Multi-centre, multi-vendor reproducibility of 7T QSM and R2* in the human brain: Results from the UK7T study

(1)

ContentslistsavailableatScienceDirect

NeuroImage

journalhomepage:www.elsevier.com/locate/neuroimage

Multi-centre,

multi-vendor

reproducibility

of

7T

QSM

and

R

₂

∗

_in

_the

human

brain:

Results

from

the

UK7T

study

Catarina

Rua

a,∗

_,

_William

_T.

_Clarke

b

_,

_Ian

_D.

_Driver

c

_,

_Olivier

_Mougin

d

_,

_Andrew

_T.

_Morgan

e

_,

Stuart

Clare

b

_,

_Susan

_Francis

d

_,

_Keith

_W.

_Muir

e

_,

_Richard

_G.

_Wise

c

_,

_T.

_Adrian

_Carpenter

a

_,

Guy

B. Williams

a

_,

_James

_B.

_Rowe

f,g

_,

_Richard

_Bowtell

d

_,

_Christopher

_T.

_Rodgers

a

a Wolfson Brain Imaging Centre, Department of Clinical Neurosciences, University of Cambridge, Box 65, Cambridge Biomedical Campus, Cambridge CB2 0QQ, United

Kingdom

b Wellcome Centre for Integrative Neuroimaging, FMRIB, Nuﬃeld Department of Clinical Neurosciences, John Radcliﬀe Hospital, University of Oxford, Oxford OX3

9DU, United Kingdom

c Cardiff University Brain Research Imaging Centre, School of Psychology, Cardiff University, Maindy Road, Cardiff CF24 4HQ, United Kingdom d Sir Peter Mansfield Imaging Centre, School of Physics and Astronomy, University of Nottingham, University Park, Nottingham NG7 2RD, United Kingdom e Imaging Centre of Excellence, Queen Elizabeth University Hospital, University of Glasgow, Langlands Dr, Glasgow G51 4LB, United Kingdom

f Department of Clinical Neurosciences and Cambridge University Hospitals NHS Trust, University of Cambridge, Herchel Smith Building, Cambridge Biomedical Campus,

Cambridge CB2 0SZ, United Kingdom

g Medical Research Council Cognition and Brain Sciences Unit, University of Cambridge, 15 Chaucer Road, Cambridge CB27EF, United Kingdom

a r t i c l e

i n f o

Keywords:

7 tesla MRI

Quantitative susceptibility mapping R 2∗ mapping

Multi-centre Reproducibility

a b s t r a c t

Introduction:Wepresentthereliabilityofultra-highﬁeldT2∗MRIat7T,aspartoftheUK7TNetwork’s“Travelling

Heads” study.T2∗-weightedMRIimagescanbeprocessedtoproducequantitativesusceptibilitymaps(QSM)and

R2∗maps.Thesereﬂectironandmyelinconcentrations,whicharealteredinmanypathophysiologicalprocesses.

TherelaxationparametersofhumanbraintissuearesuchthatR2∗mappingandQSMshowparticularlystrong

gainsincontrast-to-noiseratioatultra-highﬁeld(7T)vsclinicalﬁeldstrengths(1.5–3T).Weaimedtodetermine theinter-subjectandinter-sitereproducibilityofQSMandR2∗mappingat7T,inreadinessforfuturemulti-site

clinicalstudies.

Methods: Tenhealthyvolunteerswerescannedwithharmonisedsingle-andmulti-echoT₂∗_-weighted

gradi-entechopulsesequences.Participantswerescannedﬁvetimesateach“home” siteandonceateachoffour othersites.Theﬁvesiteshad1×Philips,2×SiemensMagnetom,and2×SiemensTerrascanners.QSMandR2∗

mapswerecomputedwiththeMulti-ScaleDipoleInversion(MSDI)algorithm(https://github.com/ﬁl-physics/ Publication-Code).Resultswereassessedinrelevantsubcorticalandcorticalregionsofinterest(ROIs)deﬁned manuallyorbytheMNI152standardspace.

ResultsandDiscussion: Meansusceptibility(𝜒)andR₂∗_values_agreed_broadly_with_literature_values_in_all_ROIs.

Theinter-sitewithin-subjectstandarddeviationwas0.001–0.005ppm(𝜒)and0.0005–0.001ms−1_(R 2∗).For𝜒

thisis2.1–4.8foldbetterthan3Treports,and1.1–3.4foldbetterforR2∗.ThemedianICCfromwithin-and

cross-siteR₂∗_data_was_0.98_and_0.91,_{respectively.}_Multi-echo_QSM_had_greater_variability_vs_single-echo_QSM

especiallyinareaswithlargeB0inhomogeneitysuchastheinferiorfrontalcortex.Acrosssites,R2∗valueswere

moreconsistentthanQSMinsubcorticalstructuresduetodiﬀerencesinB0-shimming.Onabetween-subjectlevel,

ourmeasured𝜒 andR2∗cross-sitevarianceiscomparabletowithin-sitevarianceintheliterature,suggestingthat

itisreasonabletopooldataacrosssitesusingourharmonisedprotocol.

Conclusion: TheharmonizedUK7Tprotocolandpipelinedeliversonaveragea3-foldimprovementinthe coeﬃ-cientofreproducibilityforQSMandR2∗at7Tcomparedtopreviousreportsofmulti-sitereproducibilityat3T.

Theseprotocolsarereadyforuseinmulti-siteclinicalstudiesat7T.

1. Introduction

Neurodegenerativediseasesareasigniﬁcantglobalhealthburden. Inmany instances,neurodegenerationis associated withthe

deposi-∗_{Corresponding}_author.

E-mailaddress:cr439@cam.ac.uk(C.Rua).

tion of iron in the brain. Understanding the patterns of deposition and their association with other risk factorsis a key area of clini-calresearch,butprogresshasbeenlimitedbytheneedtoscaleover multi-centre trials(Moeller etal., 2019). TheEUFIND(Düzel etal., 2019)isanexampleofanetworkfocusedonadvancementsin neurode-generativeresearchbyrunninglarge-scalemulti-centreimaging stud-ies.Also,theUK7Tnetwork(http://www.uk7t.org)hasrecentlyruna https://doi.org/10.1016/j.neuroimage.2020.117358

Received22April2020;Receivedinrevisedform3September2020;Accepted3September2020 Availableonline9September2020

(2)

multi-sitestudywithadementiacohorttoassessfeasibilityinpatient groups.ImagingaspartoftheC-MOREstudy(Capturingthe MultiOR-ganEﬀectsofCOVID-19)isalsoincludingharmonizedmulti-centre se-quenceswhichmightprovideinsightsintothelong-termimpactin sur-vivorsofCOVID-19( https://oxfordbrc.nihr.ac.uk/brc-funds-new-covid-19-research-projects/).Yet,inordertoperformsuchmulti-centre stud-ies,itisnecessarytoﬁrstguaranteetheconsistencyandreproducibility ofimagingmarkers.

Apopularapproachtoestimatingironconcentrationinthehuman brainusesgradient-echo(GE)magnetic resonanceimaging(MRI).In grey matter,ironis mainlyfound in theprotein ferritinwhich, due toitsantiferromagneticcoreandthepresenceofuncompensatedspins atthesurfaceorinthecore,exhibitsasuperparamagneticbehaviour (Makhlofetal.,1997;Langkammeretal.,2012).Thisparamagneticiron interactswiththeMRIscanner’sstaticmagneticfield(B₀)causinglocal dipolarfieldperturbations.Theseaccentuatetherateoftransverse sig-naldecaycausingT2∗relaxationinsurroundingtissue,whichisvisible asdecreasingsignalamplitudewithincreasingechotimeinaseriesof GEimages.Thiseffectcausesanincreaseintherateoftransverse re-laxation,R2∗,whichcorrelateswellwithnon-hemeironconcentrations ingreymatter(Gelmanetal.,1999;Langkammeretal.,2010),andhas beenusedtoinvestigatethedistributionofironinthehealthybrainand indisease(Haackeetal.,2005;Yaoetal.,2009;Lietal.,2019).

Thelocalpresenceofiron(andtoalesserextentmyelin and cal-cium)alsoaffectsthesignalphaseofGEimagesbecauseoftheeffect ofthefieldperturbationonthelocalLarmorfrequency(Houseetal., 2007;Heetal.,2009;Leeetal.,2012).QuantitativeSusceptibility Map-ping(QSM)methodsattempttodeconvolvethesedipolephasepatterns toidentifythesourcesofthemagneticfieldinhomogeneity.Inother words,QSMestimatesquantitativemapsoftissuemagnetic suscepti-bility𝜒 fromGEphasedata(LiandLeigh,2004;Reichenbach,2012; WangandLiu,2015).Thisapproachhasshownsensitivitytoseveral neurologicalconditions(Lotfipouretal.,2012;Acosta-Cabroneroetal., 2013;Blazejewskaetal.,2015;Acosta-Cabroneroetal.,2016)and of-fersadvantagesovermagnitudeR2∗suchashavingreducedblooming artifactsorbeingabletodistinguishbetweenparamagneticand diamag-neticsubstances(Eskreis-Winkleretal.,2017).

R₂∗_imaging_and_QSM_have_been_shown_to_provide_reproducible re-sultsinsingle-siteandcross-sitestudiesat1.5Tand3T(Hinodaetal., 2015;Cobzasetal.,2015;Dehetal.,2015;Linetal.,2015;Santinetal., 2017;Fengetal.,2018;Spincemailleetal.,2019).

Thedipole-inversionproblemattheheartofQSMmethodsbenefits fromtheincreasedsensitivitytomagneticsusceptibilityvariationand spatialresolutionatultra-highfields(B₀≥7T)(Yacoubetal.,2001; Reichenbach etal., 2001; Tie-Qianget al.,2006; Duyn etal., 2007; WhartonandBowtell, 2010).At7T, closeattentionmustbe paidto B₀shimmingandgradientlinearitytoachieveaccurateQSMandR₂∗ mapping(Yangetal.,2010).Headpositionisalsoanimportantfactor thataffectsthesusceptibilityanisotropy(Lancioneetal.,2017;Lietal., 2017).

Inthisstudy,weintroducesingle-echoandmulti-echoGEimaging protocolsforQSMandR2∗mappingat7Twhichwerestandardisedon threediﬀerent7TMRIscannerplatforms,fromtwodiﬀerentvendors. WeappliedthisstandardisedprotocolintheUK7TNetwork’s “Travel-lingHeads” studyon10subjectsscannedat5sites.Wereport repro-ducibilityforderivedR2∗andQSMmapsandmakerecommendations forthedesignoffuturemulti-centrestudies.

2. Methods

2.1. Measurementsetup

Tenhealthyvolunteers(3female,7male;age32.0±5.9years)were recruited:comprisingtwosubjectsfrom eachof theﬁve7Timaging

sites in theUK7TNetwork (describedin Table 1).Each subjectwas scannedfivetimes attheir “home” site,andonceat theothersites, under local ethicsapproval for multi-site studiesobtained at Site-4 (HBREC.2017.08).Scansforeachsubjectwerecompletedwithina pe-riodofbetween83and258days.Thefivehome-sitescanswere per-formed acrossdifferent sessions: themedian timetoacquireall five scanswas59days(range:3–71days).

