University of Groningen
On the ability to exploit signal fluctuations in pseudocontinuous arterial spin labeling for
inferring the major flow territories from a traditional perfusion scan
van Harten, T. W.; Dzyubachyk, O.; Bokkers, R. P. H.; Wermer, M. J. H.; van Osch, M. J. P.
Published in:
Neuroimage
DOI:
10.1016/j.neuroimage.2021.117813
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from
it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date:
2021
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
van Harten, T. W., Dzyubachyk, O., Bokkers, R. P. H., Wermer, M. J. H., & van Osch, M. J. P. (2021). On
the ability to exploit signal fluctuations in pseudocontinuous arterial spin labeling for inferring the major flow
territories from a traditional perfusion scan. Neuroimage, 230, 1-8. [117813].
https://doi.org/10.1016/j.neuroimage.2021.117813
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
ContentslistsavailableatScienceDirect
NeuroImage
journalhomepage:www.elsevier.com/locate/neuroimage
On
the
ability
to
exploit
signal
fluctuations
in
pseudocontinuous
arterial
spin
labeling
for
inferring
the
major
flow
territories
from
a
traditional
perfusion
scan
T.W.
van
Harten
a,∗,
O.
Dzyubachyk
b,
R.P.H.
Bokkers
c,
M.J.H.
Wermer
d,
M.J.P.
van
Osch
a a C.J. Gorter Center for High Field MRI, Department of Radiology, Leiden University Medical Center, Postbus 9600, 2300 RC Leiden, the Netherlands b Division of Image Processing (LKEB), Department of Radiology, Leiden University Medical Center, Postbus 9600, 2300 RC Leiden, the Netherlandsc Department of Radiology, Medical Imaging Center, University Medical Center Groningen, University of Groningen, Postbus 30.001, 3700 RB Groningen, the Netherlands d Department of Neurology, Leiden University Medical Center, Postbus 9600, 2300 RC Leiden, the Netherlands
a
r
t
i
c
l
e
i
n
f
o
Keywords:
Arterial spin labeling Flow territories Postprocessing Off-resonance effects
a
b
s
t
r
a
c
t
Inarterialspinlabeling(ASL)amagneticlabelisappliedtotheflowingbloodinfeedingarteriesallowing depic-tionofcerebralperfusionmaps.Thelabelingefficiencydepends,however,onbloodvelocityandlocalfield inho-mogeneitiesandis,therefore,notconstantovertime.Inthiswork,weinvestigatetheabilityofstatisticalmethods usedinfunctionalconnectivityresearchtoinferflowterritoryinformationfromtraditionalpseudo-continuous ASL(pCASL)scansbyexploitingartery-specificsignalfluctuations.Byapplyinganadditionalgradientduring labelingtheminimumamountofsignalfluctuationthatallowsdiscriminationofthemainflowterritoriesis de-termined.Thefollowingthreeapproachesweretestedfortheirperformanceoninferringthelargevesselflow territoriesofthebrain:agenerallinearmodel(GLM),anindependentcomponentanalysis(ICA)andt-stochastic neighborembedding.Furthermore,toinvestigatetheeffectoflargevesselpathology,standardASLscansofthree patientswithaunilateralstenosis(>70%)ofoneoftheinternalcarotidarterieswereretrospectivelyanalyzed usingICAandt-SNE.Ourresultssuggestthattheamountofnatural-occurringvariationinlabelingefficiencyis insufficienttodeterminelargevesselflowterritories.Whenapplyingadditionalvessel-encodedgradientsthese methodsareabletodistinguishflowterritoriesfromoneanother,butthiswouldresultinapproximately8.5% lowerperfusionsignalandthusalsoareductioninSNRofthesamemagnitude.
1. Introduction
pCASLisa readilyavailablenon-invasive techniqueforperfusion imagingthatcanbeusedinaclinicalsetting(Alsopetal.,2015),for clinicalresearch,aswellasforneuroscienceapplications,suchas iden-tificationofneuronalnetworks(Daietal., 2016).Byapplyingvessel encoding,pCASLcanalsobeusedtoinferflowterritoriesofarteries (Geversetal.,2012;Wong,2007; WongandGuo,2012).Whereas cur-rentclinicalpracticereliesongeneralatlasesoftypicalflowterritories, informationonthelayoutoftheactualflowterritoriesinanindividual patientcanbehighlyvaluableinseveralclinicalsettings,suchasforrisk assessmentandguidanceofrevascularizationtherapyinpatients suffer-ingfromstroke,aswellasforexplainingdifferencesinpatientoutcome (Hartkampetal.,2016;Hendrikseetal.,2009;vanLaaretal.,2008).
SmallchangesinASLsignalovertimeallowforidentificationof rest-ingstatenetworks(Daietal.,2016).Onecouldhypothesizethatsmall, arteryspecificsignalfluctuationsinpCASL-signalwouldallowtoalso
∗Correspondingauthor.
E-mailaddress:T.W.van_Harten@lumc.nl(T.W.vanHarten).
