole (with andwithoutstrain-correction),systole(withandwithoutstrain-correction),thetwosweet-spots,andfortheexplantedheartsarrested inadiastolic/systolic likestate.Thehistogramcurvesshowthe intersubjectmedianandinterquartilerange;binsizeis10degrees.
Figure8–Simulationoftheeffectsofanelasticstretchonafreediffusioncloudofparticles.Astretch>1loop of the material during diffusion encoding will result in a reduction of the diffusivity along thatdirection.Astretch<1loopwillhavetheoppositeeffect.
SupportingFigureS1–Simulateddiffusiontensors,andtherespectiveHA,E2AandTAmapsandhistogramswithandwithoutstrain-correctionforadiastolicandsystolictensorconformation.
18
TABLETITLESANDCAPTIONS
Table1–Diffusionparameters
Top:Diffusionparametersin-vivowithandwithoutstrain-correction,atthein-vivosweet-spotsandattheex-vivoarrestedhearts.HArange isthemeananglerangefromendocardiumtoepicardium.TAandE2AarethemedianoftheabsolutevaluesofallLVmyocardialvoxels.MDisthemeanoverallLVmyocardialvoxels.Allvaluesareshownas intersubjectmedian[interquartilerange].* (red)-denotesastatisticallysignificantdifferencewhencomparedtothecorrespondingsystolicorsweet-spot2value(P<0.05).
Bottom:Pvalueswhencomparingbetweenin-vivowithoutandwithstrain-correction(sc)andex-vivoar-rested. HA range is themean angle range from endocardium to epicardium. *(red) - denotes statisticaldifference (P<=0.01; P-value threshold with a Bonferroni correction for 3multiple tests, conservativelyroundeddownto0.01).
Supporting tableS1–Numerical simulations:effectsof strain-correctiononDTIparameters.HArange isthemeananglerangefromendocardiumtoepicardium.TAandE2Aarethemedianoftheabsolutevaluesofallmyocardialvoxels.
SupportingtableS2–SummaryoftheeffectsofeachstraindirectiontoHArangeandE2A.TAisnotcon-siderablyaffectedbyanystraindirection.↗=increase;↘=decrease;−=noconsiderableeffect.
19
ACKNOWLEDGMENTSThisworkwassupportedbytheBritishHeartFoundation;theNationalInstituteofHealthResearchCardi-
ovascularBiomedicalResearchUnitattheRoyalBromptonHospitalandImperialCollege,London;andthe
NationalHeart,Lung,andBloodInstitute,NationalInstitutesofHealth,DivisionofIntramuralResearch,
DepartmentofHealthandHumanServices(HL004607-14CPB).
DISCLOSURESProfessorDudleyPennell receives researchsupport fromSiemens,and isastockholderanddirectorof
CardiovascularImagingSolutions.ProfessorDavidFirminreceivesresearchsupportfromSiemens.
20
BIBLIOGRAPHY1.PettigrewJB.OntheArrangementoftheMuscularFibresintheVentriclesoftheVertebrateHeart,with
PhysiologicalRemarks.PhilosophicalTransactionsoftheRoyalSocietyofLondon.1864;154:445–500.
2. Streeter JrDD, SpotnitzHM,PatelDP,Ross Jr J, SonnenblickEH. Fiberorientation in the canine left
ventricleduringdiastoleandsystole.CircRes.1969;24(3):339–347.
3.SmerupM,NielsenE,AggerP,FrandsenJ,Vestergaard-PoulsenP,AndersenJ,NyengaardJ,Pedersen
M,RinggaardS,HjortdalV,etal.Thethree-dimensionalarrangementofthemyocytesaggregatedtogether
within the mammalian ventricular myocardium. Anat Rec (Hoboken). 2009;292(1):1–11.
DOI:10.1002/ar.20798
4.HarringtonKB,RodriguezF,ChengA,LangerF,AshikagaH,DaughtersGT,CriscioneJC,IngelsNB,Miller
DC. Directmeasurement of transmural laminar architecture in the anterolateralwall of the ovine left
ventricle: new implications for wall thickening mechanics. Am J Physiol Heart Circ Physiol.