In every scan session, B0 shimming was performed using the vendors’ default second-order (or third-order for Site-4 and Site-5) B0-shimming routines. B1+-calibration was performed initially using the vendor’s default adjustment scans. A 3D DREAM se-quence (Nehrke and Bornert, 2012; Ehses et al., 2019) was subse-quently acquired and the transmit voltage (or power attenuation) was then adjusted for all subsequent imaging based on the mean flip-angle from the brain in an anatomically-specified axial slice of the 3DDREAMflip anglemapas described in Clarkeet al.(2019). Single-echo 0.7mm isotropic resolution T2∗-weighted GE data were thenacquired with: TE/TR=20/31ms;FA=15°;bandwidth=70Hz/px; in-plane acceleration-factor=4 (Sites-1/2/4/5) or 2 × 2 (Site-3); FOV=224 × 224 × 157mm3_; _scan-time₌_~9min. _Multi-echo _1.4mm isotropic resolution T₂∗_-weighted _GE _data _were _acquired _with: TE1/TR=4/43ms; 8echoes with monopolar gradientreadouts; echo-spacing=5ms; FA=15°; bandwidth=260Hz/px; acceleration-factor=4 (Sites-1/2/4/5)or2×1.5(Site-3);FOV=269×218×157mm3_; scan-time~6min (Sites-1/2/4/5)and~4min(Site-3).ForSiemensdata, coilcombinationwasperformedusingacustomimplementationof Roe-mer’salgorithm,aspreviouslydescribed(Clarkeetal.,2019).Subject 6’ssingle-echoscanfailedtoreconstructusingRoemer’smethodondata fromthefirstvisitatSite-5soasum-of-squares(SoS)algorithmwasused forcoilcombinationforthatscaninstead.A0.7mmisotropicMP2RAGE scanwasusedforwithin-andcross-siteregistrationaspreviously de-scribed(Mouginetal.,2019).

2.2. QSMandR₂∗_data_processing

QSMmapsweregeneratedfromboththesingle-echoandmulti-echo T2∗-weighted datasetsusing theMulti-ScaleDipoleInversion(MSDI) algorithm,asimplementedinQSMboxv2.0(Acosta-Cabroneroetal., 2018). Briefly:first the local fieldwas estimated by phase unwrap-ping(Abdul-Rahmanetal.,2005)andmagnitude-weightedleastsquares phaseechofittingwasperformedonthemulti-echodata.Then, indepen-dentlyforbothsingle-echoandmulti-echodata,backgroundfieldwas removedusingtheLaplacianBoundaryValue(LBV)methodfollowed bythevariableSphericalMeanValue(vSMV)algorithmwithan ini-tialkernelradiusof40mm(Zhouetal.,2014;Acosta-Cabroneroetal., 2018).MSDIinversionwasestimatedwithtwoscales:theself-optimised lambdamethodwasusedonthefirstscalewithfilteringperformed us-ingakernelwith1mmradius,andonthesecondscalethe regulariza-tiontermwassetto𝜆=102.7_(the_optimal_value_for_in-vivo_7T_datasets foundin(Acosta-Cabroneroetal.,2018))andfilteringwasdonewitha kernelradiussetto5mm.Brain masksusedintheanalysiswere ob-tained with FSL’sBrain Extraction Tool(BET) with fractional inten-sitythreshold=0.2forsingle-echodata(Smith,2002).Thesewerethen mappedtomulti-echodataspace.

Onthemulti-echodata,QSMwasreconstructedsevenmoretimes: withonlyoneechoat19ms(matchingthesingle-echodata),withthe twoshortestechoes(i.e.TE₁/TE₂=4/9 ms),withthethreeshortest echoes(i.e.TE1/TE2/TE3=4/9/14ms),andsoforth.

Onthemulti-echodataset,voxel-wisequantitativemapsofR2∗were obtainedusingtheAuto-RegressiononLinearOperations(ARLO) algo-rithmforfastmonoexponentialfitting(Peietal.,2015).R2∗wasalso fit-tedfivemoretimes:withdatafromthefirstthreeechoes(TE1/TE2/TE3 =4/9/14 ms), then with the first four echoes (TE1/TE2/TE3/TE4 =4/9/14/19ms),andsoforth.

(3)

Table1

DetailsofthescannersandhardwareusedfortheUK7TNetwork’sTravellingHeadsstudy.

# Site Vendor Scanner Model Gradient Performance

Installation Date (Month-Year)

Software Version 1 Wellcome Centre for Integrative Neuroimaging (FMRIB),

University of Oxford

Siemens Magnetom 7T 70 mT m −1 _Dec-2011 _VB17A

200 mT m −1 ms −1

2 Cardiff University Brain Research Imaging Centre, Cardiff University

Siemens Magnetom 7T 70 mT m −1 _Dec-2015 _VB17A

200 mT m −1 ms −1

3 Sir Peter Mansﬁeld Imaging Centre, University of Nottingham Philips Achieva 7T 40 mT m −1 _Sep-2005 _R5.1.7.0

200 mT m −1 ms −1

4 Wolfson Brain Imaging Centre, University of Cambridge Siemens Magnetom Terra 80 mT m −1 _Dec-2016 _VE11U

200 mT m −1 ms −1

5 Imaging Centre of Excellence, University of Glasgow Siemens Magnetom Terra 80 mT m −1 _Mar-2017 _VE11U

200 mT m −1 ms −1

2.3. Dataregistration

The neckwas cropped from the magnitude data with FSL’s “ro-bustfov” command (https://fsl.fmrib.ox.ac.uk/fsl/), applied to the single-echo data and the 4th echo of the multi-echo data. High-resolution single-echo and multi-echo templates were made from this cropped data for each subject with antsMultivariateTemplate-Construction2.sh from the Advanced Normalization Tools (ANTs, http://stnava.github.io/ANTs/).Twoapproacheswerecompared: trans-formationsusingrigidregistrationwithmutualinformationsimilarity metric(denotedas“Rigid” below)orusingsymmetricdiﬀeomorphic im-ageregistrationwithcross-correlationsimilaritymetric(denoted“SyN” below).Othersettingswerekeptthesameforbothapproaches:4steps with0.1gradientstepsize,maximumiterationsperstep1000,500,250 and100,smoothingfactorsperstepof4,3,2,and1voxels,andshrink factorsperstepof12×,8×,4×,and2×.Theresultingregistrationswere thenappliedtotheQSMandR2∗mapswhichwereaveragedtocreate single-echoandmulti-echoQSMandR2∗templatesforeachsubject.

2.4. SelectionofRegionsofInterest(ROIs)

Fiveregionsofinterest(SubstantiaNigra,RedNucleus,Caudate Nu-cleus,PutamenandGlobusPallidus)weremanuallysegmentedbased onthesubject-speciﬁcQSMtemplatesofthesingle-echodataregistered withthe“SyN” approach.Inordertominimizetheamountof segmenta-tionvariability,theseROIswerethenmappedtothesingle-echo“Rigid”, andmulti-echo“SyN” andmulti-echo“Rigid” spaceswithnearest neigh-bourinterpolationandvianon-linear registrationsobtainedwiththe defaultsettingsintheantsRegistrationSyN.shcommandinANTs.

MagnitudedatawereﬁrstregisteredtotheT1-weightedMP2RAGE scans(Rigidtransformations;MIsimilaritymetric)andlatertothe stan-dardT1“MNI152brain” (MontrealNeurologicalInstitute152)(using settingsinantsRegistrationSyN.sh)appliedtothesingle-echodataand tothe1stechoofthemulti-echodata.Theseregistrationswerethen usedtomapthe48probabilisticcorticalROIs,“cortical ROIs”,from theHarvard-OxfordCorticalAtlasandthe21probabilisticsubcortical ROIs,“subcorticalROIs”,fromtheHarvardOxfordSubcorticalAtlasto theQSMandR₂∗_template_spaces.

The T1-weighted MP2RAGE data was bias-field corrected, brain extracted, and segmented into five tissues using SPM (https://www.fil.ion.ucl.ac.uk/spm/): the grey matter (GM), white matter (WM)and cerebral-spinal fluid (CSF) volumes were mapped intoeachsubject-specificQSMtemplatespace.Then,using“fslmaths” from FSL (https://fsl.fmrib.ox.ac.uk/fsl/), the mapped cortical ROIs werethresholdedat10%ofthe“robustrange” ofnon-zerovoxelsand multipliedbytheGMtissuemapinordertoobtainGM-specificcortical ROIs.Themapped subcorticalROIs werethresholded at50%of the “robustrange” ofnon-zero voxels. Fromthese,anyCSF voxelswere excludedfromtheleftandrightCaudateNucleus,PutamenandGlobus Pallidus,andthevoxelsetsfromtheleftandrightcounterpartswere mergedtogether.

Fromthesingle-echoandmulti-echodata,average𝜒 andR2∗values wereextractedfromthemanualandatlas-basedROIsforallvolunteers andsessionsintemplatespace(valuesgiveninSupplementarymaterial 1).

Inordertoestimatewherethemagneticfieldisspatiallymore vari-able,field-mapswerefirstestimatedfromthemulti-echodatasets.ΔB₀

was calculated from thebackground fieldremoval step of the QSM pipeline,andwasdefined,per-voxel,astheaveragedifferencebetween thefieldinavoxelanditsimmediatenearestneighbours.Theaverage ΔB0 wasextractedforeach ofthecorticalROIs andaveragedacross allsubjectsandsessions.ThenthecorticalROIsweredividedintotwo groupsbasedontheΔB₀values:wherever|ΔB₀|>0.005HztheROIwas groupedinto“highΔB0” regions,otherwiseitwasgroupedinto“low ΔB0” regions.ΔB0 wascalculatedfromthemulti-echopipelineonly, asdifferencestovaluescalculatedusingsingle-echodatawereminimal (Fig.1,Supplementarymaterial2).

