Research
Paper
Imaging
Registration
of
magnetic
resonance
and
computed
tomography
images
in
patients
with
oral
squamous
cell
carcinoma
for
three-dimensional
virtual
planning
of
mandibular
resection
and
reconstruction
M.Polfliet,M.S.Hendriks,J.-M.Guyader,I.tenHove,H.Mast,J.
Vandemeulebroucke,A.vanderLugt,E.B.Wolvius,S.Klein:Registrationof magneticresonanceandcomputedtomographyimagesinpatientswithoralsquamous cellcarcinomaforthree-dimensional virtualplanningofmandibularresectionand reconstruction. Int.J.OralMaxillofac.Surg.2019;xxx:xxx–xxx.ã2021TheAuthor (s). PublishedbyElsevierInc.onbehalfofInternationalAssociationofOraland MaxillofacialSurgeons.ThisisanopenaccessarticleundertheCCBYlicense(http:// creativecommons.org/licenses/by/4.0/).
M.Polfliet1,2,3,M.S.Hendriks4,
J.-M.Guyader3,5,I.tenHove4,
H.Mast4,J.Vandemeulebroucke1,2,
A.vanderLugt6,E.B.Wolvius4,
S.Klein3
1DepartmentofElectronicsandInformatics
(ETRO),VrijeUniversiteitBrussel(VUB), Brussels,Belgium;2imec,Leuven,Belgium;
3BiomedicalImagingGroupRotterdam,
DepartmentsofMedicalInformaticsand Radiology,ErasmusUniversityMedical Center,Rotterdam,TheNetherlands;
4DepartmentofOralandMaxillofacial
Surgery,ErasmusUniversityMedicalCenter, Rotterdam,TheNetherlands;5LabISEN–
Yncre´aOuest,Brest,France;6Departmentof
RadiologyandNuclearMedicine,Erasmus UniversityMedicalCenter,Rotterdam,The Netherlands
Abstract. Theaimofthisstudywastoevaluateandpresentanautomatedmethodfor registrationofmagneticresonanceimaging(MRI)andcomputedtomography(CT) orconebeamCT(CBCT)imagesofthemandibularregionforpatientswithoral squamouscellcarcinoma(OSCC).RegisteredMRIand(CB)CTcouldfacilitatethe three-dimensionalvirtualplanningofsurgicalguidesemployedforresectionand reconstructioninpatientswithOSCCwithmandibularinvasion.MRIand(CB)CT imageswerecollectedretrospectivelyfrom19patients.MRIimageswerealigned with(CB)CTimagesemployingarigidregistrationapproach(stage1),arigid registrationapproachusingamandibularmask(stage2), andtwonon-rigid Int.J.OralMaxillofac.Surg.2019;xxx:xxx–xxx
https://doi.org/10.1016/j.ijom.2021.01.003,availableonlineathttps://www.sciencedirect.com
registrationapproaches(stage3).Registrationaccuracywasquantifiedbythemean targetregistrationerror(mTRE),calculatedoverasetoflandmarksannotatedby twoobservers.Stage2achievedthebestregistrationresult,withanmTREof2.5 0.7mm,whichwascomparable tothe inter-andintra-observervariabilitiesof landmarkplacementinMRI.Stage2wassignificantlybetteralignedcomparedto all approachesinstage3.Inconclusion,thisstudydemonstratedthatrigid registrationwiththeuseofamaskisanappropriateimageregistrationmethodfor aligningMRIand(CB)CTimagesofthemandibularregioninpatientswithOSCC.
Key words: oral squamous cell carcinoma; mandibular reconstruction; X-ray computed tomography; magnetic resonance imaging; multimodalimaging;medicalimageprocessing. Acceptedforpublication
Oralsquamouscellcarcinoma(OSCC)is
thesixthmostcommoncancerworldwide.
Theoverall5-yearsurvivalrateofpatients
withOSCCislessthan50%andhasnot
shown anysignificantimprovementover
thelastdecades,despiteadvancesin
treat-mentmodalities1–5.