infertheflowterritoriesfromatraditionalperfusionscan.Suchsignal changescouldbotharisefromphysiologicalfluctuationswithinaflow territory,aswellasfromtheacquisitionprocess.Forexample,the label-ingefficiencyofpCASLisknowntobedependentonvelocityofarterial bloodandoff-resonanceeffectsatthelabeling location(Aslanetal., 2010; Wuetal., 2011, 2007).Anyvariationin labelingefficiencyof aparticulararterywillaffecttheASL-signalofthecompleteassociated flowterritory,and,differencesbetweenarteriescouldthereforeprovide asourceofflow-territory-specificsignalfluctuationsthatcouldbe ex-ploitedforidentifyingtheseterritories.Similarly,dynamicchangesin arterialtransporttimescouldalsoinduceflow-territory-specific fluctua-tions.(Qiuetal.,2010;VerbreeandvanOsch,2018)Thiscouldbeaway of acquiringflowterritory informationforfree froma2D-multi-slice pCASLscan,whichnormallyhasmany(>25)repeatedmeasurements. When naturaloccurring fluctuations wouldbe sufficient, this would enableapenalty-freeapproachforobtainingcombinedcerebralblood flow(CBF)andflowterritoryinformationfromasinglescan. Alterna-tively,smalllabeling-efficiencyfluctuationscouldbeinsertedintothe sequence,whichwouldonlyresultinaminorSNRpenaltyonthe perfu-sionscan.Toassessthefeasibilityofsuchanapproach,weanalyzedASL scansbyseveraltechniquestraditionallyusedinfunctionalconnectivity
https://doi.org/10.1016/j.neuroimage.2021.117813
Received27July2020;Receivedinrevisedform14January2021;Accepted20January2021 Availableonline29January2021
T.W. van Harten, O. Dzyubachyk, R.P.H. Bokkers et al. NeuroImage 230 (2021) 117813
research.Moreover,byincorporatingdifferentlevelsofsignal fluctua-tionsintopCASLscans,itwasdeterminedwhatlevelofsignal varia-tionsisneededtoallowassessmentofflowterritoryinformation. Vessel-encodedsignalfluctuationswereintroducedbydeliberatelychanging thelabelingefficiencyoftheleftandrightinternalcarotidarteryby addingavessel-encodinggradient(Geversetal.,2012; Wong,2007) withvariablestrengthintoanormalpCASLsequence.Thethree meth-odstestedarealreadycommonlyusedinfunctionalMRIanalyses:1) generallinearmodel(GLM)basedapproach,asgenerallyusedintask basedfMRI(Yerysetal.,2018);TheGLMbasedapproachwasselectedto representthebest-casescenario:byincludingthetimingoftheinserted signalfluctuationsintotheanalysisanupper-limitofperformanceof flow-territoryextractioncanbeestablished.2)Independentcomponent analysis(ICA),asgenerallyusedinrestingstatefMRI(Daietal.,2016); and3) t-stochasticneighbor embedding(t-SNE), aclusteringmethod withbroadapplicationsinimageprocessing(Huetal.,2020).ICAand t-SNEareinputnaïveapproachesastheydonotrelyontheknowledge ofthetimingofsignalfluctuationsandcould,therefore,alsobeused fortraditionalpCASLdatawhennoadditionalsignalfluctuationsare inserted. TheICAwasperformedbothontheoriginaldata,aswellas onthedataafterblurring witha10 mmkernel. Theadditionofthe 10mmblurredversionwasbasedonthehypothesisthattherelatively lowSNRofsingleaverageASLimageswouldhamperidentificationof flowterritoryspecificfluctuations.Sincesuchfluctuationswouldper definitionbespatiallycoherent,someblurringcouldhelpinidentifying flowterritories,albeitatalossofeffectivespatialresolution.
Finally,toinvestigatethisapproachinpatientcases,we retrospec-tivelyanalyzedpCASLscansofpatientswithaninternalcarotid steno-sisandcomparedtheoutcomewiththeflowterritoriesobtainedfroma separatelyacquiredvessel-encodedpCASLscan.Inthisproof-of-concept studyonlytheflowterritoriesoftherightandleftinternalcarotidartery wereconsidered.
2. Methods
pCASLscanswithdifferentlevelsofvessel-encodedsignal fluctua-tionswereacquiredin5volunteerswithtraditionalflowterritoryscans obtainedas groundtruth.Thescansweresubsequentlyanalyzed us-ingthreedifferent statisticalmethods:agenerallinearmodel(GLM), anindependent componentanalysis (ICA), anda t-Stochastic neigh-borembedding(t-SNE)approach.TheFMRIB’sSoftwareLibrary(FSL,
www.fmrib.ox.ac.uk/fsl) wasused fortheGLM andICA approaches, whilethet-SNEwasperformedusinganinhousedevelopedpipeline,as furtherdescribedin Applied Models .ThepCASLscansacquiredwithout avessel-encodinggradientwereanalyzedonlyusingtheICAandt-SNE methods,sincetheGLM-approachreliesonpriorknowledgeonthe tim-ingofthefluctuationsinlabelingefficiency,whichisnotavailablewhen thereisnopriorknowledgeonthepatternofsignalfluctuations.
2.1. Participants
This research has been performed in accordance with the dec-laration of Helsinki. Written informed consent was obtained from all participants. Five healthy participants (mean age 30.4 ± 7.6, male/female=4/1) were recruited for this study. Since one might expect that patientsshow larger naturaloccurring fluctuationsthan healthyparticipants,datafromthreepatients(meanage:69.7±9.5, male/female=2/1) with a recently symptomatic unilateral internal carotidstenosis(> 50%occluded)wereretrospectivelyanalyzed.