WeexploredthreepossiblesusceptibilityreferenceregionsforQSM processing.TheaverageQSMsignalwasextractedfrom:

1 Awholebrainmask,“wb”;

2 Awhole-brainCSFmaskerodedintwosteps,“csf”;

3 AmanuallyplacedcylindricalROIintherightventricle,“cyl” (across allsubjectstheROIvolumewas104±11mm3_).

2.5. Statisticalanalysis

StatisticalanalysiswasperformedwithR3.5.3(RCoreTeam,2013). Cross-siteanalysisusedonlythe1stscanatthe“home” sitealongwith thescansattheotherfoursites.Toobtainthewithinsubjectaverage, AVw,the𝜒 andR2∗valueswereaveragedwithinthesamesiteandacross thesitesandthenaveragedacrosssubjects:

AV_𝑤= ∑𝑚 𝑖=1( ∑𝑛 𝑗=1𝑥𝑖𝑗∕𝑛) 𝑚 (1)

wherenisthenumberofsessions(𝑛=5forwithin-siteandcross-site) andmthenumberofsubjects.Relativereliabilitywasmeasuredusing theintra-classcorrelationcoeﬃcient(ICC)fromwithin andcross-site dataindependentlyforeachROI(Weir,2005):

ICC= 𝑀𝑆𝑏−𝑀𝑆𝑤

𝑀𝑆𝑏+𝑀𝑆𝑤(𝑛−1) (2)

whereMS_bandMS_warethebetween-subjectsandwithin-subjectsmean squarefromarandom-eﬀects,one-wayanalysisofvariance(ANOVA) model.Intra-subjectabsolutevariabilityisassessedbymeasuringthe within-subject standard-deviation (SD_w) calculated as (Santin et al., 2017): 𝑆𝐷𝑤= √_∑ 𝑚 𝑖=1𝜎𝑖2 𝑚 with𝜎𝑖= √_∑ 𝑛 𝑗=1(𝑥𝑖𝑗−̄𝑥𝑖)2 𝑛−1 (3) where𝑥𝑖= 𝑛 ∑

𝑗=1𝑥𝑖𝑗∕𝑛isthereplicateaverageforeachsubject.SDwwas computedusingwithin-sitedataandcross-sitedataindependently.

(4)

Simi-larly,cross-subjectvariabilitywascalculatedbymeasuringthe between-subjectstandard-deviation(SD_b): 𝑆𝐷𝑏= √_∑ 𝑚 𝑖=1 ∑𝑛 𝑗=1 ( 𝑥𝑖𝑗−𝑥𝑎𝑣𝑔)2 𝑛×𝑚−1 (4) where𝑥𝑎𝑣𝑔= 𝑚 ∑ 𝑖=1 𝑛 ∑

𝑗=1𝑥𝑖𝑗∕(𝑛×𝑚)isthemeasurementaverageacross sub-jectsandsessions.NotethatSDbiscomputedusingdatafromallsites.

StatisticaltestingonAV_w,SD_wandICCvaluesextractedfrom man-ualandtemplate-basedROIswasdonebyfirstfittingthedatawith nor-mal,log-normal,gammaandlogisticdistributions.Thegoodness-of-fit statisticsfortheparametricdistributionswerecalculatedandthe dis-tributionwhichshowedthelowestAkaikesInformationCriterion(AIC) wasthenusedonagenerallinearmodelfitting.Allmodelsincludedas fixedmaineffectsROInumberanddatatype(within-andcross-site). Whenevaluating thedataregistrationtype, themodelalsoincluded registrationtype(“Rigid” and“SyN”)asafixedmaineffect.When test-ingforQSMreference,themodelalsoincludedreferenceregion(“wb”, “csf”,and“cyl”)asafixedmaineffect.Onmulti-echoQSMdata,amodel wasfittedwhichalsoincludedthenumberofechoesprocessedasafixed maineffect.WhencomparingthemanualandsubcorticalROIs,theROI type(manualvs.atlas-based)wasalsoincludedasafixedmaineffect. Finally,onthedatafromthecorticalROIs,ROInumberwasreplaced with“highΔB0” and“lowΔB0” ROItypeascovariate.Ap-valueless than0.05wasconsideredsignificant.

2.6. Headorientation

WeinvestigatedtheeﬀectofheadorientationonQSMvariability. Sinceallourdatawasacquiredwithtransversesliceorientation,the slicenormalvectorintheacquiredimageswasparalleltoB0.Weused theper-subjectrotationmatricesoftheaﬃnetransformsfromthis ac-quiredtransversedatatoMNIspacetoestimatethez-axis rotation𝜃

withrespecttotheB0vector(0,0,1)(Fig.7(A)):

𝜃 =cos−1(_𝑀 33

)

whereM₃₃isthe3rdrow,3rdcolumnoftheaffinetransformmatrix. Twolinearmixedeffectsmodels,‘mod1’and‘mod2’,werefittedon thewithin-siteandcross-site𝜒 dataseparately:bothmodelsincluded site,ROI,andsessionnumberasfixedeffects,andsubjectnumberasa randomeffect,while‘mod2’alsoincluded𝜃 asafixedeffect.Foreach model,theR2_was_evaluated_and_both_models_were_compared_with_a chi-squaredtest.

Finally,from‘mod2’the𝜃 ﬁtcoeﬃcientswereusedtoestimate cor-rected𝜒-values basedonachosen standard𝜃 forallofthe measure-ments(𝜃_{𝑛𝑜𝑟𝑚}=0.52radians).Then,newwithin-siteandcross-siteSDw ofthecorrected𝜒 were calculatedbasedonthesameapproachasin Section2.5.

3. Results

Fig. 1 shows QSM and R2∗ maps for one example subject. Basal ganglia structures, including Caudate Nucleus, Putamen and GlobusPallidus are clearlyvisible consistentwith previous ﬁndings (Langkammeretal.,2010;Wangetal.,2015;Bettsetal.,2016; Acosta-Cabroneroetal.,2016).Fig.2,Supplementarymaterial2highlightsthe diﬀerenceinQSMdataqualitywhenusingourchosenRoemercoil com-binationmethodvsusingsum-of-squarescoilcombination.

3.1. QSMandR2∗resultsandliterature

Fig.2comparesaverage𝜒 andR2∗valuescalculatedinthisstudy intheﬁvemanualROIsandthreecorrespondingatlas-basedsubcortical ROIsagainstliteratureranges.Thesingle-echo𝜒-valuesandmulti-echo

𝜒-valuesfromthisstudyareconsistentwithliteraturevaluesat1.5T,3T and7T.R2∗valuesfromthisstudyalsoagreecloselywith7Tliterature values.

3.2. ReproducibilityofQSMandR2∗

Fig.3showsboxplotsoverROIsofthewithin-andcross-siteAVw(A), SDw(B)andICC(C)valuesforthemanualROIsonthe𝜒 andR2∗maps. TheAV_w fromR₂∗_maps_measured_on_the_same_site _is_{systematically} highercomparedtotheAVwmeasuredacrosssites(p<0.0001;e.g.,on thePutamenROI,AVw_within-site=0.0493ms−1vsAVw_cross-site=0.0489 ms−1_)._On_this_comparison,_QSM_data_did_not_show_signiﬁcant diﬀer-encesbetweenwithin-siteandcross-sitegroupsforeithersingle-echo data(p=0.053)ormulti-echodata(p=0.65).

FromallthedatainthemanualROIs,themedianSDwofsingle-echo

𝜒-values wasapproximately29%lowerthanformulti-echo𝜒-values

(p=0.0010).TherewasasigniﬁcantlylargerSDwcross-sitecompared towithin-siteonsingle-echo𝜒 data(p<0.0001;e.g.,onthePNROI, SD_{w_within-site}=0.00088ppmvsSD_{w_cross-site}=0.0014ppm),multi-echo

𝜒 (p=0.033)andonR2∗data(p<0.0001).

The ICC values for within- and cross-site R2∗ data (medianICC was0.98and0.91,respectively)werefoundtobesigniﬁcantlyhigher thanvaluesforsingle-echo𝜒 (medianICCwas0.89and0.64, respec-tively)orformulti-echo𝜒 (medianwasICC0.76and0.38,respectively) (p=0.00011).Forallmeasurements,theICCforcross-sitedatawas sig-niﬁcantlylowerthanforwithin-sitedata(single-echoQSM:p<0.0001; multi-echoQSM:p=0.017;R2∗:p<0.0001).

SimilarstatisticswereobtainedforAVw,SDwandICCmeasurements intheAltas-basedcorticalROIs(Table2Supplementarymaterial2).

3.3. Registration

Thewithin-andcross-sitestandarddeviationsforoneaxialslicefrom oneexamplesubjectusing“Rigid” and“SyN” registrationapproaches areshowninFig.4.Generally,withbothregistrationmethods, within-siteandcross-siteSD_wincreasesinveins,intheorbitofrontalregionsand atthecorticalsurface(whiteandgreenarrows,Figure4).Theseare ar-easassociatedwithlargeB0inhomogeneitiesandgradientnon-linearity. However,thereisadecreaseinthecross-sitestandarddeviationinthe orbitofrontalregionandclosetotheedgesofthecortexwhenusingthe “SyN” comparedtothe“Rigid” method(greenarrows,Fig.4).

OnthemanualROIsincreasedvariabilitywasobservedforR₂∗_on “Rigid” registereddatacomparedto“SyN” (SDw:p<0.0001;ICC:p= 0.013)butnotforsingle-echoormulti-echo𝜒:forexample,themedian cross-siteR2∗SDwfromallROIswas0.00066ms−1using“SyN” method and0.00086ms−1_using_the_“Rigid_{” registration}_method._On_the atlas-basedcorticalROIs,thesamesigniﬁcanttrendwasobservedforR2∗and single-echo𝜒 data(Table2Supplementarymaterial2).

3.4. QSMreferencing

ToassesstheoptimalQSMsusceptibilityreferencing,Fig.5shows boxplotsoftheSD_wforsingle-echoandmulti-echo𝜒 usingdiﬀerent ref-erencingmethodsonthemanualROIs.Onsingle-echo𝜒 data,compared to“wb” correction(chosencorrectionforthisstudy),the“csf” reference didnotincreasesigniﬁcantlytheSD_w(p=0.93)butwith“cyl” the me-dianSDwincreasedbyapproximately14%(p<0.0001).

Multi-echo𝜒 datashowedanincreaseinthemedianSDwof, respec-tively,11%(p=0.00096)and8%(p=0.00064)whenusing“csf” and “cyl” methodsforcorrection.Theeﬀectofvaryingthereferencingof QSMdatawassimilarinwithin-siteandcross-sitedata,forallmethods tested.

3.5. Multi-echoQSMandR2∗

OnaverageacrossallthemanualROIsandcomparedtosingleecho data,multi-echodata(usingtwoormoreechoes)showedasigniﬁcant 14%increaseoftheSDw(Fig.6)and3%oftheICC(Table1, Supplemen-tarymaterial2).Thissupportsthesingle-echoandmulti-echo𝜒 compar-isoninSection3.2.Similarbehaviourwasobservedontheatlas-based

(5)

Fig.1. Representative slicesof single-echo𝜒

(A)multi-echo𝜒 (B)andR2∗maps(C)froman

examplesubject templates.TherightCaudate Nucleus(a),Putamen(b)andGlobusPallidus (c)areshowningreen.Multi-echo𝜒 maps cal-culatedwithdatafromall8echoes.