Segmentalmandibularresectionand
re-constructionwithafreevascularized
osseo-cutaneous flap is currently the
recommendedtreatmentforOSCC
invad-ing the mandible6.Themaingoal of surgical
treatmentistoobtaintumour-freeresection
margins with acceptable remaining
func-tion(chewing,swallowing,andspeaking)
and physical appearance. Achieving
tu-mour-freeresectionmarginsischallenging,
but crucial for disease control and
survival1,3,7–12.Recentstudiesreportedthat
inadequateresectionmarginswerefoundin
20% oftheboneresections,which
nega-tivelyimpactedthe5-yearsurvivalrateof
patients1,3,7. The inability to distinguish
tumourfrom healthybone tissue
intraopera-tivelyduringresectionisthemostcommon
causeofsuchinadequatemargins10.
Addi-tionally,are-resectionofpositivemargins
inasecondoperationisnotdesirabledueto
technical difficulties, and has a negative
effect onthesurvivalofthetransplant11.
Duringresection, tumour-free surgical
mar-ginsaretheonlyprognosticfactorthatthe
surgeoncancontrol11.
Thestate-of-the-art mandibular
recon-structionmethodisbasedonpreoperative
three-dimensional (3D) virtual surgical
planning using 3D-printed surgical
guides12–14.Withthismethod,thepatient
undergoes the necessary imaging, after
which the surgeon virtually defines the
cutting planes and plans the resection
and subsequent reconstruction.
Thereaf-ter, the surgical guides are printed and
the virtual planning is translated to the
surgical procedure. Cutting guides for
themandibleandfibulahavebeenshown
toprovide abetterfitofthefibulaparts,
resultinginareducedsurgicaltime.
How-ever, accurate3Dvirtualplanningofthe
surgical cuttingguides remains essential
inordertoachieveacompleteresection.
Current3Dvirtualplanningisbasedon
computedtomography(CT)orconebeam
CT(CBCT)images,which offerdetailed
informationonbonegeometryandcortical
bonedestruction,butdonotprovide
accu-rateinformationonbonemarrow
involve-mentand perineural spreadof the tumour. In
recentyears,magneticresonanceimaging
(MRI)hasincreasinglybeenusedfor
diag-nosticpurposesduetoitsbetter
visualiza-tion of tumour tissue, mandibular bone
marrowinvolvement,andperineuralspread
alongtheinferioralveolarnerve3,15–17.
The uncertainty about the location of
the tumour boundaries inthe 3Dvirtual
planning based on (CB)CT acquisitions
could be eliminated by including MRI
aquisitions18. Overlaying orfusingthese
MRIand(CB)CTimagesbeforethe
pre-operative virtual planning could aid the
surgeonindefiningthesurgicalguidesand
subsequentreconstructionthroughamore
accurate determination ofthe osteotomy
locationandabetterunderstandingofthe
surroundingstructures.Theintegrationof
fusedCTandMRIimagesformandibular
resectionplanninginclinicalpracticehas
been shown to be a safe and accurate
alternative19.However,duetothe
differ-entorientationandpositionofthe
mandi-bleintheMRIand(CB)CTacquisitions,
an imageregistrationmethodisrequired
to establish the spatial correspondences
betweenthedifferentimages.
Theaimofthisstudywas toevaluate
andpresentanautomatedmethodto
per-formimageregistrationofMRIand(CB)
CT in the mandibular region inpatients
withOSCC,whichcouldsubsequentlybe
integrated into a pipelinefor the virtual
planning of mandibular resections and
reconstructions.
Materialsandmethods
Dataset
Thestudywasreviewedandapprovedby
thelocalmedicalethicsreviewcommittee
(MEC-2016-143) and was performed in
accordancewithnationalandinternational
legislation.Theneedforinformedconsent
waswaivedowingtotheretrospectiveand
anonymizednatureofthestudy.
Preoper-ative3D MRIand (CB)CT scans ofthe
headandneckregion werecollected
ret-rospectively from 19 patients diagnosed
between 2014 and 2016 with untreated
primaryOSCCwithinvasionofthe
man-dible.Theimageswereanonymizedprior
to processing. The MRI scans were
ac-quired with a Spin Echo T1-weighted
sequence. The in-plane voxel size of
MRI was between 0.40.4mm2 and
0.50.5mm2, and slice thickness was
between3mmand4mm.Theechotime
(TE)rangedfrom10.8msto13.6ms,the
repetition time (TR) from 416ms to
689ms,andtheflipangle(FA)was90,
111, or160. TheCTimaging in-plane
voxelsizerangedfrom0.30.3mm2to
0.50.5mm2, and slice thickness from
0.3mmto0.6mm.TheCBCTimaging
in-planevoxelsizewas 0.30.3mm2 and
slicethicknesswas1mm.Themeantime
betweentheMRIand(CB)CTscanswas9
days(range2–23days,standarddeviation
(SD)5.3days).Nopre-orpost-processing
was applied to the images and they
remainedunmodifiedfortheregistration.