2.2. Scans
Allhealthyparticipantswerescannedona3TMRIscanner(Achieva, PhilipsMedicalSystems,theNetherlands)usingaprotocolcontaining
a standardbackground suppressedpCASL(TR/TE 4290/16ms, FOV 240×240×119mm3matrixsize80×80,17sliceswithnointerslicegap
resultinginavoxel-sizeof3×3×7mm3,labelingduration=1800ms,
post-labelingdelay(PLD)of1800mswith35repetitionsresultingina totalscandurationof5min09s),atime-of-flightangiogramofthe label-ingplane(TR/TE35/3.5ms,flipangle60°,9sliceswithaslicethickness of3mm,FOV230×230×19mm3,matrixsize192×96,resultingina
pixel-sizeof1.2×2.4mm2andascandurationof31s),avessel-encoded
flowterritorymappingscan,consideredasthegoldstandard,with simi-larsettingsasthestandardpCASLscan,butacquiredin5conditions:1) label,2)control,3)vessel-encodingintheRLdirectionwithadistanceof 50mmbetweenfulllabelandcontrol,4)and5)twovessel-encodings intheAPdirection withadistanceof15mmwiththelastcondition shifted7.5mmcomparedtothefirst(Geversetal.,2012)(12averages, resultinginascandurationof4min39s)Thevesselencodingdistance isdefinedasthedistancebetweenthelocationwhereoptimallabeling occurs(alwaysplannedtobeontopofaninternalcarotidartery)and thelocationwherethecontrolconditionisachieved,anequivalent def-initionisthedistanceoverwhichapiphasewilloccurduetothevessel encodinggradient.E.g.avesselencodingdistanceof50mm(average dis-tancebetweeninternalcarotidarteries)wouldhavethetargetedinternal carotidarteryinlabelconditionwhilethecontralateralarterywouldwe incontrolcondition.TheprotocolalsoincludedarangeofpCASLscans withanadditionalvessel-selective gradientintheleft-rightdirection withvaryinggradientstrengthsaccordingtoapredefinedscheme(see
Fig.1),whosescanparameterswereotherwiseidenticaltothestandard pCASLscan.Thevaryinggradientstrengthvalueswerechosentocreate artery-specificsignalfluctuationsbyhavingmaximallabelingefficiency atthelocationofoneinternalcarotidartery,whilethecontralateral in-ternalcarotidarterywillexperiencesub-optimallabeling.Thephases ofthepCASLlabelingpulseswereadaptedtoensurethattheoptimal labelingconditionwasachievedontopofoneoftheinternalcarotid arteries.Bychangingtheareaunderthevessel-encodinggradientby adjustingitsamplitude,thespatialfrequencyoftheinducedlabel ef-ficiency changesin thedirection of thevessel-encoding gradientcan becontrolled.Withoptimallabelinginoneinternalcarotidartery,the contralateralinternalcarotidarterywillexperienceoff-resonanceeffects andthusalowerlabelingefficiency.Themagnitudeoftheoff-resonance effectswilldeterminetheextentofthelabelingefficiencydecrease.The term“vessel-encodingdistance” willbeusedforthedistanceresulting in 𝜋 phaseshiftduringtheinterpulseintervalof thepCASLtrain.At thisdistancefromtheinternalcarotidarterywithoptimallabeling,the controlandlabelconditionwillbeswitchedcomparedtothetargeted vessel.
Thepositionofoptimallabelingisswitchedtothecontralateral in-ternalcarotidarteryafteracquisitionofeachlabelcontrolpair, result-inginaswitchinsignalintensityoftheflowterritoriesafterevery la-bel/controlpair.Thiswasdoneinordertocreatetemporalfluctuations inlabelingefficiencyforbothinternalcarotidarteries,whileprovidinga normalperfusionmapwhenaveragingalldata,albeitwithlowermean labelingefficiency.
The data from patients with a unilateral internal carotid artery stenosis was acquired as part of a previously published study. (Hartkampetal.,2011)Allpatientshadsufferedatransientischemic attack (TIA) or non-disabling ischemic stroke ipsilateral to the in-ternal carotidarterystenosis. TheprotocolincludedstandardpCASL (TR/TE 4000/14 ms, FOV 240×140 mm2, matrix size 89×79, 17
sliceswithnointerslicegapresultinginavoxel-sizeof3×3×7mm3,
labeling duration=1650 ms, post-labeling delay (PLD) of 1525 ms with 38 repetitions resulting in a total scan duration of 5 min 26s),andaplanning-freeflowterritorymappingscanwiththesame settings.
Fig.1. Theeffectoftheareaofthevessel-encodinggradientontheASL-signalinanexamplesubject;withamplitudeofthevessel-encodinggradientvaryingfrom strong(top)toweak(bottom).ThefirstcolumnshowsTime-Of-Flightangiographyofthelabelingplanewiththebaraboveindicatinghowthevessel-encoding gradientaffectsthelabelingefficiency:whiteisoptimallabelingandblackthelocationwhereanadditional𝜋 phaseshiftoccursduringthepulse-interval,whichwill resultintooppositebehaviorduringlabeling(i.e.controlduringthelabel,andlabelduringthecontrolcondition).Thenumbersreflectthevessel-encodingdistance. Sinceoptimallabelingisswitchedbetweenthelefttotherightinternalcarotidarteryaftereverylabel-controlpair,thedataweresplitintotwo.Columnstwoand threeshowtheresultingmeanASLsubtractionimagewhentheleftorrightinternalcarotidarterywaslabeledoptimally.Thefinalcolumnshowsthemeansignal intensityoverthecompletepCASLscans,i.e.theaverageofthesecondandthirdcolumn,showingahomogenousimage,exceptforthefirstrowthatshowssignal cancelations.Thepercentagesontherightindicatetherelativesignalfluctuationsfortheimagesshowninthisfigure,ascalculatedbyEq.(1).Thebottomrow showsthedesignforGLMsetup.
T.W. van Harten, O. Dzyubachyk, R.P.H. Bokkers et al. NeuroImage 230 (2021) 117813 2.3. Post-processing
First,theflowterritoriesof bothinternalcarotid arteriesandthe basilararteryweredeterminedbyprocessingvessel-encodedflow terri-torymappingscans,usingpreviouslydescribedmethods(Geversetal., 2012).The effect of the applied vessel-encoding gradient on the la-belingefficiency wasquantified for eachflow territory as themean signalintensityaftersubtracting andaveragingtheimageswith sub-optimalconditionsduetothepresenceofthevessel-encodinggradient (SIoff-resonance)fromtheimageswiththearteryinperfectlabeling
condi-tion(SIfull);andsubsequentlynormalizingbythemeansignalintensity
asmeasuredbythestandardpCASLscan(SIpCASL): Relativesignalfluctuations=(SIf ull−SIof f-resonance
)
∕SIpCASL∗100% (1)
2.3. Applied models
Fourdifferentapproacheswereselectedtoextracttheflowterritories baseduponflow-territory-specificsignalfluctuations.