Fig.2. MeanandstandarddeviationliteraturevaluesofQSM(A)andR2∗(B).Themeanandstandarddeviationresultsfromthisstudyarealsoplotted.Fordatawith

thesymbol‘§’thestandarderrorofthemeanwasoriginallyreportedandhasbeenrescaledbyreportedN.Shadedregionscorrespondtoliteraturedata.Multi-echo

𝜒-mapswerecalculatedwithdatafromalleightechoes(Choietal.,2019;Deistungetal.,2013;KeuKenetal.,2017;Barbosaetal.,2015;Schweseretal.,2011;

(6)

Fig.3. BoxplotsfromdataobtainedonthemanualROIsofwithin-andcross-siteAV_w(A),SD_w(B)andICC(C)ofsingle-echoandmulti-echoQSM,andR₂∗_._Data

fromeachROIisshownwithadiﬀerentmarkerforeachboxplot.Legend:SN=SubstantiaNigra;RN:RedNucleus;CN:CaudateNucleus;Pu:Putamen;GP:Globus Pallidus.ThevariabilityinAVwreﬂectsthenaturalvariationofironcontentinsubcorticalstructuresinthehealthybrain.Multi-echo𝜒-mapswerecalculatedwith

datafromalleightechoes.

corticalROIs(Table2Supplementarymaterial2).OnthemanualROIs, thereisnosigniﬁcantdiﬀerenceinAVw(p=0.79)orinSDw(p=0.11) from𝜒 computedfrommultipleechoes (i.e.2ormoreechoesinthe QSManalysis).Yet,in theatlas-basedcorticalROIs,long echotimes (i.e.using6ormoreechoes)showedanaverageincreaseof15.7%in SD_w (p<0.0001)comparedtousing2to5echoesandadecreaseof 1.75%inICC(p<0.0001)(Table2Supplementarymaterial2).

InthemanualROIs,R2∗showednosignificantchangeinvariability acrossallROIswhendifferentnumberofechoeswereusedinthe fit-ting(SDw:p=0.11;ICC:p=0.95)(Fig.6(B))oronAVw(p=0.97). Intheatlas-basedcorticalROIs,thenumberofechoesusedinfluenced theaverageR₂∗_value_(AV

w:p<0.0001),weaklyICC(p=0.021),but

notSDw(p=0.61).Tables1and2Supplementarymaterial2display individualstatistics.

3.6. ROIselection

There is a smallbut significanthigher average 𝜒 frommanually drawnROIscomparedtotheatlas-basedsubcorticalROIsinsingle-echo QSMdata (p < 0.0001;e.g. 0.042±0.009 ppmvs 0.033±0.010 ppm inthecaudatenucleus)andinmulti-echoQSMdata(p<0.0001;e.g. 0.048±0.010ppmvs0.038±0.011ppminthecaudatenucleus)(Fig.2). Similarly,forR2∗(e.g.0.041±0.004ms−1vs0.039±0.006ms−1inthe caudatenucleus)thisdifferencewassignificant(p<0.0001).In

(7)

addi-Fig.4. Voxel-wisewithin-andcross-site stan-dard deviation of an example subject from single-echoandmulti-echoQSMandR2∗data

with data registered with “Rigid” (A) and “SyN” (B) transformations. Arrows point to regions where the SDw decreased with the

“SyN” transformations(green)arecomparedto “Rigid” (white).TherightCaudateNucleus(a), Putamen(b)andGlobusPallidus(c)are out-linedinwhite.Multi-echo𝜒-mapswere calcu-latedwithdatafromalleightechoes.

Fig.5.BoxplotsfromdataobtainedonthemanualROIsofwithin-andcross-siteSDw(redandgreen,respectively)ofsingle-echoQSM(A)andmulti-echoQSM

(B)withawhole-brainreference(wb),withacsfreference(csf),andwithacylinderreference(cyl).DatafromeachROIisshownwithadiﬀerentmarkerforeach boxplot.Legend:SN=SubstantiaNigra;RN:RedNucleus;CN:CaudateNucleus;Pu:Putamen;GP:GlobusPallidus.Multi-echo𝜒-mapswerecalculatedwithdatafrom alleightechoes.

tion,theSDwwas,onaverage,approximatelytwotimeshigherandthe ICClowerintheatlas-basedsubcorticalROIscomparedtothemanual ROIsinalldatasets(SDw:single-echoQSMp<0.0001,multi-echoQSM

p<0.0001,R2∗p<0.0001;ICC:single-echoQSMp=0.00021, multi-echoQSMp=0.0023,R₂∗_p₌_0.012)._So,_ROI_selection_should_be_done consistentlyinastudy.

3.7. Spatialdistributionofthemagneticﬁeld

Ontheatlas-basedcorticalROIstheSDwincreasedbyapproximately 28%and88%on“highΔB0” regionscomparedto“lowΔB0” regionson multi-echo𝜒 andR₂∗_data,_respectively₍_p₌_0.0011_and_p_<_0.0001) (Table2Supplementarymaterial 2).Similarly,ICCvalues decreased signiﬁcantlyforsingle-echoandmulti-echo𝜒 andR2∗values.

(8)

Fig.6. BoxplotsfromdataobtainedonthemanualROIsofwithin-andcross-siteSD_wformulti-echoQSM(A)andR₂∗_(B)_calculated_with_diﬀerent_number_of

echoes.IncreasingtrendonthemedianSDwobservedwithincreasingnumberofechoeswasobservedontheQSMdata(dottedgreenandreddottedlinesin(A)).

Legend:SN=SubstantiaNigra;RN:RedNucleus;CN:CaudateNucleus;Pu:Putamen;GP:GlobusPallidus.

Table2

Within-sitetoCross-siteratioofthemedianSDwobtainedfromallﬁvemanually-deﬁnedROIsonsingle-echoandmulti-echo

𝜒 withoutandwith𝜃-correction.

Parameter within-site/cross-site median SD w (without 𝜃-correction) within-site/cross-site median SD w (with 𝜃 -correction)

Single-echo 𝜒 0.82 0.67

Multi-echo 𝜒 0.91 0.82

3.8. QSMvariabilitywithheadorientation

Whenanalysing𝜒 inthemanually-deﬁnedROIswithrespectto𝜃,

aconsistentnegativetrendwasobservedforallsubjects.Fig.7(C)and (D)showsanexamplefortheanalysisintheGlobusPallidusROI. Fit-tingalinearmodelon𝜒,with𝜃 andROIasﬁxedvariables,𝜃 showed

asigniﬁcantnegativecorrelationwithsingle-echo𝜒 (p<0.0001)and multi-echo𝜒 (p=0.015).

Inaddition,for𝜃,thewithin-siteSDwwasnearlyhalfofthe cross-siteSDw(0.011and0.028radians,respectively),indicatingthatthere waslargervariabilityinheadorientationacrosssites(subject-wise vari-abilityof𝜃,𝜎iofequation[3],isplottedinFig.7(B)).

Separately for within-site andcross-site 𝜒 data, we assessed the goodness-of-ﬁt of a modelcontaining 𝜃 as an explanatory variable. Onsingle-echowithin-site data,themarginalR2 _increased_from_0.71 with‘mod1’to0.76with‘mod2’(whichincludes𝜃)(Chi-squaredtest,

p=0.041).Thecorrespondingcross-siteR2_s_were:_0.77_and_0.80 (Chi-squaredtest,p=0.057).Onmulti-echodata,themarginalR2_increased from0.75with‘mod1’to0.79with‘mod2’onwithin-site data (Chi-squared test,p=0.041)andmaintainedat0.79onbothmodelsfor cross-sitedata(Chi-squaredtest,p=0.14).

Fromthecorrected𝜒-valuesat𝜃norm,resultsshowaslightdecrease intheratioofwithin-sitetocross-siteSDw(Table2),butvariabilityof

𝜒 obtainedfromcross-sitedatawasstillhigherthanfromwithin-site(𝜒 with𝜃-correction,p=0.01;uncorrected𝜒,p<0.0001(Section3.2)). For multi-echodata, theSDw obtained from the corrected 𝜒-values

weresimilaronwithin-sitecomparedtocross-site(𝜒 with𝜃-correction,

p=0.11;uncorrected𝜒,p=0.033(Section3.2)). 4. Discussion

Inthispaper,thereproducibilityofQSM𝜒 andR2∗measurements incorticalandsubcorticalregionsofthebrainwasassessedforthefirst timeinamulti-sitestudyat7Tfortwodifferentprotocols(asingle-echo 0.7mmisotropicT2∗-weightedscananda1.5mmisotropicmulti-echo T₂∗_-weighted_scan),_using_three_different_scanner_platforms_provided_by twodifferentvendors.

Previousstudiesat1.5Tand3Thaveshowngoodreproducibilityfor

𝜒 andR₂∗_data_acquired_on_the_same_scanner_or_across_sites_(1.5T_and 3T)(Hinodaetal.,2015;Cobzasetal.,2015;Dehetal.,2015;Linetal., 2015;Santinetal.,2017;Fengetal.,2018;Spincemailleetal.,2019). IntermsofQSManddependingonthesubcorticalregion,intra-scanner 3TrepeatabilitystudiesreportanSDwof0.002–0.005ppm(Fengetal., 2018)and0.004–0.006ppm(Santinetal.,2017),andthecross-site3T studybyLinetal.(2015)reportedanaverageSDWof0.006–0.010ppm. Weobservedawithin-siteSDwrangeof0.0009–0.004ppmand cross-siteSDwrangeof0.001–0.005ppmat7T.Comparedto3Tstudies,this isa2.0–5.3folddecreaseinthewithin-siteSDw,anda2.1–4.8decrease inthecross-siteSD_w.

Therangeofwithin-siteSDwvaluesforR2∗wasaveraged0.0003– 0.001ms−1_in_our_study_and_the_cross-site_SD

wrangewas0.0005–0.001 ms−1_. _The_cross-site _values_are_comparable_to_the_same _site_reported

(9)

Fig.7. InQSM,itisassumedthatthemacroscopicsusceptibilityinanimagingvoxelisisotropic.However,ithasbeenshownthatthisassumptionistoosimplistic forsingleheadorientationQSMmethods,complicatingtheinterpretationoftheQSMresults(Lietal.,2017).Weinvestigatedtheeﬀectofheadorientationon QSMestimationinourdata:(A)ConsideringthatdatawasallacquiredinthetransverseplanewithB0perpendiculartotheimagingslice,subjectshadavariable

headrotation𝜃 withrespecttoB0.Toestimate𝜃,weusedMNIspaceasacommonheadorientation(zMNI)acrossallscans.FromtheaﬃneregistrationmatrixM

convertingacquireddataintoMNIspace,theangleofrotationfromtherotatedz-axis,z2,willbegivenby𝜃 =cos−1(𝑀33)whereM33isthe3rdrow,3rdcolumnof

theaﬃnetransformmatrix.(B)Subject-wisewithin-siteandcross-site𝜎imeasurementson𝜃.(C)Single-echoand(D)multi-echoscatterplotsof𝜒 measurements

accordingto𝜃 ontheGlobusPallidusmanualROI.Foreachsubjectalineartrendisalsoplottedandthefitcoefficientsaregivenintheplotlegend.Datafromeach siteisdisplayedwithadifferentsymbol.Multi-echo𝜒-mapswerecalculatedwithdatafromalleightechoes.

at 3T: 0.0005–0.0009ms−1 ₍_Feng _et_al., ₂₀₁₈_), _{0.0006–0.002} _ms−1 (Santinetal.,2017).Comparedtothelatter,ourcross-siteresultsshow a1.1–3.4improvementoverthesamebrainregionsinR2∗_variability.