Registration
Anautomatedimageregistrationmethod
toaligntheMRIwiththe(CB)CTimages
was investigated. The alignment was
achievedinthreestages.Inthefirststage,
aninitialrigid alignmentwas estimated.
Thereafteramorerefinedrigidalignment
wasestimated,focusedaroundthe
mandi-ble.Inthethirdandfinalstage,a
deform-ablealignmentwasperformed,forwhich
different approaches were compared. In
all stages, an automated intensity-based
3D registration framework (Elastix)20
wasused, basedonthemaximization of
mutual information21 using a stochastic
gradientdescentoptimizationmethod22.
Inthefirststage,twoconsecutive
regis-trations were performed to achieve an
initial rigid alignment. First, a global
imagesneededtoberoughlyalignedtothe
(CB)CT image domain. Subsequently, a
rigidregistrationwasconducted,
estimat-ingbothtranslationsandrotations
(param-eterizedbyEulerangles).
In the second stage, the initial rigid
alignment was fine-tuned by restricting
thefocusofthealgorithmona3Dregion
of interest encompassing the mandible,
manually drawn in the (CB)CT image.
As such, all image information outside
oftheregionofinterestwasignoredand
potential registration difficulties due to
pose orappearance changescouldbe
al-leviated.Themandibularmaskwasdrawn
slice by slice, using open-source
ITK-SNAPsoftware23.
In the third stage, an evaluation was
performedtodeterminewhetherthe
align-mentcouldberefinedfurtherusinga
non-rigid(ordeformable)registrationto
com-pensateforanygeometricdistortionsinthe
MRIimages24.AparametricB-spline
free-form deformation model was employed,
and (isotropic) control point spacings of
64mm,32mm,and16mmwere
evalua-ted25. Furthermore, two different
approachesforthedeformableregistration
were compared: (a) an asymmetric
ap-proachwiththe(CB)CTimageasthefixed
(ortarget,reference)imageandtheMRIas
the moving(ortemplate, source)image; and
(b) a symmetricapproach in which both
imageswereregisteredtoacommon
mid-space26.Resultsfromtheliteraturesuggest
thatsymmetricregistrationtechniquescan
leadtoimprovedregistrationaccuracyand
inverse-consistency27,28, which are
espe-ciallycriticalfortreatmentplanning29.
AllstagesareillustratedinFig.1.The
threeregistrationstageswereexecutedin
a consecutive manner. The registration
resultafter stage1was employedto
ini-tialize the registration in stage 2. The
registration result after stage 2 was
employed toinitializethe registration in
stage 3,whereeitherapproach3aor
ap-proach3bwastaken.
Evaluation
Registration accuracy in the mandibular
regionwasevaluatedintermsofthemean
targetregistrationerror(mTRE),by
com-puting the Euclidean distance between
corresponding landmarks in MRI and
(CB)CT and then averaging it over all
landmarks30. An extensive landmark set
was designed with 39 anatomical
refer-encepointsinordertoevaluatethe
regis-tration error for the entire mandible
specifically.Thesetconsistedof22
land-marks placed at theroots ofeach lower
tooth (for the molars, both roots were
consideredaslandmarks)and17
anatom-icalreferencepointsonthemandible.The
set oflandmarksisdescribedindetailin
Table1andFig.2.Duetotumourinvasion
intheboneorremovedtoothelements,not
all landmarks from thedataset couldbe
annotatedforallimages.
Thelandmarkswereannotatedmanually
in eachMRIand(CB)CTacquisition by two
researchers (M.S.H., J.-M.G.) who were
trainedinadvance.Thereliabilityofeach
landmarkwas investigatedbycalculating
theinter-observervariabilityforbothMRI
and(CB)CTimagesseparately.Inaddition,
oneobserver(M.S.H.)repeatedthe
annota-tionofalllandmarks,toenablethe
assess-mentoftheintra-observervariability.