2005;288(3):H1324-30.DOI:10.1152/ajpheart.00813.2004
5.SandsGB,GernekeDA,HooksDA,GreenCR,SmaillBH,LegriceIJ.Automatedimagingofextendedtissue
volumesusingconfocalmicroscopy.MicroscResTech.2005;67(5):227–239.DOI:10.1002/jemt.20200
6.GilbertSH,BenoistD,BensonAP,WhiteE,TannerSF,HoldenAV,DobrzynskiH,BernusO,Radjenovic
A. Visualization and quantification of whole rat heart laminar structure using high-spatial resolution
contrast-enhanced MRI. Am J Physiol Heart Circ Physiol. 2012;302(1):H287-98.
DOI:10.1152/ajpheart.00824.2011
7.LeGriceIJ,SmaillBH,ChaiLZ,EdgarSG,GavinJB,HunterPJ.Laminarstructureoftheheart:ventricular
myocyte arrangement and connective tissue architecture in the dog. American Journal of Physiology -
HeartandCirculatoryPhysiology.1995;269(2):H571LP-H582.
8. LeGrice IJ, Takayama Y, Covell JW. Transverse Shear Along Myocardial Cleavage Planes Provides a
MechanismforNormalSystolicWallThickening.CirculationResearch.1995;77(1):182–193.
9. Takayama Y, Costa KD, Covell JW. Contribution of laminarmyofiber architecture to load-dependent
changes in mechanics of LV myocardium. Am J Physiol Heart Circ Physiol. 2002;282(4):H1510-20.
DOI:10.1152/ajpheart.00261.2001
21
10.CostaKD,TakayamaY,McCullochAD,Covell JW. Laminar fiberarchitectureand three-dimensional
systolicmechanicsincanineventricularmyocardium.AmJPhysiol.1999;276(2Pt2):H595-607.
11.SpotnitzHM,SpotnitzWD,CottrellTS,SpiroD,SonnenblickEH.Cellularbasisforvolumerelatedwall
thicknesschangesintheratleftventricle.JMolCellCardiol.1974;6(4):317–331.
12. Axel L, Wedeen VJ, Ennis DB. Probing dynamic myocardial microstructure with cardiac magnetic
resonancediffusiontensorimaging.JCardiovascMagnReson.2014;16:89.DOI:10.1186/s12968-014-0089-
6
13.SonnenblickEH,RossJ,CovellJW,SpotnitzHM,SpiroD.TheUltrastructureoftheHeartinSystoleand
Diastole. Changes In Sarcomere Length. Circulation Research. 1967;21(4):423–431.
DOI:10.1161/01.RES.21.4.423
14.HalesPW,SchneiderJE,BurtonRAB,WrightBJ,BollensdorffC,KohlP.Histo-anatomicalstructureof
thelivingisolatedratheartintwocontractionstatesassessedbydiffusiontensorMRI.ProgBiophysMol
Biol.2012;110(2–3):319–330.DOI:10.1016/j.pbiomolbio.2012.07.014
15. Holmes AA, Scollan DF, Winslow RL. Direct histological validation of diffusion tensor MRI in
formaldehyde-fixedmyocardium.MagnResonMed.2000;44(1):157–161.
16. Scollan DF, Holmes A, Winslow R, Forder J. Histological validation of myocardial microstructure
obtainedfromdiffusiontensormagneticresonanceimaging.AmJPhysiol.1998;275(6Pt2):H2308-18.
17.ScollanDF,HolmesA,ZhangJ,WinslowRL.Reconstructionofcardiacventriculargeometryandfiber
orientationusingmagneticresonanceimaging.AnnBiomedEng.2000;28(8):934–944.
18. KungGL,Nguyen TC, ItohA, Skare S, Ingels JrNB,Miller DC, Ennis DB. The presence of two local
myocardialsheetpopulationsconfirmedbydiffusiontensorMRIandhistologicalvalidation.JMagnReson
Imaging.2011;34(5):1080–1091.DOI:10.1002/jmri.22725
19. Helm PA, Tseng H-J, Younes L,McVeigh ER,Winslow RL. Ex vivo 3D diffusion tensor imaging and
quantification of cardiac laminar structure. Magn Reson Med. 2005;54(4):850–859.
DOI:10.1002/mrm.20622
20.HelmP,BegMF,MillerMI,WinslowRL.Measuringandmappingcardiacfiberandlaminararchitecture
22
usingdiffusiontensorMRimaging.AnnNYAcadSci.2005;1047:296–307.DOI:10.1196/annals.1341.026
21. Teh I, Burton RAB, McClymont D, Capel RA, Aston D, Kohl P, Schneider JE. Mapping cardiac
microstructureofrabbitheartindifferentmechanicalstatesbyhighresolutiondiffusiontensorimaging.