ThestudyfromHinodaetal.(2015)measuredQSMreproducibility at1.5Tand3Tbyscanningsubjectstwiceoneachofthescanners.They showedthereisa1.1–2.1folddecreaseintheupperandlowerlimitsin Bland-Altmanplotsat3Tcomparedto1.5T,whichisinlinewiththe expectedsignal-to-noiseratio(SNR)increasebetweenthesetwoﬁeld strengths(Edelsteinetal.,1986;Wardlawetal.,2012).Comparedto 3Treports,thereis,onaverage,animprovementofapproximately 3-foldinourQSMandR2∗7Tmeasurementsofreproducibility.Thisis inlinewiththeexpectedSNRincreaseinbrainimagingfrom3Tto7T (Pohmannetal.,2016).

Thehighervaluesofcross-siteSDwcomparedtothewithin-site val-uesinourstudymaybeattributedtothediﬀerentgradientsystemsand automaticdistortioncorrectionsusedinthediﬀerentscannerplatforms

andtothedifferentapproachestoshimming,whichleadtodifferent ge-ometricaldistortionsanddropoutregions(Figs.3and4Supplementary material2)(Yangetal.,2010).Inourstudyweverifiedthatnotonly regionsinthecortexclosetoair-tissueinterfacesshowdifferencesinB0 acrossscanners,butalsolargesubcorticalregionssuchastheCN,the PuandtheGPROIs.

We alsoshowedthat theuse of anon-linear registrationmethod (here,“SyN” inANTs)significantlyreducedtheinter-scanner variabil-ityofcorticalQSMcomparedtorigid-bodyregistration,indicatingthat differences ingeometricdistortionacrossscanners werepresent.The R2∗resultsforbothcorticalandsubcorticalstructuresalsoshow sig-nificantlylowerinter-scannervariabilitywhenanon-linearregistration wasused.ForQSM,highercross-sitevariabilitymayalsobeattributed totheheadorientationwithrespecttoB0(Lancioneetal.,2017;Lietal., 2017).Ourresultsindicateheadorientationvariedsomewhatbetween scansandtherewasgreatervariationbetweensitesthanintra-site;we

(10)

alsoobservedaconsistentnegativecorrelationbetween𝜒 andhead ori-entation(𝜃).Usingalinearmodeltoattempttoregress-outtheeﬀects ofheadrotationimprovedthereproducibilityofbothwithin-siteand cross-sitedata. Italsoreduced thepenaltyformulti-site scanningvs single-sitescanning,butnotcompletely.

Inthisstudy, thereproducibility ofQSMusingsingle-echo, high-resolution(0.7 mm isotropicresolution; TE= 20ms)andmulti-echo standard-resolution(1.4 mm isotropicresolution; TE= 4,9, 14,19, 24,29,34and39ms)protocolswerecompared,andtheresultsshow that the multi-echo QSM data has a significantly higher variability thansingle-echoQSM.Althoughmulti-echophasedatahasbeen com-binedwithamagnitude-weightedleastsquaresregressionofphaseto echotime,itmaycarryinconsistentphaseaccumulationacrossechoes thatwereinconsistentlyunwrapped.Thisisalsoparticularlyrelevant forregionsoflargefieldinhomogeneities,wherephaseaccumulation could exceed±𝜋 between neighbouringvoxels, resulting in multiple phasewraps,whichtheunwrappingalgorithmmaybeunableto cor-rect(Croninetal.,2017).Thishasalsobeenverifiedontheanalysisof QSMdatafromthecorticalROIsreconstructedwithdifferentnumbersof echoes:longechotimesincreasesignificantlythetest-retestvariability. Alternatively,phaseunwrappingcanbecompletelyomittedby calculat-ingthephasechangeoverallecho-timesusingacomplexfittingroutine (Liuetal.,2012;Liuetal.,2013)(fit_ppm_complex.mofMEDItoolbox) andcalculatingtheLaplaciandirectlyfromtheresulting,stillwrapped, temporalphasechangedata(Schweseretal.,2013).

It has been shown that resolution influences QSM estimation. Haacke et al. (2015) showed on phantom data that by decreasing slice thicknessfrom 3mm to0.5mm partialvolume effects are re-duced,absolutesusceptibilityvaluesdecrease,andaccuracyimproves up to 25%. Similar findings on in vivo brain data are reported in Sunetal.(2017)(single-echodata)andKarsaetal.(2019)(multi-echo data).Ourresultssupportthesuggestionthatareductionofpartial vol-umeeffectsathigher-resolutionmightplayaroleindecreasingboth test–retestandcross-sitevariabilityonthesingle-echohigh-resolution datacomparedtothemulti-echolow-resolutiondata.

R2∗ values show signiﬁcantly lower variability, reﬂected in the higherICCwithinandacross-sitescomparedtocorrespondingvaluesfor

𝜒 insubcorticalareas.Thismaybebecausethe𝜒 estimationisglobally moresensitivetobackgroundfieldinhomogeneitycomparedto magni-tudedata.However,inorbitofrontalandlowertemporalregionslarge through-planefieldvariationsfromtissue-airinterfacesdominatethe fieldchangesandproducedropoutsinthesignalmagnitudeandincrease thebackgroundphase,affectingbothQSMandR2∗mapsbyincreasing variabilityanddecreasingICCacrosssites.Inaddition,becauseoflarge fieldvariations,theestimatedcorticalR2∗increasessignificantlywhen lateechotimes areusedforthefitting,butthiseffect isnotseen in subcorticalareas.

QSM can only determine relative susceptibility differences (Cheng et al., 2009) and most approaches to calculation of sus-ceptibilityfrommeasuredphaseyieldmapsinwhichtheaveragevalue ofsusceptibilityiszerooverthemaskedimagingvolume.Issuesrelated toreferencingofQSMdatahavebeeninvestigated(Fengetal.,2018; Straubetal.,2017),withaimoffindingareferenceregionortissueto whichallsusceptibilityvaluesarereferredthatproduceswell-defined andreproducible values of susceptibility. Here we investigated how thechoiceofreferenceaffectsthewithin-siteandcross-sitevariability ofmeasuredsusceptibilityatultra-high-field.Wetestedthreeaccepted reference regions: total whole brain signal, “wb”, whole brain CSF erodedinordertoexcludeanypialorskullsurfaces,“csf”,anda manu-allyselectedcylindricalROIintherightventricle,“cyl”.Wefoundthat the“cyl” referencinggenerallyincreasedthevariabilityofthecross-site andwithin-sitesusceptibilitymeasurementsincorticalandsubcortical ROIs compared to “wb” referencing. In the case of the multi-echo acquisitionthe“csf” referencingalsoincreasedthevariabilityrelative to“wb” data. This maybe becauseof imprecisionin systematically obtainingaverageQSMsignalfromCSFregions.Referencingusinga

small ROI inthe ventriclesmightbe prone tosubjectivitygiven the natural variationinventricle sizeinhealthysubjectsandindisease. Furthermore,theventriclesdonotcontainpureCSF:theyaretraversed bybloodvesselswithadiﬀerent𝜒 (Sullivanetal.,2002).Thismakes whole-brainreferencingattractive inmanysituations. Yet,inpatient cohorts wherethereis substantialironloadin subcorticalstructures (SnyderandConnor,2009),wholebrainreferencingmightnotbean appropriate approach. In this case, the more appropriate approach willbetochooseasmallreferenceregionwhichshowsnochangesin theparticulardiseasetobe“zero” susceptibility ata costofaslight increaseinSD.

Toeliminateoperator-dependentbiasinsegmentationwhen deter-miningbrainstructures,wehaveanalyseddatausingbothmanualand atlas-based segmentation.Fromourresults,manualROIsshowed sig-niﬁcantlylowervariabilitycomparedtoatlas-basedmethods.This hap-pensbecauseofimprecisioninregistrationbetweenMNIandsubject spaceaswellastheempiricalthresholdingthatwaschosentoobtain thesubcorticalROIs.Thisresultedin largerROIsbeingderivedfrom theatlas-basedmethodcomparedtothemanualmethod(Wilcoxontest, CN:p=0.014;Pu:p=0.00018;GP:p=0.0010).Overestimationofthe re-gion(Fig.5Supplementarymaterial2)meantincludingboundaryvoxels that,generally,havelowersusceptibility(white-matter,forexample), loweringtheaverage𝜒 andR2∗.However,traditionalmanualdrawing ofROIsforcohortstudiesisdiﬃcult,timeconsumingandpotentially unsuitableasitbiasesresultstowardsparticularcohorts(Collinsetal., 2003)soitmaynotalwaysbethemostappropriateapproach.

Inthisstudy,harmonizedprotocolswereproducedforallfive scan-nerswithoutanysignificantsequencealterations,asaproduct3D gra-dientecho(GE)sequencewasreadilyavailableonallsystems(the prod-uct‘gre’sequencefromSiemensandtheproduct‘ffe’fromPhilips).The protocolsandanexampledatasetareprovidedin(Clarke,2018). Gener-ally,wealsoreliedonthevendors’reconstruction.However,attheend ofthereconstructionpipelineoftheSiemenssystemsweadopteda dif-ferentcoilcombination approachbasedonRoemeretal.(1990)and Walsh et al.(2000), tomatch theSENSE approachimplemented on Philipsscanners(Pruessmannetal.,1999;Robinsonetal.,2017).This was requireddue toartifactsappearingon phaseimagesin Siemens datareconstructedwiththevendor’spipeline,suchasopen-endedfringe linesorsingularities(Chavezetal.,2002)(Fig.2,Supplementary mate-rial2).ThesereducetheconsistencyoftheQSMresults(Santinetal., 2017).However,othercoil combinationmethods suchasaselective channelcombinationapproach(Veghetal.,2016)ortheCOMPOSER (COMbiningPhasedatausingaShortEcho-timeReferencescan)method (Bollmannetal.,2018) havealsobeenshown toreduce open-ended fringelinesandnoiseinthesignalphase.Forfutureinvestigations,the rawk-spacedatacollectedfromallsitesinthisstudyhasbeenstored andisavailablefromtheauthorsuponrequest.

OntheQSMreconstruction,animperfect background field filter-ingcaninfluencethereproducibilityofQSMdata.Forthisreason,we performedbackgroundremovalintwostepsasimplementedin QSM-boxv2.0andasdescribedin(Acosta-Cabroneroetal.,2018):firstwith theLBVapproachandthenfollowedbythevSMVmethod. Regular-izedfield-to-susceptibilityinversionstrategieshavebeenproposed to overcometheill-posedprobleminQSMwithdataacquiredatasingle headorientation(deRochefortetal.,2010).WeoptedtousetheMSDI implementationinQSMboxv2.0(Acosta-Cabroneroetal.,2018),asit rankedtop-10inallmetricsofthe2016QSMReconstructionChallenge (Langkammeretal.,2018),andalsonowincludesanewself-optimized local scale,whichresultsin abetterpreservationofphasenoise tex-ture andlowsusceptibilitycontrast features.Onthesecondstep,the regularizationfactor,𝜆,usedforthisstudywassetto102.7_,_as recom-mendedbyAcosta-Cabroneroetal.(2018)basedonanL-curveanalysis (Hansenetal.,1993)withhigh-resolution7Tdata.