Intra-andinter-observervariabilitywere
quanti-fiedbycalculatingtheEuclideandistance
(similartothemTRE)between
correspond-ing landmarks ofthe same anddifferent
observer,respectively.
If the inter-observer variability for a
landmarkwasgreaterthan5mm,the
land-markwasexcluded(whenapplicableboth
for leftandright sides).After the
exclu-sion of the unreliable landmarks, the
mTREwascalculatedforeachregistration
stage toassesstheregistrationaccuracy.
Statisticalanalysis
Two statisticalanalyseswereperformed.
First,acomparisonwasmadebetweenthe
stagethatperformedbest(withthelowest
Fig.1. Theproposedthree-stageregistrationmethodtoalignMRIwith(CB)CTimagesinthemandibularregion.Inthefirststage,aninitialrigid alignmentwasestimated.Thereafter,inthesecondstage,amorerefinedrigidalignmentfocusedaroundthemandiblewasestimated.Inthethird (andfinal)stage,twoapproachesfordeformablealignmentwereinvestigatedsidebyside:anasymmetric(3a)andasymmetricapproach(3b). Notethat,unliketheschematic2Dillustrationsinthisfigure,allregistrationswereperformedcompletelyinthe3Dspace.
mTRE) and allother registration stages.
Second, forthethirdstage acomparison
was made between the asymmetric and
symmetric approachfor each spacing of
the control points (64mm, 32mm, and
16mm). A two-sided Wilcoxon signed
rank test at a significance level of0.05
wasusedtoevaluatethedifferencesinthe
distributionofthemTRE.
Results
Theinter-observervariabilityforthe
land-marks gonion (C) and coronoid process
(E) wasgreater than5mm(6.5mmand
6.0mm, respectively). These landmarks
onbothsidesofthemandiblewere
there-foreexcludedfromthecalculationofthe
mTRE andtheobservervariabilities.On
average, 20 landmarks per patient
remainedtocalculatethemTRE.
Theinter-observervariabilityand
intra-observer variability (meanSD over all
subjects) for landmark placement were
found to be 2.40.7mm and
2.00.5mm,respectively,forMRIimages
and 1.50.8mm and 1.00.3mm,
re-spectively,for(CB)CTimages.
After rigid registration instage 1, the
mTRE(meanSDoverallsubjects)was
3.11.8mm. After rigid registration
withtheuseofamaskaroundthe
mandi-ble (stage 2), the mTRE was 2.5
0.7mm.Afterasymmetricnon-rigid
reg-istration (stage 3a)withB-splinecontrol
point spacings of 64mm, 32mm, and
16mm, the mTREs were 3.61.1mm,
3.41.1mm,and3.31.2mm,
respec-tively. Inthe symmetricnon-rigid
regis-tration(stage3b),themTREvalueswere
foundtobe3.51.2mm,3.31.2mm,
and 3.11.3mm, respectively. Fig. 3
shows the distributions of mTRE over
all subjects for each registration stage,
as well as the inter- and intra-observer
variability.
Comparedtostage2,whichyieldedthe
lowestaveragemTRE, stage 3produced
significantly different mTREvalues (3a,
64mm: W=5, P<0.001; 3a, 32mm: W=2, P<0.001; 3a, 16mm: W=6, P<0.001; 3b, 64mm: W=3, P<0.001; 3b, 32mm: W=15, P=0.001; 3b, 16mm: W=26, P=0.005). No significant difference
was found between stage 1 and stage 2
(W=47, P=0.054)orbetween stage 3a
andstage3b(64mm:W=59,P=0.147;
32 mm: W=69, P=0.294; 16 mm:
W=48,P=0.059).Arepresentativecase
illustratingtheresultoftheregistrationof
MRIandCTimagesafterstage2isgiven
inFig.4.
Table1. Descriptionoftheanatomicallandmarksontheteethandthemandible.