DatapreprocessingfortheGLMandICA basedmethodswas per-formedusingFSL(FMRIB’sSoftwareLibrary, www.fmrib.ox.ac.uk/fsl). The following preprocessing was applied: motion correction using MCFLIRT;non-brainremovalusingBET(Smith,2002);grand-mean in-tensitynormalizationoftheentire4Ddatasetbyasinglemultiplicative factor; high-pass temporal filtering (Gaussian-weighted least-squares straightlinefitting,withsigma=50.0sforbothGLMandICA).
2.3.1. General linear model
IntheGLMtheappliedlabelingfluctuationswereintroducedinto themodelasanexplanatoryvariable.Becausethemultipleregression of theGLM requiressuch priorinformation, itis not possible todo this analysis on scanswhere no intentionalsignal fluctuationswere insertedbymeansofadditionalgradientsswitchedoninthelabeling plane.ThegenerallinearmodelwascarriedoutusingFEAT(FMRI Ex-pertAnalysisTool)Version6.00,partofFSL(FMRIB’sSoftwareLibrary,
www.fmrib.ox.ac.uk/fsl).Thetime-seriesstatisticalanalysiswascarried outusingFILMwithlocalautocorrelationcorrection.Onlylowerlevel analysiswasappliedasthiswasranonasingle4-DMRIexperiment.
T -statisticmapsresultingfromthisanalysiswereusedwithout thresh-oldingorclustering.
2.3.2. Independent component analysis
ICAanalysiswascarriedoutusingProbabilisticIndependent Compo-nentAnalysisimplementedinMELODIC(MultivariateExploratory Lin-earDecompositionintoIndependentComponents)Version3.15,part ofFSL.Insummary,pre-processeddatawerewhitenedandprojected intoamulti-dimensionalsubspaceusingaprobabilisticPrincipal Com-ponentAnalysiswherethenumberofdimensionswasestimatedusing theLaplaceapproximationtotheBayesianevidenceofthemodel or-der(Beckmannetal.,2001; Minka,2000).Thewhitenedobservations weresubsequentlydecomposed intosetsof vectorsthatdescribe sig-nalvariationacrossthetemporaldomain(time-courses)andacrossthe spatialdomain(maps)by optimizingfor non-Gaussianspatialsource distributionsusingafixed-pointiterationtechnique(Hyvarinen,1999). Estimatedcomponentmapsweredividedbythestandarddeviationof theresidualnoiseandthresholdedbyfittingamixturemodeltothe his-togramofintensityvalues(BeckmannandSmith,2004).Fromthe iden-tifiedcomponents,thebestfittingwasselectedbaseduponthelargest differencebetweenleftandrightintracranialregiondefinedbyaleft andrighthemispheremaskcontainingintracranialvoxelsseparatedby a(mid-)sagittalplanecrossingthroughthegenuandsplenumofthe corpuscallosum.Thecomponentwiththehighestcorroborationwith thecorrespondinghemispheremaskwasusedasoutcomeforthatflow territoryforfurtherevaluation.
Finally,thecompleteMELODICwasrepeatedafterblurringtheinput datawitha10mmGaussiankerneltominimizetheeffectofvoxel-level inconsistencies.
2.3.3. t-Distributed stochastic neighbor embedding
Sinceweareonlyinterested indifferentiatingleft andright flow-territories, whichis effectivelya 1Dproblem, we setthetarget em-beddingdimensionalitytounity.TheASLdifferencevolumeswere re-ducedinthetimedimensiontoasinglevaluebyapplyingthe Hierarchi-calStochasticNeighborEmbedding(HSNE)algorithm(Pezzottietal., 2016).HSNEisamodificationoftheclassicalt-SNEalgorithm,which wasproventobetterpreservethedatastructureduringtheembedding process(vanderMaatenandHinton,2008).Fullembeddingofeach ofthedatasetswasperformedasahierarchicalprocesswith4levels. Drillingdownintothenextlevelofthehierarchywasperformedafter convergence(definedasthemaximumnumberofiterations)ofthe em-beddingonthecurrentlevel.Allthelandmarksavailableonthecurrent levelwereusedforthis.Initialpositionsofthelower-levellandmarks werecalculatedbyinterpolatingcurrentpositionsofthecorresponding higher-levellandmarks.
2.4. Validation
Therobustnessofthemethodswastestedviareceiveroperating char-acteristic(ROC)curves,onlyincludingintracranialvoxelsthatshow suf-ficientASL-signal.Moreover,sincesignalfluctuationswereonly intro-ducedintheinternalcarotidarteries,voxelsthatarepartofthe pos-teriorcirculationterritoryasdeterminedfromtheflowterritorymaps wereexcluded.SufficientASL-signalwasoperationalized,asthelower limitofvalueswithinthegraymatter,whichwasdeterminedviaa con-sensusvalueoverallparticipantsandwaskeptconstantforallscans. Voxelswithlowersignal,suchasthoseinthewhitematter,were ex-cludedfromtheanalysis,becausetheycouldnotbeclassified onthe flowterritorymappingdataandare,therefore,notacorrectreflection of theperformanceofthedifferentapproaches,andwouldintroduce biasintheevaluationprocedure.
The z -statisticsmapoftheGLM-basedapproach,theIC-mapofthe ICA-based approaches,as wellasthedistance mapsfrom the t-SNE-basedapproachwereprojectedonthegroundtruthflowterritorymaps. ROCcurvesweregeneratedbythresholdingthestatisticalmaps reflect-ingthez-score/distancetoyieldbinarysegmentationsandsubsequently calculatingtheoverlapwiththeflowterritorymap.Thisanalysiswas donewithanin-housedevelopedpipelinecombiningMeVisLab3.2and Matlab2016a.