ProgBiophysMolBiol.2016Jun.DOI:10.1016/j.pbiomolbio.2016.06.001
22. Nielles-Vallespin S,Mekkaoui C, Gatehouse P, Reese TG, Keegan J, Ferreira PF, Collins S, Speier P,
FeiweierT,deSilvaR,etal.InvivodiffusiontensorMRIofthehumanheart:reproducibilityofbreath-hold
andnavigator-basedapproaches.MagnResonMed.2013;70(2):454–465.DOI:10.1002/mrm.24488
23.NguyenC, Fan Z, Sharif B,He Y,DharmakumarR, BermanDS, LiD. In vivo three-dimensional high
resolutioncardiacdiffusion-weightedMRI:Amotioncompensateddiffusion-preparedbalancedsteady-
statefreeprecessionapproach.MagnResonMed.2013Nov.DOI:10.1002/mrm.25038
24.StoeckCT,vonDeusterC,GenetM,AtkinsonD,KozerkeS.Second-ordermotion-compensatedspin
echodiffusiontensorimagingofthehumanheart.MagnResonMed.2015May.DOI:10.1002/mrm.25784
25.ScottAD,FerreiraPFADCPFADC,Nielles-VallespinS,GatehouseP,McgillL-AL-A,KilnerP,PennellDJDJ,
FirminDNDN.Optimaldiffusionweightingforinvivocardiacdiffusiontensorimaging.MagnResonMed.
2015;74(2):420–430.DOI:10.1002/mrm.25418
26. Aliotta E,Wu HH, Ennis DB. Convex optimized diffusion encoding (CODE) gradient waveforms for
minimum echo time and bulk motion-compensated diffusion-weighted MRI. Magnetic resonance in
medicine.2017;77(2):717–729.DOI:10.1002/mrm.26166
27.DouJ,ReeseTG,TsengW-YI,WedeenVJ.CardiacdiffusionMRIwithoutmotioneffects.MagnReson
Med.2002;48(1):105–114.DOI:10.1002/mrm.10188
28.FerreiraPF,KilnerPJ,McGillL-A,Nielles-VallespinS,ScottAD,HoSY,McCarthyKP,HabaMM,Ismail
TF, Gatehouse PD, et al. In vivo cardiovascular magnetic resonance diffusion tensor imaging shows
evidence of abnormalmyocardial laminar orientations andmobility in hypertrophic cardiomyopathy. J
CardiovascMagnReson.2014;16:87.DOI:10.1186/s12968-014-0087-8
29. Stoeck CT, Kalinowska A, von Deuster C, Harmer J, Chan RW, NiemannM,Manka R, Atkinson D,
SosnovikDE,MekkaouiC,etal.Dual-phasecardiacdiffusiontensorimagingwithstraincorrection.PLoS
One.2014;9(9):e107159.DOI:10.1371/journal.pone.0107159
23
30.Nielles-VallespinS,KhaliqueZ,FerreiraPF,deSilvaR,ScottAD,KilnerP,McGill L-A,GiannakidisA,
GatehousePD,EnnisD,et al.AssessmentofMyocardialMicrostructuralDynamicsby InVivoDiffusion
TensorCardiacMagneticResonance.JournaloftheAmericanCollegeofCardiology.2017;69(6):661–676.
DOI:10.1016/j.jacc.2016.11.051
31.EdelmanRR,GaaJ,WedeenVJ,LohE,HareJM,PrasadP,LiW.Invivomeasurementofwaterdiffusion
inthehumanheart.MagnResonMed.1994;32(3):423–428.
32.vonDeusterC,StoeckCT,GenetM,AtkinsonD,KozerkeS.Spinechoversusstimulatedechodiffusion
tensor imaging of the in vivo human heart. Magnetic Resonance in Medicine. 2016;76(3):862–872.