Abbreviation Landmark Description
t31,t32,t33,t34,t35, t36,t37,t38, t41,t42,t43, t44,t45,t46,t47,t48
Teethofthethirdandfourthquadrants 22landmarksontheteeth;themolars(36,37,38,46,47, 48)havetworoots
A Menton Themostinferiorpointofthemandibularsymphysis
B Mentalforamen Foramenlocatedontheanteriorsideofthemandible
C Gonion Apointdefinedasthemandibularangle,representingthe
intersectionofthelinesoftheposteriorramusandthe inferiorborderofthemandible
D Mandibularforamen Foramenlocatedontheinternalsurfaceontheramus
E Coronoidprocess Thetipofthecoronoidprocess
F Leftcondylion Leftmostaspectofthecondylarhead
G Rightcondylion Rightmostaspectofthecondylarhead
H Topcondylion Topofthecondylarhead
I Mandibularnotch Notchlocatedatthemostsuperiorpointoftheramus,
whichseparatesthecoronoidprocessanteriorlyandthe condyloidprocessposteriorly
Fig.2. Toothandmandibularevaluationlandmarks.Thepurpleandbluelandmarksindicatetheapicesoftheteethandtheanatomicalpositionsof themandible,respectively.These39landmarkscorrespondwiththelandmarksdescribedinTable1.LandmarkDisnotillustratedinthisfigure, becausethislandmarkisbehindthefieldofview.Landmarkst38andt48werenotpresentinthissubject.Notethatthe3Dmodelshownherewas generatedforillustrationpurposesonly;duringannotationofthelandmarks,theoriginal(CB)CTandMRIacquisitionswereused,inspecting imageslicesinthreeorthogonalplanes(axial,sagittal,coronal).
Regardingthecomputationtime,stage
1required26665stocomplete;stage2
required an additional 13137s. Stage
3a required an additional 28548s,
28834s, and 29451s, for 64mm,
32mm,and16mm,respectively,whereas
in stage 3b these computations required
1138140s, 1234139s, and
1356151s, respectively. All
experi-ments were performed single-threaded
onthelocaluniversityCPUcluster.
Discussion
Thisstudyshowedthattherigid
registra-tionwithamask (stage2)isthe
recom-mendedmethodfor registeringMRIand
(CB)CTimagesinthemandibularregion.
Stage2achievedalowermTREcompared
tostage1,althoughthedifferencewasnot
significant (P=0.054). However, the
lo-calizedfocusinstage2shouldgeneralize
bettertootherpatientsandbemorerobust
to outliers. A protocol for applying the
recommendedmethodisprovidedinthe
AppendixA.
In this work, two approaches for
de-formableregistrationwereapplied.A
con-ventionalasymmetricapproachwherethe
(CB)CTimagewasemployedasthefixed
imageandtheMRIimageasthemoving
image (stage 3a) and a symmetric
ap-proachwheretheimageswereregistered
toacommonreferencesystem(stage3b).
Fig.3. Boxplotsrepresentingthedistributionsoftheinter-andintra-observervariabilitiesforCTandMRIandofthemTREvaluesforeach registrationstage.Thesquarebracketsbetweentwoconnectedboxplotsindicatestatisticalsignificanceatalevelof0.05.
Fig.4. ThisfigureillustratestheresultoftheregistrationofMRIandCTimagesafterstage2ofarepresentativecase(subject8).Themandible wassegmentedintheregisteredMRIimagesandoverlaidinredwiththecorrespondingCTimagesinthreerandomlyselectedslices.Theseresults werecombinedina3DrenderingwiththesegmentedskullandmandiblefromtheCTimages.Thegapatthefrontrepresentstumourtissuewhich isinvadingthemandibleinandaroundtheanteriorpartofthemandible.
Althoughthedifferenceswerenot
statisti-callysignificant,stage3bachieveda
mar-ginalimprovementcomparedtostage3a
for all control pointspacingsof64mm,
32mm,and16mm.Furthermore,the
reg-istrationerrorofallapproachesinstage3
was significantlyhigherthantheerrorin
stage 2.Assuch,theresultsindicatethat
non-rigid registration (stage 3) has no
addedvalueinthisapplication.
Withthelargenumberoflandmarks(35
aftertheexclusions),themTREcouldbe
estimatedreliably.Thiswaseventhecase
for patients for whomnotall landmarks
couldbeannotated,e.g.caseswheretooth
elementswereextracted,caseswithbone
invasion by tumour, and cases with a
landmark outside the fieldof view. The
inter- and intra-observer variability of
landmark annotations in the (CB)CT
images were consistent with those
reported in the literature31,32.