3. Results
Thevessel-encodinggradientsweintroduceddidleadtotheintended changesinASLsignalfluctuation.Thesefluctuationsweredirectly pro-portionaltothegradientareaoftheemployedvessel-encoding gradi-ents.InFig.1theeffectofvessel-encodinggradient-areaonthe ASL-imagesisshownforonesubject.Fordisplayingthemagnitudeofthe induced signalfluctuations,theASL-scans weresplitintothose mea-surementsforwhichtheleftinternalcarotidarterywasoptimally la-beledandthosethat targetedtherightinternal carotidartery. From theseimageswecanconcludethattheproposedapproachindeedleads toatunablemagnitudeoflabelingefficiencyfluctuationsontopofa normalASL-scan.Thechosenstrength valuesforthevessel-encoding leadtofluctuationsbetween3%and150%.
The averaged effect over all participants of the area under the vessel-encoding gradienton therelative signalfluctuationsis shown in Fig. 2, showcasingthedirectrelationship between theintroduced off-resonanceeffectsinthelabelingplaneandASL-signalfluctuations downstreamwithintheflowterritoriesofthesearteries.Thesmallerthe areaunderthevessel-encodinggradient,thesmallertheoff-resonance effectsinthecontralateralinternalcarotidartery,andthesmallerthe differenceinrelativesignalfluctuationsbetweenthetwoflowterritories ofthesearteries.
Fig.3showshowwellthefourdifferentpost-processingapproaches coulddetecttheunderlyingflowterritoriesforasingleexamplesubject
Fig.2. Meanrelativesignalfluctuationsasafunctionofvessel-encodingdistance;(distanceresultingin𝜋 phaseshiftduringtheinterpulseintervalofthepCASL train)withstandarderrorofthemeanforallparticipants.
Fig.3.Post-processingapproachesandtheirabilitytodetectflowterritories;exampleofhowwellthefourpost-processingapproachescandetecttheflowterritories onthedatafromthesecondrowofFig.1,i.e.avessel-encodingdistanceof200mm,correspondingtoarelativesignalfluctuationof41.3%.Fromtoptobottom: GLM,ICAwith10mmblurring,ICAwithoutblurring,t-SNE,andflowterritorymapping.Intherightpanel,theROCareshownforthesamedata.Thedashedline indicatesahypothetical50/50randomchanceclassifier.
(samesubjectasFig.1).Itcanbeseenthat,asthestrengthofthe vessel-encodinggradientisstrongenough,all3methodscanidentifytheflow territorycorrectly,butspecificityandsensitivityvarybetweenthefour approaches,whichisalsoreflectedintheROCcurvesdisplayedonthe right.
Fig.4summarizesthemeanandstandarddeviationoftheareaunder theROCcurvesoverall5participants,overalllabelingconditions.The averagesoftheareaunderthecurveoftheROCcurvesshowthatthe GLM-basedapproachdoesoutperformtheothermethodsinthisdataset, whichwasexpectedbecausetheGLMexploitspriorknowledgeofthe appliedlabelingfluctuations.Inthescanswiththestrongestsignal fluc-tuationstheICAwasfoundtosplitflowterritoriesinmultiple compo-nents.Onlythecomponentwiththestrongestdifferencebetweenleft andrightwasusedinfurtheranalyses.Furthermore,formostlevelsof relativesignalfluctuations,blurringimprovedtheabilityof the ICA-basedapproachtodifferentiatebetweenflowterritories,althoughthis wasnotthecaseforallparticipantsorforallconditions.Whenthe differ-encesinfluctuationswereespeciallylarge(>200mmgradientlength) orsmall(<500mmgradientlength),therewasanagreementbetween
theblurringandnon-blurringresults,whileforconditionsbetweenthese twoextremes,betterresultswereobtainedwhenblurringtheinputdata. TheICAaswellasthet-SNEanalysisonthepCASLdatawithout imputedsignalfluctuationsdidnotyieldconvincingflowterritory in-formation,ineitherhealthyvolunteersnorthepatientswithinternal carotidarterystenosis(Fig.5).
4. Discussion
Themainfindingsofthisstudyarethreefold.First,weareableto reportthatwiththeintroductionofweakoff-resonanceeffectsbymeans of a vessel-encoding gradientit becomes possible todetermineflow territoriesusingaGLMstatisticalanalysis.Theperformancebecomes weakerinrobustnesswhentheintroducedfluctuationsaresmalleror when“naivemethods” areemployed,i.e.whennopriorinformationon thetimingofthelabelingefficiencyfluctuationsisexploitedinthe sta-tisticalmodel.WeobserveastatisticallysignificanthigherROCforGLM, butthisshouldbeinterpretedwithcareduetotheverysmallsample sizeofourstudySecond,whenmaintainingaminimumof0.85forthe
T.W. van Harten, O. Dzyubachyk, R.P.H. Bokkers et al. NeuroImage 230 (2021) 117813
Fig.4. MeanareaundertheROC-curveforeach methodasafunctionofthevesselencoding dis-tance;error-barsindicatestandarderrorofthe mean.Asmalloffsetinthex-directionwas intro-ducedforeachmethodforillustrationpurposes.
Fig.5. MeanareaundertheROC-curveforeachmethodtestedonstandardpCASLscans;allmethodswereslightlybetterthana50/50chanceclassifierbutunable todetermineflowterritories.Therewasnosignificantdifferencebetweenhealthyvolunteersandpatientswithaunilateralstenosisintheinternalcarotidartery. Errorbarsindicatestandarderrorofthemean.
areaundertheROC-curvetoreflectcorrectperformance,theICAand t-SNEapproacheslosttheirabilitytodifferentiatebetweenleftandright flowterritoryatarelativesignalfluctuationsof41.3%and64.1%, re-spectively.Third,itwasnotfeasibletodeterminereliableflowterritory informationfromtraditionalpCASLscansinhealthyvolunteersor pa-tientswith>70%unilateralcarotidstenosisusingICAort-SNEanalysis,
i.e.naturaloccurringfluctuationsinlabelingefficiencyinthetemporal domainwereinsufficienttoyieldreliableflowterritoryinformation.