Thestandardmulti-echoGEprotocolinthisstudywasproducedas aharmonisedsequencethatcouldbeperformedatallsites,witha rel-ativelyshortacquisitiontime(approximately5min),whichis

(11)

accept-Fig.8. Illustrationofthefeasibilityofa7TQSMclinicalstudy.𝜒 (A)andR2∗(B)forfourROIs(SubstantiaNigra,SN;CaudateNucleus,CN;Putamen,Pu;Globus

Pallidus,GP)fromhealthyvolunteer(HV)andsynthetic“patient” (PT)dataforwhichAVlitandSDlitwereobtainedfromLangkammeretal.(2016)andSDbwere

calculatedfromdataofthecurrentstudy.AVlitvaluesforR2∗werelinearlyscaledto7TaccordingtoYaoetal.(2007).BluebarsshowtheAVlit±SDlitandgreen

barstheAVlit±SDb.StatisticaldiﬀerencesbetweenHVandPTobtainedfromLangkammeretal.(2016)arealsoshown.ForeachROI,thesamplesizethatwould

havebeenneededtogiveasigniﬁcanteﬀectwascalculatedfromthegroupmeans,AVlit,andtheSDbperROIandisshownincircles.Multi-echo𝜒-mapswere

calculatedwithdatafromalleightechoes.

ableforpatientstudies.Mid-brainstructuressuchasthebasalganglia areidentifiable,yetsmallsubcorticalstructureswillsufferfrom partial-volumeeffects,whichcouldbealimitationofthisharmonizedprotocol forfutureultra-highfieldmulti-sitestudies.

Atultra-highfieldtherecanbevariationsinSNRinmagnitudedata causedbythevariableB1+acrossthebrain(Abduljaliletal.,2003).As R2∗ isestimatedvoxel-wise,andasthereisalwaysareasonableSNR inthemagnitudedata,thecoefficientintheexponentialfitthat esti-matesR2∗willnotbestronglyaffectedbyvariationsinB1+.QSMmaps areestimatedfromfilteredphasedatawhichisnotstronglyaffectedby transmitB1+ variations.Onourdata,nocorrelationswerefound be-tweenQSMorR2∗mapsandB1+flip-anglemapscollectedinthesame session(Fig.6Supplementarymaterial2).

Tominimiseconfoundingeffectsofageorpathology,weassessed test-retestreliabilityandcross-sitevariabilitywithtenhealthyyoung subjects.Thecross-site,between-subjectstandard-deviation,SD_b, mea-suredinthisstudywasevaluatedtogetherwithhealthyandParkinson’s diseasedatafrom(Langkammeretal.,2016).Apoweranalysisrevealed asamplesizethatwouldhavebeenrequiredforamulti-site clinical studyineachROIasshowninFig.8.ForallthesignificantROIsthe numberofsubjectsthatwouldhavebeenrequiredpergroupwaslessor equalto44.Sincethisislowerthanthesamplesizewehaveusedinthis study(90healthyvolunteerscans)andthenumbersintheLangkammer study(66patientsand58controlsubjects),itgivesstrongconfidence offeasibilityforfuture7TQSMclinicalstudies.

5. Conclusion

We investigated test-retest reliabilityand reproducibility of T₂∗ -weightedimagingprotocolsatultra-highfieldMRI.Consideringthe in-creaseinsusceptibilityeffectsat7T,wefoundthatvariabilityof mea-surementsofQSM𝜒 andR2∗inthebasalgangliaarereducedcompared toreportsfromlowerfieldstrengths,1.5Tand3T.Scannerhardware differencesgivemoremodestimprovementsforcorticalmeasurements ofQSM𝜒 andR2∗.Multi-echoprotocolsdonotbenefitfromlongecho timesastheseincreasetheimprecisionintheestimationofQSM.We suggestthat7TMRIissuitableformulticentrequantitativeanalysesof brainiron,inhealthanddisease.

Centrefunding

TheWellcomeCentreforIntegrativeNeuroimagingissupportedby corefundingfromtheWellcomeTrust(203139/Z/16/Z).

Cardiﬀ UniversityBrain ResearchImagingCentreissupportedby theUKMedicalResearchCouncil(MR/M008932/1)andtheWellcome Trust(WT104943).

Thisresearchwasco-fundedbytheNIHRCambridgeBiomedical Re-searchCentre.Theviewsexpressedarethoseoftheauthor(s)andnot necessarilythoseoftheNHS,theNIHRortheDepartmentofHealthand SocialCare.TheCambridge7TMRIfacilityisco-fundedbythe Univer-sityofCambridgeandtheMedicalResearchCouncil(MR/M008983/1).

(12)

Individualfunding

CTRisfundedbyaSirHenryDaleFellowshipfromtheWellcome TrustandtheRoyalSociety[098436/Z/12/B].JBRissupportedbythe WellcomeTrust(WT103838).

Creditauthorstatement

CatarinaRua:Conceptualization,Methodology,Formalanalysis, In-vestigation,Writing– OriginalDraft,Writing– Review&Editing

WilliamT Clarke: Conceptualization, Methodology, Investigation, Writing– Review&Editing

IanDDriver:Conceptualization,Methodology,Investigation, Writ-ing– Review&Editing

Olivier Mougin: Conceptualization, Methodology, Investigation, Writing– Review&Editing

AndrewT.Morgan:Conceptualization,Methodology,Investigation. StuartClare:Conceptualization, Methodology,Project administra-tion,Fundingacquisition.

SusanFrancis:Conceptualization,Methodology,Project administra-tion,Fundingacquisition.

KeithMuir:Conceptualization,Methodology,Projectadministration, Fundingacquisition.

RichardWise:Conceptualization,Methodology,Project administra-tion,Fundingacquisition.

AdrianCarpenter:Conceptualization,Methodology,Project admin-istration,Fundingacquisition.

GuyWilliams:Conceptualization,Methodology,Project administra-tion,Fundingacquisition.

JamesBRowe:Conceptualization,Methodology,Project administra-tion,Writing– Review&Editing,Fundingacquisition.

RichardBowtell:Conceptualization,Methodology,Project adminis-tration,Writing– Review&Editing,Fundingacquisition.

ChristopherTRodgers:Conceptualization,Methodology,Project ad-ministration,Investigation,Writing– Review&Editing,Funding acqui-sition.

Acknowledgements

TheUK7TNetworkandthisworkwasfundedbytheUK’sMedical ResearchCouncil(MRC)[MR/N008537/1].WethankDr.Julio Acosta-CabroneroformakingtheQSMboxpubliclyavailableandProf.David PorterforhissupportintheUK7Tnetwork.

Supplementarymaterials

Supplementarymaterialassociatedwiththisarticlecanbefound,in theonlineversion,atdoi:10.1016/j.neuroimage.2020.117358. References

Abdul-Rahman, H., Gdeisat, M., Burton, D., Michael, L., 2005. Fast three-dimensional phase-unwrapping algorithm based on sorting by reliability following a non- continuous path. App. Optic. 46, 6623–6635. doi: 10.1117/12.611415 .

Abduljalil, A.M., Schmalbrock, P., Novak, V., Chakeres, D.W., 2003. Enhanced gray and white matter contrast of phase susceptibility ‐weighted images in ultra ‐high ‐ﬁeld magnetic resonance imaging. J. Magn. Resonanc. Imaging: Oﬀ. J. Int. Soc. Magn. Reso- nanc. Med. 18 (3), 284–290. doi: 10.1002/jmri.10362 .

Acosta-Cabronero, J., Milovic, C., Mattern, H., Tejos, C., Speck, O., Callaghan, M.F., 2018. A robust multi-scale approach to quantitative susceptibility mapping. NeuroImage 183, 7–24. doi: 10.1016/j.neuroimage.2018.07.065 .

Acosta-Cabronero, J., Betts, M.J., Cardenas-Blanco, A., Yang, S., Nestor, P.J., 2016. In Vivo MRI mapping of brain iron deposition across the adult lifespan. J. Neurosci. 36, 364–374. doi: 10.1523/JNEUROSCI.1907-15.2016 .

Acosta-Cabronero, J., Williams, G.B., Cardenas-Blanco, A., Arnold, R.J., Lupson, V., Nestor, P.J., 2013. In vivo quantitative susceptibility mapping (QSM) in Alzheimer’s disease. PLoS One 8, e81093. doi: 10.1371/journal.pone.0081093 .

Barbosa, J.H.O., Santos, A.C., Salmon, C.E.G., 2015. Susceptibility weighted imaging: dif- ferentiating between calciﬁcation and hemosiderin. Radiol. Brasil. 48 (2), 93–100. doi: 10.1590/0100-3984.2014.0010 .

Betts, M.J., Acosta-Cabronero, J., Cardenas-Blanco, A., Nestor, P.J., Düzel, E., 2016. High- resolution characterisation of the aging brain using simultaneous quantitative susceptibility mapping (QSM) and R2 ∗ measurements at 7 T. Neuroimage 138, 43–63.

doi: 10.1016/j.neuroimage.2016.05.024 .

Bilgic, B., Pfeﬀerbaum, A., Rohlﬁng, T., Sullivan, E.V., Adalsteinsson, E., 2012. MRI es- timates of brain iron concentration in normal aging using quantitative susceptibility mapping. Neuroimage 59 (3), 2625–2635. doi: 10.1016/j.neuroimage.2011.08.077 . Blazejewska, A.I., Al-Radaideh, A.M., Wharton, S., Lim, S.Y., Bowtell, R.W., Constanti-

nescu, C.S., Gowland, P.A., 2015. Increase in the iron content of the substantia nigra and red nucleus in multiple sclerosis and clinically isolated syndrome: a 7 Tesla MRI study. J. Magn. Reson. Imaging 41, 1065–1070. doi: 10.1002/jmri.24644 . Bollmann, S., Robinson, S.D., O’Brien, K., Vegh, V., Janke, A., Marstaller, L., Reutens, D.,

Barth, M., 2018. The challenge of bias ‐free coil combination for quantitative susceptibility mapping at ultra ‐high ﬁeld. Magn. Reson. Med. 79 (1), 97–107. doi: 10.1002/mrm.26644 .

Chavez, S., Xiang, Q., Li, A., 2002. Understanding phase maps in MRI: a new cutline phase unwrapping method. IEEE Trans. Med. Imaging 21, 966–977. doi: 10.1109/TMI.2002.803106 .

Cheng, Y.-C.N., Neelavalli, J., Haacke, E.M., 2009. Limitations of calculating ﬁeld distributions and magnetic susceptibilities in MRI using a Fourier based method. Phys. Med. Biol. 54 (5), 1169–1189. doi: 10.1088/0031-9155/54/5/005 .

Choi, S., Li, X., Harrison, D.M., 2019. The impact of coregitration of gradient re- called echo images on quantitative susceptibility and R2 ∗ mapping at 7T. bioRxiv.

doi: 10.1101/529891 .