Further-more, the mTRE achieved by the best
registration method (stage 2, 2.5mm)
wassimilartotheinter-andintra-observer
variabilitiesforMRIimages(2.4mmand
2.0mm, respectively). As such, lower
mTRE values based on the landmarks
employed in this study can hardly be
expected. Note that a recent study on
landmarkaccuracyinMRIimagesfound
lower inter- andintra-observer
variabili-ty33. However, this difference can be
explainedbytheconsiderablylowerslice
thicknessemployedinthatpreviousstudy
(0.53mmvsatleast3mminthepresent
study). Basedontheinter- and
intra-ob-servervariabilities, thegonionand
coro-noid process landmarks were excluded;
their annotations were most likely
hin-dered bythe poor delineationand fuzzy
boundariesofthelandmarklocations34.Of
note,severalotherevaluation
methodolo-giesforregistrationaccuracyexist,suchas
theEuclideandistancebetweencentroids,
overlapmeasures,andsurfacedistancesof
manually segmented anatomical
struc-tures35,36.Inthisstudy,weoptedfor
man-ual landmark annotations, since a
relatively large number of well-defined
landmarks could be identified, allowing
areliableestimationofthemTRE,while
manual segmentation would have been
muchmoretime-consuming.
Itappearsthatdataontheregistrationof
MRIand(CB)CTimagesoftheheadand
neck region are scarce. Previous studies
havefoundregistrationerrorsintherange
of 1.7–3.3mmin datasets offour to 16
patients37–41.Noneoftheseprevious
stud-ies focusedspecificallyonthemandible,
hinderingathoroughcomparisonwiththe
present study. However, the results of
these studies suggest that the mTRE of
2.5mm(forstage2)achievedindicatesa
state-of-the-art errorlevel.
Fortunatietal.42suggestedthatpatient
immobilization during imaging leads to
better registration of the MRI and (CB)
CTimages oftheheadandneck region.
Their study founda registration errorof
7.0mmwithoutimmobilizationanda
reg-istration error of1.9mmwith
immobili-zation. The implementation of
immobilizationequipmentforourspecific
applicationmightnotaddvalueinclinical
practice,sinceacompetitivemTREof2.5
mmwasalreadyachievedwithout
immo-bilization. Moreover, rigorous
immobili-zation of the mandible would be
challenging and likely not comfortable
forthepatient.
Themandibular mask used in stage 2
wasdrawnmanuallyaroundthemandible
ineach slice ofthe(CB)CTimages.
Al-though the mask does not have to be
delineated very precisely (it just serves
to indicatean approximate region of
in-terest), this manual interaction step may
notbe desirableinclinicalpractice. The
developmentofarobustsemi-automated
orevenfullyautomatedsegmentation43is
thereforerecommendedtoacceleratethis
step.Wereferthereadertoarecentreview
of such methodologies for a full
over-view44.
Althoughtheopen-sourceElastix
soft-ware was used to implement the image
registrations in this study, other
(open-source or commercial) software
applica-tions that implement similarregistration
algorithmsbasedonmaximizationof
mu-tualinformationcouldhavebeenusedas
well. Some well-known open-source
examples include NiftyReg45 and
ITKv4/ANTS46. After proper
configura-tion, these tools areexpected toachieve
similarregistrationaccuracy.
Correctdeterminationoftheosteotomy
location depends onseveral factors(e.g.
the waiting time between imaging and
surgery,theprocessofthetranslationfrom
3D virtual planning to the patient, the
accurate placement of the cutting guide
during surgery, the length of the fibula
reconstruction,therelationshipofthe
tu-mourtothementalnerveandthe
remain-ing teeth), which could independently
contribute to positiveresection margins.
Forexample, severalstudieshaveshown
that CBCT results in less accurate 3D
planning models than CT47. When the
proposedregistrationmethodistranslated
into clinical practice, image registration
errors havetobe consideredin
conjunc-tionwithallother sourcesoferrors.
No clinical outcome criteria were
employedinthiswork,suchastheresection
margin,thefrequencyofrevisingthe
surgi-calguideduringplanning,orfrequencyof
local tumour progression, as this was a
retrospectivestudy.Arandomizedclinical
trialisneededtodeterminetheaddedvalue
ofregisteredMRIandCB(CT)invirtual
planningofmandibularresection and
re-construction.