DOI:10.1002/mrm.25998
33.ScottAD,Nielles-VallespinS,FerreiraP,KhaliqueZ,McGillL-A,KilnerPJ,PennellDJ,FirminD.In-vivo
cardiacDTI:AninitialcomparisonofM012compensatedspin-echoandSTEAM.JournalofCardiovascular
MagneticResonance.2016;18(Suppl1):W19.DOI:10.1186/1532-429X-18-S1-W19
34.NguyenC,SpeierP,BiXiaoming,LiD.ComparisonBetweenSpinEchoandStimulatedEchoDiffusion
EncodingforDiffusion-WeightedCardiacMagneticResonance(DW-CMR)at3T. In:Proc. Intl.Soc.Mag.
Reson.Med.2016.
35. Reese TG,Weisskoff RM, SmithRN,RosenBR,DinsmoreRE,WedeenVJ. Imagingmyocardial fiber
architectureinvivowithmagneticresonance.MagnResonMed.1995;34(6):786–791.
36.Reese,Wedeen,Weisskoff.MeasuringDiffusioninthePresenceofMaterialStrain.JMagnResonB.
1996;112(3):253–258.
37.TsengWY,ReeseTG,WeisskoffRM,WedeenVJ.CardiacdiffusiontensorMRI invivowithoutstrain
correction.MagnResonMed.1999;42(2):393–403.
38.McGillL-A,IsmailTF,Nielles-VallespinS,FerreiraP,ScottAD,RoughtonM,KilnerPJ,HoSY,McCarthy
KP,GatehousePD,etal.Reproducibilityofin-vivodiffusiontensorcardiovascularmagneticresonancein
hypertrophiccardiomyopathy.JCardiovascMagnReson.2012;14(1):86.DOI:10.1186/1532-429X-14-86
39.TunnicliffeEM,ScottAD,FerreiraP,ArigaR,McGillL-A,Nielles-VallespinS,NeubauerS,PennellDJ,
RobsonMD,FirminDN.Intercentrereproducibilityofcardiacapparentdiffusioncoefficientandfractional
anisotropyinhealthyvolunteers.JCardiovascMagnReson.2014;16(1):31.DOI:10.1186/1532-429X-16-31
24
40. Ferreira PF, Nielles-Vallespin S, de Silva R, Scott AD, Ennis D, Auger DA, Suever JD, Zhong X,
SpottiswoodeBS,PennellDJ,etal.Studyontheimpactofstraincorrectiononthesecondaryeigenvector
ofdiffusionwithinvivoandexvivoporcinehearts.In:ProceedingsofSocietyforCardiovascularMagnetic
Resonance.2017.
41.ZhongX,SpottiswoodeBS,MeyerCH,KramerCM,EpsteinFH.Imagingthree-dimensionalmyocardial
mechanicsusingnavigator-gatedvolumetricspiralcineDENSEMRI.MagnResonMed.2010;64(4):1089–
1097.DOI:10.1002/mrm.22503
42. Auger DA, Zhong X, Epstein FH, Meintjes EM, Spottiswoode BS. Semi-automated left ventricular
segmentation based on a guide point model approach for 3D cine DENSE cardiovascular magnetic
resonance.JCardiovascMagnReson.2014;16:8.DOI:10.1186/1532-429X-16-8
43. Spottiswoode BS, Zhong X, Hess AT, Kramer CM, Meintjes EM, Mayosi BM, Epstein FH. Tracking
myocardialmotionfromcineDENSEimagesusingspatiotemporalphaseunwrappingandtemporalfitting.
IEEETransMedImaging.2007;26(1):15–30.DOI:10.1109/TMI.2006.884215
44.GilliamAD,SueverJD,contributors.DENSEanalysis.
45.RüsselIK,TecelãoSR,KuijerJPA,HeethaarRM,MarcusJT.Comparisonof2Dand3Dcalculationofleft
ventricular torsionas circumferential-longitudinal shearangleusing cardiovascularmagnetic resonance
tagging.JCardiovascMagnReson.2009;11:8.DOI:10.1186/1532-429X-11-8
46. SosnovikDE,WangR,DaiG,WangT,AikawaE,NovikovM,RosenzweigA,GilbertRJ,WedeenVJ.
DiffusionspectrumMRItractographyrevealsthepresenceofacomplexnetworkofresidualmyofibersin
infarcted myocardium. Circ Cardiovasc Imaging. 2009;2(3):206–212.