Clarke, W.T., 2018. UK7T Network harmonized neuroimaging protocols. https://ora.ox.ac.uk/objects/uuid:55ca977f-62df-4cbf-b300-2dc893e36647 . Clarke, W.T., Mougin, O., Driver, I.D., Rua, C., Morgan, A., Asghar, M., Clare, S., Fran-

cis, S., Wise, R., Rodgers, C.T., Carpenter, T.A., Muir, K., Bowtell, R., 2019. Multi-site harmonization of 7 Tesla MRI neuroimaging protocols. NeuroImage 206, 116335. doi: 10.1016/j.neuroimage.2019.116335 .

Cobzas, D., Sun, H.F., Walsh, A.J., Lebel, R.M., Blevins, G., Wilman, A.H., 2015. Subcorti- cal gray matter segmentation and voxel-based analysis using transverse relaxation and quantitative susceptibility mapping with application to multiple sclerosis. J. Magn. Reson. Imaging 42 (6), 1601–1610. doi: 10.1002/jmri.24951 .

Collins, D.L. , Zijdenbos, A.P. , Paus, T. , Evans, A.C. , 2003. Use of registration for cohort studies. Med. Image Regist. .

Cronin, M.J. , Wang, N. , Decker, K.S. , Wei, H. , Zhu, W.Z. , Liu, C. , 2017. Exploring the origins of echo-time-dependent quantitative susceptibility mapping (QSM) measurements in healthy tissue and cerebral microbleeds. Neuroimage 149, 98–113 . de Rochefort, L., Liu, T., Kressler, B., Liu, J., Spincemaille, P., Lebon, V., Wu, J., Wang, Y.,

2010. Quantitative susceptibility map reconstruction from MR phase data using Bayesian regularization: validation and application to brain imaging. Magn. Reson. Med. 63, 194–206. doi: 10.1002/mrm.22187 .

Deh, K., Nguyen, T.D., Eskreis-Winkler, S., Prince, M.R., Spincemaille, P., Gauthier, S., Kovanlikaya, I., Zhang, Y., Wang, Y., 2015. Reproducibility of quantitative susceptibility mapping in the brain at two ﬁeld strengths from two vendors. J. Magn. Reson. Imaging 42, 1592–1600. doi: 10.1002/jmri.24943 .

Deistung, A., Schäfer, A., Schweser, F., Biedermann, U., Güllmar, D., Trampel, R., Turner, R., Reichenbach, J.R., 2013. High-resolution MR imaging of the human brainstem in vivo at 7 Tesla. Front. Hum. Neurosci. 7, 710. doi: 10.3389/fn- hum.2013.00710 .

Düzel, E., Acosta ‐Cabronero, J., Berron, D., Biessels, G.J., Björkman ‐Burtscher, I., Bottlaen- der, M., Bowtell, R., Buchem, M.V., Cardenas ‐Blanco, A., Boumezbeur, F., Chan, D., 2019. European Ultrahigh ‐Field Imaging Network for Neurodegenerative Diseases (EUFIND). Alzheimer’s Dementia: Diagn. Assessm. Dis. Monit. 11 (1), 538–549. doi: 10.1016/j.dadm.2019.04.010 .

Duyn, J.H., van Gelderen, P., Li, T.Q., de Zwart, J.A., Koretsky, A.P., Fukunaga, M., 2007. High-ﬁeld MRI of brain cortical substructure based on signal phase. Proc. Natl. Acad. Sci. 104 (28), 11796–11801. doi: 10.1073/pnas.0610821104 .

Edelstein, W.A., Glover, G.H., Hardy, C.J., Redington, R.W., 1986. The intrinsic signal ‐to ‐noise ratio in NMR imaging. Magn. Resonanc. Med. 3 (4), 604–618. doi: 10.1002/mrm.1910030413 .

Ehses, P., Brenner, D., Stirnberg, R., Pracht, E.D., Stöcker, T., 2019. Whole ‐brain B1 ‐mapping using three ‐dimensional DREAM. Magn. Resonanc. Med. 82 (3), 924– 934. doi: 10.1002/mrm.27773 .

Eskreis ‐Winkler, S., Zhang, Y., Zhang, J., Liu, Z., Dimov, A., Gupta, A., Wang, Y., 2017. The clinical utility of QSM: disease diagnosis, medical management, and surgical planning. NMR Biomed. 30 (4), e3668. doi: 10.1002/nbm.3668 .

Feng, X., Deistung, A., Reichenbach, J.R., 2018. Quantitative susceptibility mapping (QSM) and R2 ∗ in the human brain at 3 T: Evaluation of intra-scanner repeatabil-

ity. Z. Med. Phys. 28, 36–48. doi: 10.1016/j.zemedi.2017.05.003 .

Gelman, N., Gorell, J.M., Barker, P.B., Savage, R.M., Spickler, E.M., Windham, J.P., Knight, R.A., 1999. MR imaging of human brain at 3.0 T: preliminary report on transverse re-laxation rates and relation to estimated iron content. Radiology 210, 759– 767. doi: 10.1148/radiology.210.3.r99fe41759 .

Haacke, E.M., Cheng, N., House, M.J., Liu, Q., Neelavalli, J., Ogg, R.J., Khan, A., Ayaz, M., Kirsch, W., Obenaus, A., 2005. Imaging iron stores in the brain using magnetic resonance imaging. Magn. Resonanc. Imaging 23, 1–25. doi: 10.1016/j.mri.2004.10.001 . Haacke, E.M., Liu, S., Buch, S., Zheng, W., Wu, D., Ye, Y., 2015. Quantitative susceptibility mapping: current status and future directions. Magn. Resonanc. Imaging 33 (1), 1–25. doi: 10.1016/j.mri.2014.09.004 .

Hansen, P.C., O’Leary, D.P., 1993. The use of the l-curve in the regularization of discrete ill-posed problems. SIAM J Sci. Comput. 14 (6), 1487–1503. doi: 10.1137/0914086 . He, X., Yablonskiy, D.A., 2009. Biophysical mechanisms of phase contrast in gradient echo

MRI. Proc. Natl. Acad. Sci. USA 106, 13558–13563. doi: 10.1073/pnas.0904899106 . Hinoda, T., Fushimi, Y., Okada, T., Fujimoto, K., Liu, C., Yamamoto, A., Okada, T.,

(13)

Kido, A., Togashi, K., 2015. Quantitative susceptibility mapping at 3 T and 1.5 T: evaluation of consistency and reproducibility. Investig. Radiol. 50, 522–530. doi: 10.1097/RLI.0000000000000159 .

House, M.J., Pierre, T.G.S., Kowdley, K.V., Montine, T., Connor, J., Beard, J., Berger, J., Siddaiah, N., Shankland, E., Jin, L.W., 2007. Correlation of proton transverse relaxation rates (R2) with iron concentrations in postmortem brain tissue from Alzheimer’s disease patients. Magn. Resonanc. Med. 57, 172–180. doi: 10.1002/mrm.21118 . Karsa, A., Punwani, S., Shmueli, K., 2019. The eﬀect of low resolution and coverage on

the accuracy of susceptibility mapping. Magn. Resonanc. Med. 81 (3), 1833–1848. doi: 10.1002/mrm.27542 .

KeuKen, M.C., Bazin, P.L., Backhouse, K., Beekhuizen, S., Himmer, L., Kandola, A., Lafeber, J.J., Prochazkova, L., Trutti, A., Schäfer, A., Turner, R., 2017. Eﬀects of aging on T1, T2 ∗ , and QSM MRI values in the subcortex. Brain Struct. Funct. 222 (6), 2487–2505. doi: 10.1007/s00429-016-1352-4 .

Lancione, M., Tosetti, M., Donatelli, G., Cosottini, M., Costagli, M., 2017. The impact of white matter ﬁber orientation in single-acquisition quantitative susceptibility mapping. NMR Biomed. 30, e3798. doi: 10.1002/nbm.3798 .

Langkammer, C., Schweser, F., Shmueli, K., Kames, C., Li, X., Guo, L., Milovic, C., Kim, J., Wei, H., Bredies, K., Buch, S., 2018. Quantitative susceptibility mapping: report from the 2016 reconstruction challenge. Magn. Reson. Med. 79 (3), 1661–1673. doi: 10.1002/mrm.26830 .

Langkammer, C., Pirpamer, L., Seiler, S., Deistung, A., Schweser, F., Franthal, S., Homayoon, N., Katschnig-Winter, P., Koegl-Wallner, M., Pendl, T., Stoegerer, E.M., 2016. Quantitative susceptibility mapping in Parkinson’s disease. PLoS One 11 (9), e0162460. doi: 10.1371/journal.pone.0162460 .

Langkammer, C., Schweser, F., Krebs, N., Deistung, A., Goessler, W., Scheurer, E., Sommer, K., Reishofer, G., Yen, K., Fazekas, F., Ropele, S., Reichenbach, J.R., 2012. Quantitative susceptibility mapping (QSM) as a means to measure brain iron? A post mortem validation study. Neuroimage 62 (3), 1593–1599. doi: 10.1016/j.neuroimage.2012.05.049 .

Langkammer, C., Krebs, N., Goessler, W., Scheurer, E., Ebner, F., Yen, K., Fazekas, F., Ropele, S., 2010. Quantitative MR imaging of brain iron: a postmortem validation study. Radiology 257 (2), 455–462. doi: 10.1148/radiol.10100495 .

Lee, J., Shmueli, K., Kang, B.T., Yao, B., Fukunaga, M., van Gelderen, P., Palumbo, S., Bosetti, F., Silva, A.C., Duyn, J.H., 2012. The contribution of myelin to magnetic susceptibility-weighted contrasts in high-ﬁeld MRI of the brain. Neuroimage 59, 3967–3975. doi: 10.1016/j.neuroimage.2011.10.076 .

Li, G., Zhai, G., Zhao, X., An, H., Spincemaille, P., Gillen, K.M., Ku, Y., Wang, Y., Huang, D., Li, J., 2019. 3D texture analysis within substantia nigra of Parkinson’s disease patients on quantitative susceptibility maps and R2 ∗ maps. NeuroImage 188, 465–472.

doi: 10.1016/j.neuroimage.2018.12.041 .

Li, L., Leigh, J.S., 2004. Quantifying arbitrary magnetic susceptibility distributions with MR. Magn. Reson. Med. 51, 1077–1082. doi: 10.1002/mrm.20054 .

Li, W., Liu, C., Duong, T.Q., van Zijl, P.C., Li, X., 2017. Susceptibility tensor imaging (STI) of the brain. NMR Biomed. 30 (4), e3540. doi: 10.1002/nbm.3540 .

Lin, P.Y., Chao, T.C., Wu, M.L., 2015. Quantitative susceptibility mapping of human brain at 3T: a multisite reproducibility study. AJNR Am J. Neuroradiol. 36, 467–474. doi: 10.3174/ajnr.A4137 .