Coherentfluctuationswithinaflowterritorywerefoundtobetoo smalltoidentifythemfromconventional pCASLscans,whereas fluc-tuationscausedbyhemodynamicfluctuationswerepreviouslyfoundto allowdetectionofrestingstatenetworks,albeitatgrouplevel(Daietal., 2016). This can be considered a surprisingfinding: coherent
dynamic fluctuations as observed in resting-state fMRI/ASL have a more substantial influence than any flow-territory specific hemody-namicchangesoroff-resonanceeffects.Correctperformanceofour ap-pliedpost-processingmethodologywasprovenbyshowingthatimputed labelingefficiencyfluctuationsdidallowdifferentiationbetweentheleft andrightinternal carotidarteryterritory.One explanationmightbe thatfluctuationsinlabelingefficiencyaremainlycausedbyglobal sys-temicprocesses.Forexample,heartrateorrespiratoryvariationswill affectallbrain-feedingsarteriesterritoriessimilarly,justasfluctuations inend-tidalCO2 willhaveaglobalrather thananarteryspecific
ef-fect.Theimpactofbreathingonthelabelingefficiencycanbeestimated frompreviouslyreportedresonancefrequencychanges(Rajetal.,2001;
Versluisetal.,2012).Forexample, Birnetal.(1998)and Zhaoetal. (2017) showedthatoff-resonancefrequenciesduringswallowingand speechreachup to−9.6Hzat 3Tin theinferiorpartof thebrain, whichwouldresultinaphaseshiftof4.14° duringthe1.2msbetween thepCASLpulses,i.e.aratherlimitedeffectalthoughtheeffectatthe labelingplanecouldbeexpectedtobeabitmorepronounced.
When16.5%orstrongerfluctuationsofASL-signalwereimputedon theperfusionsignal,reliablediscriminationbetweentherightandleft internalcarotidarteryflowterritorycouldbeachievedbyusingtheGLM approach.Thesefluctuationsarelargerthanthefluctuationsdescribed earlierinducedbyrespirationandotherphysiology(Birnetal.,1998;
Zhaoetal.,2017).Thepresenceofsuchagradientimpliesthatthemean labelingefficiencywillbe8.3%loweroverthewholeASL-acquisition (labelingisonlydecreasedinhalfofthemeasurements),which trans-latesdirectlyto8.3%lowerSNRoftheperfusionmapwhencomparedto anASLscanwiththesamescantime.Thisseemstobeontheborderof whatcouldbeconsideredacceptableinaclinicalsettingtoallow com-binedflowterritoryandperfusionimaging.Also,slightinterindividual differencesinlocationofthearteriescouldaffecttheinducedlabeling efficiencyreductionbetweentheleftandrightinternalcarotidartery, whenrelyingonpopulationaveragedsettingsfortheplanningofthe lo-cationofperfectlabelingconditions.Thiswouldimplythattheaverage perfusionmapcouldbecomeasymmetrical,whichmightaffectclinical interpretation,althoughalsoinnormalpCASLscanslabelingefficiency differencesbetweenarterieshavebeenreported(Chenetal.,2018).
Otherapproachesforimprovingvesselencodedscansinvolve apply-ingnovelacquisitiontechniques,suchasadifferentlabeling scheme toreducescantimewhilemaintainingsufficientSNRforreliableflow territorydetermination(Gunther,2006; Zimineetal.,2006);and ap-proachestocalculateoptimallabelingschemestofurtherimproveSNR (Berryetal.,2015).Theseapproaches,however,requirespecific plan-ning approaches and may therefore introduce differences in perfor-mancebetweentechnicians.Furthermore,theseapproacheswouldnot beabletoextractanyadditionalflowterritoryinformationfrom tradi-tionalpCASLscans.
Inthisstudy,fourdifferentpost-processingapproachesweretested intheirabilitytodetermineflowterritoryinformationfromsmall fluc-tuationsinlabelingefficiencyinthefeedingarteries.Basedonobserved trendsindataofthisstudythemostrobustmethodappearstobea GLM-basedapproach.Thisiscausedbythepriorinformationthatisfedinto themodel,therebyincreasingthestatisticalpowerofthemodel.The relativeperformanceofthenaïvemethodsseemsatfirstglancemore difficulttodescribe,sincet-SNEoutperformsbothICA-basedmethods onthebiggestandthesmallestamountofsignalfluctuations,whereas in-betweentheICA-basedapproachesperformbetter.However,forthe strongestsignalfluctuations,theICAwas foundtosplitflow territo-riesinmultiplecomponents,whereasforthevalidationonlyasingle componentwasincluded.Thisexplainsthepoorerperformanceofthe ICA-methodsforthedatawiththelargestsignalfluctuations.
Wehypothesizedthatpatientswithlargevesseldiseasecouldshow moreprominentdifferencesinASL-signalfluctuationsbetweenflow ter-ritoriesthanhealthyvolunteers.However,inthethreepatientsthatwe includedinthisstudy,theICAandt-SNEanalysesdidnotallow trust-worthydeterminationoftheflowterritories.Apparently,alsoincarotid
stenoticdiseasepatients,fluctuationsinASL-signalaretoocorrelated betweenflowterritoriesor,ingeneral,toosmalltoallow differentia-tion.