DOI:10.1161/CIRCIMAGING.108.815050
47.DelattreBMA,ViallonM,WeiH,ZhuYM,FeiweierT,PaiVM,WenH,CroisilleP.Invivocardiacdiffusion-
weightedmagneticresonanceimaging:quantificationofnormalperfusionanddiffusioncoefficientswith
intravoxel incoherent motion imaging. Invest Radiol. 2012;47(11):662–670.
DOI:10.1097/RLI.0b013e31826ef901
48.ScottAD,FerreiraPFADC,Nielles-VallespinS,GatehouseP,McGillL-A,KilnerP,PennellDJ,FirminDN.
Optimaldiffusionweightingforinvivocardiacdiffusiontensorimaging.MagneticResonanceinMedicine.
25
2015;74(2).DOI:10.1002/mrm.25418
withoutstrain-correction withstrain-correction sweet-spots exvivoarrests
diastole systole diastole systole sweet-spot1 sweet-spot2 diastole systole
HArange
(deg)
[36.9[6.9]:-
52.4[9.3]]
(P=0.14:P=0.58)
[40.3[4.7]:-
53.3[11.8]]
-
[35.7[6.7]:-
55.7[16.0]]
(P=0.79:P=0.39)
[37.2[3.9]:-
47.5[15.1]]
-
[37.8[4.0]:-
60.6[9.0]]
(P=0.34:P=0.92)
[31.4[10.9]:-
53.7[13.3]]
-
[35.0*[5.0]:-
51.8*[2.1]]
(P=0.004:P=0.03)
[46.5[3.7]:-
60.1[5.2]]
-
absolute
E2A(deg)
12.2*[2.6]
(P<0.001)
57.1[14.2]
-
19.3*[8.64]
(P=0.005)
34.3[11.4]
-
22.9[19.2]
(P=0.91)
24.7[10.0]
-
18.2*[5.42]
(P=0.004)
60.5[5.05]
-
absolute
TA(deg)
6.08*[1.46]
(P=0.015)
7.45[2.94]
-
8.3[2.38]
(P=0.89)
8.67[1.85]
-
6.42[2.17]
(P=0.18)
7.25[1.29]
-
5.92[0.56]
(P=0.13)
7.62[2.34]
-
MD(10-3
mm2s-1)
1.11*[0.07]
(P=0.010)
0.99[0.12]
-
1.00*[0.08]
(P=0.001)
1.14[0.07]
-
1.05[0.12]
(P=0.31)
1.10[0.19]
-
0.61[0.09]
(P=0.66)
0.66[0.11]
-
Betweengroupscomparison(Kruskal-Walliswithfollow-onpairwisecomparisons)
withoutscvswithsc withoutscvsex-vivo withscvsex-vivo
diastole
HArange(deg) P=0.98:P=0.72 P=0.59:P=0.99 P=0.48:P=0.77
absoluteE2A(deg) P=0.003 P=0.037 P=0.94
absoluteTA(deg) P=0.007 P=0.99 P=0.044
systole
HArange(deg) P=0.26:P=0.72 P=0.047:P=0.57 P<0.001:P=0.23
absoluteE2A(deg) P=0.01 P=0.74 P=0.006
absoluteTA(deg) P=0.91 P=0.76 P=0.56
Table1–Diffusionparameters
Top:Diffusionparametersin-vivowithandwithoutstrain-correction,atthein-vivosweet-spotsandattheexvivoarrestedhearts.HArange is themeananglerangefromendocardiumtoepicardium.TAandE2Aarethemedianoftheabsolutevalues of all LVmyocardial voxels.MD is themeanover all LVmyocardial voxels. All values are shownas intersubjectmedian[interquartilerange].*(red)-denotesastatisticallysignificantdifferencewhencomparedtothecorrespondingsystolicorsweet-spot2value(P<0.05).
Bottom:Pvalueswhencomparingbetweenin-vivowithoutandwithstrain-correction(sc)andex-vivoarrested.HArangeis the mean angle range from endocardium to epicardium. *(red) - denotes statistical difference (P<=0.01; P-valuethresholdwithaBonferronicorrectionfor3multipletests,conservativelyroundeddownto0.01).
SUPPORTINGINFORMATION
STRAINCORRECTIONTherightstretchtensorU(t)throughoutthecardiaccyclewascalculatedfromthestraintensordata
E(t):
! " = 2& " + ((1)
whereIistheidentitymatrix.