Liu, T., Wisnieﬀ, C., Lou, M., Chen, W., Spincemaille, P., Wang, Y., 2013. Nonlinear formu- lation of the magnetic ﬁeld to source relationship for robust quantitative susceptibility mapping. Magn. Resonanc. Med. 69 (2), 467–476. doi: 10.1002/mrm.24272 . Liu, J., Liu, T., de Rochefort, L., Ledoux, J., Khalidov, I., Chen, W., Tsiouris, A.J., Wisni-

eﬀ, C., Spincemaille, P., Prince, M.R., Wang, Y., 2012. Morphology enabled dipole inversion for quantitative susceptibility mapping using structural consistency between the magnitude image and the susceptibility map. Neuroimage 59 (3), 2560–2568. doi: 10.1016/j.neuroimage.2011.08.082 .

Lotﬁpour, A.K., Wharton, S., Schwarz, S.T., Gontu, V., Schaefer, A., Peters, A.M., Bowtell, R.W., Auer, D.P., Gowland, P.A., Bajaj, P.S., 2012. High resolution magnetic susceptibility mapping of the substantia nigra in Parkinson’s disease. J. Magn. Reso- nanc. Imaging 35, 48–55. doi: 10.1002/jmri.22752 .

Makhlouf, S.A., Parker, F.T., Berkowitz, A.E., 1997. Magnetic hysteresis anomalies in fer- ritin. Phys. Rev. B 55 (22). doi: 10.1103/PhysRevB.55.R14717 , R14 717-R14 720. Moeller, H.E., Bossoni, L., Connor, J.R., Crichton, R.R., Does, M.D., Ward, R.J., Zecca, L.,

Zucca, F.A., Ronen, I., 2019. Iron, myelin, and the brain: Neuroimaging meets neuro- biology. Trends Neurosci. 42 (6), 384–401. doi: 10.1016/j.tins.2019.03.009 . Mougin, O. , Clarke, W. , Driver, I. , Rua, C. , Morgan, A.T. , Francis, S. , Muir, K. , Carpen-

ter, A. , Rodgers, C. , Wise, R. , Porter, D. , Clare, S. , Bowtell, R. , 2019. Robustness of PSIR segmentation and R1 mapping at 7T: a travelling head study. Proc. Intr. Soc. Mag. Reson. Med. 27, 237 .

Nehrke, K., Bornert, P., 2012. DREAM–a novel approach for robust, ultrafast, multislice B(1) mapping. Magn. Reson. Med. 68 (5), 1517–1526. doi: 10.1002/mrm.24158 . Pei, M., Nguyen, T.D., Thimmappa, N.D., Salustri, C., Dong, F., Cooper, M.A., Li, J.,

Prince, M.R., Wang, Y., 2015. Algorithm for fast monoexponential ﬁtting based on auto-regression on linear operations (ARLO) of data. Magn. Reson. Med. 73, 843– 850. doi: 10.1002/mrm.25137 .

Pohmann, R., Speck, O., Scheﬄer, K., 2016. Signal ‐to ‐noise ratio and MR tissue parameters in human brain imaging at 3, 7, and 9.4 tesla using current receive coil arrays. Magn. Resonanc. Med. 75 (2), 801–809. doi: 10.1002/mrm.25677 .

Pruessmann, K.P., Weiger, M., Scheidegger, M.B., Boesiger, P., 1999. SENSE: sensitivity encoding for fast MRI. Magn. Resonanc. Med. 42 (5), 952–962. doi: 10.1002/(SICI)1522-2594(199911)42:5 < 952::AID-MRM16>3.0.CO;2-S .

R Core team, 2013. R: A Language and Environment for Statistical Computing. R Founda- tion for Statistical Computing, Vienna, Austria. http://www.R-project.org/ . Reichenbach, J.R., 2012. The future of susceptibility contrast for assessment of anatomy

and function. Neuroimage 62, 1311–1315. doi: 10.1016/j.neuroimage.2012.01.004 . Reichenbach, J.R., Jonetz-Mentzel, L., Fitzek, C., Haacke, E.M., Kido, D.K.,

Lee, B.C., Kaiser, W.A., 2001. High-resolution blood oxygen-level dependent MR venography (HRBV): a new technique. Neuroradiology 43, 364–369. doi: 10.1007/s002340000503 .

Robinson, S.D., Bredies, K., Khabiova, D., Dymerska, B., Marques, J.P., Schweser, F., 2017. An illustrated comparison of processing methods for MR phase imaging and QSM: combining array coil signals and phase unwrapping. NMR Biomed. 30, e3601. doi: 10.1002/nbm.3601 .

Roemer, P.B., Edelstein, W.A., Hayes, C.E., Souza, S.P., Mueller, O.M., 1990. The NMR phased array. Magn. Reson. Med. 16 (2), 192–225. doi: 10.1002/mrm.1910160203 . Santin, M.D., Didier, M., Valabregue, R., Yahia Cherif, L., García ‐Lorenzo, D., Loureiro

de Sousa, P., Bardinet, E., Lehéricy, S., 2017. Reproducibility of R2 ∗ and quantitative

susceptibility mapping (QSM) reconstruction methods in the basal ganglia of healthy subjects. NMR Biomed. 30 (4), e3491. doi: 10.1002/nbm.3491 .

Schweser, F., Deistung, A., Lehr, B.W., Reichenbach, J.R., 2011. Quantitative imaging of intrinsic magnetic tissue properties using MRI signal phase: an approach to in vivo brain iron metabolism? Neuroimage 54 (4), 2789–2807. doi: 10.1016/j.neuroimage.2010.10.070 .

Schweser, F., Deistung, A., Sommer, K., Reichenbach, J.R., 2013. Toward online reconstruction of quantitative susceptibility maps: superfast dipole inversion. Magn. Reso- nanc. Med. 69 (6), 1582–1594. doi: 10.1002/mrm.24405 .

Smith, S.M., 2002. Fast robust automated brain extraction. Hum. Brain Mapp. 17 (3), 143–155. doi: 10.1002/hbm.10062 .

Snyder, A.M., Connor, J.R., 2009. Iron, the substantia nigra and related neurological dis- orders. Biochim. Biophys. Acta 1790, 606–614. doi: 10.1016/j.bbagen.2008.08.005 . Spincemaille, P., Liu, Z., Zhang, S., Kovanlikaya, I., Ippoliti, M., Makowski, M., Watts, R.,

de Rochefort, L., Venkatraman, V., Desmond, P., Santin, M.D., 2019. Clinical in- tegration of automated processing for brain quantitative susceptibility mapping: multi ‐site reproducibility and single ‐site robustness. J. Neuroimaging 29 (6), 689– 698. doi: 10.1111/jon.12658 .

Straub, S., Schneider, T.M., Emmerich, J., Freitag, M.T., Ziener, C.H., Schlemmer, H.P., Ladd, M.E., Laun, F.B., 2017. Suitable reference tissues for quantitative susceptibility mapping of the brain. Magn. Reson. Med. 78, 204–214. doi: 10.1002/mrm.26369 . Sullivan, E.V., Pfeﬀerbaum, A., Adalsteinsson, E., Swan, G.E., Carmelli, D., 2002. Diﬀer-

ential rates of regional brain change in callosal and ventricular size: a 4-year longi- tudinal MRI study of elderly men. Cereb. Cortex 12 (4), 438–445. doi: 10.1093/cer- cor/12.4.438 .

Sun, H., Seres, P., Wilman, A.H., 2017. Structural and functional quantitative susceptibility mapping from standard fMRI studies. NMR Biomed. 30 (4), e3619. doi: 10.1002/nbm.3619 .

Tie-Qiang, T., Gelderen, P., Merkle, H., Talagala, L., Koretsky, A.P., Duyn, J., 2006. Extensive heterogeneity in white matter intensity in high-resolution T2 ∗ -

weighted MRI of the human brain at 7.0 T. NeuroImage 32, 1032–1040. doi: 10.1016/j.neuroimage.2006.05.053 .

Vegh, V., O’Brien, K., Barth, M., Reutens, D.C., 2016. Selective channel combination of MRI signal phase. Magn. Reson. Med. 76 (5), 1469–1477. doi: 10.1002/mrm.26057 . Walsh, D.O., Gmitro, A.F., Marcellin, M.W., 2000. Adaptive reconstruc-

tion of phased array MR imagery. Magn. Reson. Med. 43 (5), 682–690. doi: 10.1002/(SICI)1522-2594(200005)43:5 < 682::AID-MRM10>3.0.CO;2-G . Wang, Y., Liu, T., 2015. Quantitative susceptibility mapping (QSM): decoding MRI data for

a tissue magnetic biomarker. Magn. Reson. Med. 7, 82–101. doi: 10.1002/mrm.25358 . Wardlaw, J.M., Brindle, W., Casado, A.M., Shuler, K., Henderson, M., Thomas, B., Macfarlane, J., Maniega, S.M., Lymer, K., Morris, Z., Pernet, C., 2012. A system- atic review of the utility of 1.5 versus 3 Tesla magnetic resonance brain imaging in clinical practice and research. European radiology 22 (11), 2295–2303. doi: 10.1007/s00330-012-2500-8 .

Weir, J.P. , 2005. Quantifying test-retest reliability using the intraclass correlation coeﬃ- cient and the SEM. J. Strength Cond. Res. 19 (1), 231–240 .

Wharton, S., Bowtell, R., 2010. Whole-brain susceptibility mapping at high ﬁeld: a comparison of multiple-and single-orientation methods. Neuroimage 53 (2), 515–525. doi: 10.1016/j.neuroimage.2010.06.070 .

Yacoub, E., Shmuel, A., Pfeuﬀer, J., Van de Moortele, P.F., Adriany, G., Andersen, P., Vaughan, J.T., Merkle, H., Ugurbil, K., Hu, X., 2001. Imaging brain function in humans at 7 Tesla. Magn. Reson. Med. 45, 588–594. doi: 10.1002/mrm.1080 .

Yang, X., Sammet, S., Schmalbrock, P., Knopp, M.V., 2010. Postprocessing correction for distortions in T2 ∗ decay caused by quadratic cross-slice b0 inhomogeneity. Magn.

Reson. Med. 63 (1), 1258–1268. doi: 10.1002/mrm.22316 .

Yao, B., Li, T., van Gelderen, P., Shmueli, K., De Zwart, J.A., Duyn, J.H., 2009. Neuro image susceptibility contrast in high ﬁeld MRI of human brain as a function of tissue iron content. Neuroimage 44 (4), 1259–1266. doi: 10.1016/j.neuroimage.2008.10.029 . Yao, B. , van Gelderen, P. , de Zwart, J.A. , Duyn, J.H. , 2007. Brain iron in MR imaging:

R2 ∗ and phase shift at diﬀerent ﬁeld strengths. Proc. Intr. Soc. Mag. Reson. Med. 15, 2165 .

Zhou, D., Liu, T., Spincemaille, P., Wang, Y., 2014. Background ﬁeld removal by solving the Laplacian boundary value problem. NMR Biomed 27 (3), 312–319. doi: 10.1002/nbm.3064 .