Alimitationofourexperimentalapproachisthatweonlyfocused ondifferentiationbetweentheleftandrightinternalcarotidartery ter-ritoryandignoredtheposteriorcirculation.This approachwastaken forpracticalreasons:weaimedatprimarilydemonstratingthe possibil-itiesforthemaininternalcarotidarteryterritoriesandwouldexpand oureffortstothethreeregionswhensuccessful.Moreover,thedistance betweenthetwointernalcarotidarteriesislargerthanthedistance be-tweentheanteriorandposteriorcirculation,whichcouldresultinlarger differenceinfluctuationscausedbyoff-resonanceeffects,i.e.indata withoutadditionalvessel-encodinggradients.Tobeabletodiscriminate theposteriorfromtheanteriorcirculation,strongergradientsshouldbe employedthanusedinthecurrentstudies,sincethedistancebetween arteriesissmaller.Discriminationofvesselsdistancedmorethan5mm couldbeachieved,althoughplanningthelocationofoptimallabeling wouldrequireanangiographicscoutandcarefulplanningInthedata withoutadditionalflow-encodinggradients,wealsotriedto discrimi-natetheanteriorfromtheposteriorcirculation,butthisdidnot pro-videsatisfactoryresultsbyeitherICAort-SNEanalysisoftheASL-data. Apparently,thedifferenceinhemodynamicsbetweentheanteriorand posteriorcirculationdoesnottranslatetoflow-territoryspecific fluctu-ationsoftheASL-signaloverdifferentrepeatedmeasurements.
5. Conclusions
SignalfluctuationspresentinstandardpCASLscansdueto fluctua-tionsinoff resonanceeffectsinthelabelingplaneortransittimeare not sufficientforextractingflowterritory mappinginformationfrom standard pCASL scans when using ICA or t-SNE, neitherin healthy participants nor in patients with severe (>70%) unilateral internal carotidarterystenosis.Whenapplyingadditionalvessel-encoded gra-dientsthesemethodsareabletodistinguishflowterritoriesfromone another,butthiswouldresultinapproximately8.5%lowerperfusion signalandthusalsoareductioninSNRofthesamemagnitude.
Acknowledgments
TheauthorswouldliketothankSophieSchmidt,WouterTeeuwisse andAnnemariekevanOpstalfortheirhelpacquiringscandata.
Funding
ThisresearchhasbeenmadepossiblebytheDutchHeart Founda-tionandtheNetherlandsOrganisationforScientificResearch(NWO), aspartoftheirjointstrategicresearchprogramme:"Earlierrecognition ofcardiovasculardiseases”.ThisprojectispartiallyfinancedbythePPP AllowancemadeavailablebyTopSectorLifeSciences&Healthtothe DutchHeartfoundationtostimulatepublic-privatepartnerships.This researchwasalsosupportedbytheEUundertheHorizon2020program (project:CDS-QUAMRI),andtheCAVIAproject(nr.733050202),which hasbeenmadepossiblebyZonMW
Authorshipstatement
T.W.vanHarten– conceptualization,formalanalysis,investigation, writing-originaldraft.
O.Dzyubachyk– methodology,formalanalysis,writing– review& editing.
R.P.H.Bokkers– investigation,writing– review&editing. M.J.H.Wermer– writing– review&editing,supervision.
M.J.P.vanOsch– conceptualization,methodology,writing– review &editing,supervision.
T.W. van Harten, O. Dzyubachyk, R.P.H. Bokkers et al. NeuroImage 230 (2021) 117813 Disclosureofinterests
T.W.vanHarten– nothingtodisclose. O.Dzyubachyk– nothingtodisclose. R.P.H.Bokkers– nothingtodisclose.
M.J.H.Wermer– receivesapersonalZonMwgrantVIDI(91717337) andDekkergrant,theNetherlandsHeartFoundation2016T086.
M.J.P.vanOsch– receivesresearchsupportfromPhilips.
Dataandcodeavailabilitystatement
Thedatausedinthisprojectisconfidential,butmaybe obtained uponrequestaftersigningadatauseagreement.Researchersinterested inaccesstothedatamaycontactthelast author(M.J.P.vanOsch). Priortosharingthedatawillbeanonymizedandscansstrippedofall identifiableinformation,includingfacialfeatures.
References
Alsop, D.C. , Detre, J.A. , Golay, X. , Gunther, M. , Hendrikse, J. , Hernandez-Garcia, L. , Lu, H. , MacIntosh, B.J. , Parkes, L.M. , Smits, M. , van Osch, M.J. , Wang, D.J. , Wong, E.C. , Zaharchuk, G. , 2015. Recommended implementation of arterial spin-labeled perfusion MRI for clinical applications: a consensus of the ISMRM perfusion study group and the European consortium for ASL in dementia. Magn. Reson. Med. 73, 102–116 . Aslan, S. , Xu, F. , Wang, P.L. , Uh, J. , Yezhuvath, U.S. , van Osch, M. , Lu, H. , 2010. Estimation
of labeling efficiency in pseudocontinuous arterial spin labeling. Magn. Reson. Med. 63, 765–771 .
Beckmann, C., Noble, J., Smith, S., 2001. Investigating the Intrinsic Dimensionality of FMRI Data for ICA. pp. S76-S76.
Beckmann, C.F. , Smith, S.M. , 2004. Probabilistic independent component analysis for functional magnetic resonance imaging. IEEE Trans. Med. Imaging 23, 137–152 . Berry, E.S. , Jezzard, P. , Okell, T.W. , 2015. An optimized encoding scheme for plan-
ning vessel-encoded pseudocontinuous arterial spin labeling. Magn. Reson. Med. 74, 1248–1256 .
Birn, R.M. , Bandettini, P.A. , Cox, R.W. , Jesmanowicz, A. , Shaker, R. , 1998. Magnetic field changes in the human brain due to swallowing or speaking. Magn. Reson. Med. 40, 55–60 .
Chen, Z. , Zhao, X. , Zhang, X. , Guo, R. , Teeuwisse, W.M. , Zhang, B. , Koken, P. , Smink, J. , Yuan, C. , van Osch, M.J.P. , 2018. Simultaneous measurement of brain perfusion and labeling efficiency in a single pseudo-continuous arterial spin labeling scan. Magn. Reson. Med. 79, 1922–1930 .