Justlikethediffusiontensors,thestretchtensorsarepositive-definite.Inordertoregisterthestretch
tensorstothediffusiontensorsandmaintainthepositive-definitenessalog-Euclideanmetric[1]was
usedwithanon-rigiddemonregistration[2].Thisregistrationstepwasperformedforthediffusion
datameasured(Dobs)atdiastasisandpeaksystole.
Toperformstrain-correction,thestretchtensorsareintegratedoveranentirecardiaccycleDwith
theinitialintegrationtimet=0atthesamecardiacphaseasthediffusiondatatobecorrected:
,obs =1Δ !(")12,!(")12d"
∆
5(2)
Dobs is the diffusion tensor without strain-correction and D is the diffusion tensor with strain-
correction.TheequationaboveneedstobesolvedforDasdescribedbyReeseetal.[3].
NUMERICALSIMULATIONSStretchtensorsanddiffusiontensorsweresimulatedbyusingthemedianeigenvaluesandthemedian
HAandE2Afromallthemeasuredin-vivodata.Ahollowdiscwasusedtosimulatethemyocardium
oftheLVinamid-slice.Thetensorshapeisgivenbytheeigenvalues,andthetensororientationby
thetransmuralHAdistributionwithahomogenousE2A.Tensordirectionalnoisewasintroducedby
3D rotating the diffusion tensors with a distribution of pseudorandom angles between [-10, 10]
degrees.Thisintroducedsometensororientationuncertaintywithoutchangingthetensorshape.
Theaveragestraincurvesfromfigure3wereusedtosimulatestretchtensorsthroughoutthecardiac
cycle with eigenvectors aligned with the local cardiac coordinates (radial, circumferential,
longitudinal).Theeffectsofstrain-correctingthediffusiontensorswerethenanalyzedforanaverage
diastolicandsystolicconformation.
Thenumericallysimulateddiffusiontensorsareshown insupportingfigureS1.Thevaluesandthe
effects of strain-correction agreewell with the animal data. E2A is themost affected parameter,
especiallyinsystole.ThiseffectcanbeseenasarotationofthediffusiontensorsaroundE1,withE2
towardsamorewallparallelorientationwithstrain-correction(frompanelCtopanelD).HAandTA
remain very similar throughout. The effects of the simulated strain-correction are summarized in
supportingtableS1.
SupportingtableS2showstheeffectsofeachstraindirectionindividuallytoHAandE2A.TAisnot
considerablyaffectedbyanystraindirection.
Supporting Figure S1 - Simulated diffusion tensors, and the respective HA, E2A and TA maps and
histogramswithandwithoutstrain-correctionforadiastolicandsystolictensorconformation.
diastole systole
input output input output
HArange(deg) [-48.631.9] ~ [-49.732.8] [-4934.2] ~ [-48.533.4]
absoluteE2A(deg) 12.2 ↗ 20.0 56.6 ↘ 40.7
absoluteTA(deg) 5.4 ↗ 7.3 5.6 ↘ 4.2
SupportingtableS1-Numericalsimulations:effectsofstrain-correctiononDTIparameters.HArange
isthemeananglerangefromendocardiumtoepicardium.TAandE2Aarethemedianoftheabsolute
valuesofallmyocardialvoxels.
diastole systole
Err Ecc Ell Err Ecc Ell
HArange(deg) − ↗ ↘ − ↘ ↗
absoluteE2A(deg) ↗ − ↗ ↘ − ↘
SupportingtableS2–SummaryoftheeffectsofeachstraindirectiontoHArangeandE2A.TAisnot
considerablyaffectedbyanystraindirection.↗=increase;↘=decrease;−=noconsiderableeffect.
BIBLIOGRAPHY1. Arsigny V, Fillard P, Pennec X, AyacheN. Log-Euclideanmetrics for fast and simple calculus on
diffusiontensors.MagneticResonanceinMedicine.2006;56(2):411–421.DOI:10.1002/mrm.20965
2.KroonD-JJ,SlumpCH.MRImodalitiytransformationindemonregistration.In:Proceedingsofthe
SixthIEEEInternationalConferenceonSymposiumonBiomedicalImaging:FromNanotoMacro.Vol.
2.Piscataway,NJ,USA:IEEEPress;2009.p.963–966.(ISBI’09).
3.Reese,Wedeen,Weisskoff.MeasuringDiffusioninthePresenceofMaterialStrain.JMagnReson
B.1996;112(3):253–258.