Dai, W. , Varma, G. , Scheidegger, R. , Alsop, D.C. , 2016. Quantifying fluctuations of resting state networks using arterial spin labeling perfusion MRI. J. Cereb. Blood Flow Metab. 36, 463–473 .
Gevers, S. , Bokkers, R.P. , Hendrikse, J. , Majoie, C.B. , Kies, D.A. , Teeuwisse, W.M. , Ned- erveen, A.J. , van Osch, M.J. , 2012. Robustness and reproducibility of flow territo- ries defined by planning-free vessel-encoded pseudocontinuous arterial spin-labeling. AJNR Am. J. Neuroradiol. 33, E21–E25 .
Gunther, M. , 2006. Efficient visualization of vascular territories in the human brain by cycled arterial spin labeling MRI. Magn. Reson. Med. 56, 671–675 .
Hartkamp, N.S. , Bokkers, R.P.H. , van der Worp, H.B. , van Osch, M.J.P. , Kappelle, L.J. , Hendrikse, J. , 2011. Distribution of cerebral blood flow in the caudate nucleus, lentiform nucleus and thalamus in patients with carotid artery stenosis. Eur. Radiol. 21, 875–881 .
Hartkamp, N.S., Hendrikse, J., De Cocker, L.J.L., de Borst, G.J., Kappelle, L.J., Bokkers, R.P.H., 2016. Misinterpretation of Ischaemic Infarct Location in Relationship to the Cerebrovascular Territories. 87, 1084-1090.
Hendrikse, J. , Petersen, E.T. , Cheze, A. , Chng, S.M. , Venketasubramanian, N. , Golay, X. , 2009. Relation between cerebral perfusion territories and location of cerebral infarcts. Stroke 40, 1617–1622 .
Hu, Y. , Li, X. , Wang, L. , Han, B. , Nie, S. , 2020. T-distribution stochastic neighbor embed- ding for fine brain functional parcellation on rs-fMRI. Brain Res. Bull. 162, 199–207 . Hyvarinen, A. , 1999. Fast and robust fixed-point algorithms for independent component
analysis. IEEE Trans. Neural Netw. 10, 626–634 .
Minka, T.P. , 2000. Automatic Choice of Dimensionality for PCA. MIT Media Lab Vision and Modeling Group Technical Report 514 .
Pezzotti, N., Höllt, T., Lelieveldt, B., Eisemann, E., Vilanova, A., 2016. Hierarchical Stochastic Neighbor Embedding. 35, 21-30.
Qiu, M. , Paul Maguire, R. , Arora, J. , Planeta-Wilson, B. , Weinzimmer, D. , Wang, J. , Wang, Y. , Kim, H. , Rajeevan, N. , Huang, Y. , Carson, R.E. , Constable, R.T. , 2010. Ar- terial transit time effects in pulsed arterial spin labeling CBF mapping: insight from a PET and MR study in normal human subjects. Magn. Reson. Med. 63, 374–384 . Raj, D. , Anderson, A.W. , Gore, J.C. , 2001. Respiratory effects in human functional mag-
netic resonance imaging due to bulk susceptibility changes. Phys. Med. Biol. 46, 3331–3340 .
Smith, S.M. , 2002. Fast robust automated brain extraction. Hum. Brain Mapp. 17, 143–155 .
van der Maaten, L. , Hinton, G. , 2008. Visualizing data using t-SNE. J. Mach. Learn. Res. 9, 2579–2605 .
van Laar, P.J. , van der Grond, J. , Hendrikse, J. , 2008. Brain perfusion territory imag- ing: methods and clinical applications of selective arterial spin-labeling MR imaging. Radiology 246, 354–364 .
Verbree, J. , van Osch, M.J.P. , 2018. Influence of the cardiac cycle on pCASL: cardiac triggering of the end-of-labeling. Magma 31, 223–233 .
Versluis, M.J., Sutton, B.P., de Bruin, P.W., Börnert, P., Webb, A.G., van Osch, M.J., 2012. Retrospective Image Correction in the Presence of Nonlinear Temporal Magnetic Field Changes Using Multichannel Navigator Echoes. 68, 1836-1845.
Wong, E.C. , 2007. Vessel-encoded arterial spin-labeling using pseudocontinuous tagging. Magn. Reson. Med. 58, 1086–1091 .
Wong, E.C. , Guo, J. , 2012. Blind detection of vascular sources and territories using ran- dom vessel encoded arterial spin labeling. Magn. Reson. Mater. Phys. Biol. Med. 25, 95–101 .
Wu, W.-C. , Jiang, S.-F. , Yang, S.-C. , Lien, S.-H. , 2011. Pseudocontinuous arterial spin la- beling perfusion magnetic resonance imaging —a normative study of reproducibility in the human brain. Neuroimage 56, 1244–1250 .
Wu, W.C. , Fernández-Seara, M. , Detre, J.A. , Wehrli, F.W. , Wang, J. , 2007. A theoretical and experimental investigation of the tagging efficiency of pseudocontinuous arterial spin labeling. Magn. Reson. Med. 58, 1020–1027 .
Yerys, B.E. , Herrington, J.D. , Bartley, G.K. , Liu, H.S. , Detre, J.A. , Schultz, R.T. , 2018. Arterial spin labeling provides a reliable neurobiological marker of autism spectrum disorder. J. Neurodev. Disord. 10, 32 .
Zhao, L. , Vidorreta, M. , Soman, S. , Detre, J.A. , Alsop, D.C. , 2017. Improving the robust- ness of pseudo-continuous arterial spin labeling to off-resonance and pulsatile flow velocity. Magn. Reson. Med. 78, 1342–1351 .
Zimine, I. , Petersen, E.T. , Golay, X. , 2006. Dual vessel arterial spin labeling scheme for regional perfusion imaging. Magn. Reson. Med. 56, 1140–1144 .