Thisstudypresentedanimage
registra-tionmethodforaligningMRIand(CB)CT
images of the mandibular region in
patientswithOSCC.Itshowedthatrigid
registration within a region of interest
drawnaroundthemandibleisthe
recom-mendedregistrationmethodforthe
align-mentofMRIand(CB)CTimages inthe
mandibularregion.
Funding
ThisstudywaspartiallyfundedbyFonds
NutsOhra(nr.1404-048).
Competinginterests
Theauthorsdeclarethatthereisno
con-flictofinterest.
Ethicalapproval
Thestudywasreviewedandapprovedby
thelocalmedicalethicsreviewcommittee
(MEC-2016-143) and was performed in
accordancewithnationalandinternational
legislation.
Patientconsent
The need for informed consent was
waived owing to the retrospective and
anonymizednatureofthestudy.
AppendixA
Protocol
Wehavemadethecodetoexecutethe
registrationavailableinapublicGitHub
re-pository (https://github.com/
MathiasLPolfliet/
mandible_ct_mri_registration).Theprotocol
canbesubdividedintothreesteps:
1) Obtain a rough registration mask of
the mandible.Thebinary registration
must overlay the mandible entirely.
Furthermore, the mask must have a
boundary around the mandible of
roughly16pixelsinalldimensions.
2) Perform initial rigid registration
(stage 1). Perform a sequential
transformationfollowed bya
transla-tionandrotation,toroughlyalignthe
anatomical structures in the
acquisi-tions. Store the resulting
transforma-tiontoinitializestage2.
3) Perform rigid registration with
maskedfieldofview(stage2).
Initial-ize anew registration withthe result
fromstage1andlimitthefieldofview
withthebinaryregistrationmask.
Op-timizeforthetranslationandrotation
transformation.
References
1. BinahmedA,NasonRW,AbdohAA.The clinicalsignificanceofthepositivesurgical margin in oral cancer. Oral Oncol 2007;43:780–4.
2. CarvalhoAL,NishimotoIN,Califano JA, KowalskiLP.Trendsinincidenceand prog-nosisforheadandneckcancerintheUnited States:asite-specificanalysisoftheSEER database.IntJCancer2005;114:806–16.
3. NieberlerM,Ha¨ußlerP,KestingMR,Kolk A,DeppeH,WeirichG,WolffK-D.Clinical impactofintraoperativecytological assess-mentofboneresectionmarginsinpatients with headand neck carcinoma.Ann Surg Oncol2016;23:3579–86.
4. ShahJP,GilZ.Currentconceptsin manage-ment oforalcancer—surgery.OralOncol 2009;45:394–401.
5. SafiAF,KaukeM,GrandochA,Nickenig HJ,Zo¨llerJE,KreppelM.Analysisof clini-copathologicalriskfactorsforlocoregional recurrenceof oral squamouscell carcino-ma—retrospectiveanalysisof517patients.J CraniomaxillofacSurg2017;45:1749–53.
6. Succo G,Berrone M,Battiston B,Tos P, GoiaF,AppendinoP,CrosettiE. Step-by-step surgicaltechniqueformandibular re-constructionwithfibularfreeflap: applica-tionofdigitaltechnologyinvirtualsurgical planningEur Arch Otorhinolaryn-gol272201514911501.EurArch Otorhino-laryngol2015;272:1491–501.
7. Smits RWH, TenHoveI, Dronkers EAC, Bakker Schut TC, Mast H, de Jong RJ Baatenburg, Wolvius EB, Puppels GJ, KoljenovicS.Evaluationofboneresection margins ofsegmental mandibulectomy for oralsquamouscellcarcinomaIntJOral Max-illofacSurg472018959964.IntJOral Max-illofacSurg2018;47:959–64.
8. KademaniD,BellRB,BagheriS,Holmgren E,DierksE,PotterB,HomerL.Prognostic factorsinintraoralsquamouscellcarcinoma: the influence of histologic grade. J Oral MaxillofacSurg2005;63:1599–605.
9. KreppelM,DrebberU,EichHT, Dreisei-dler T, Zo¨ller JE, Mu¨ller RP, Scheer M. Combined-modalitytreatmentinadvanced oralsquamouscellcarcinoma:Primary sur-gery followed by adjuvant concomitant
radiochemotherapyStrahlenther
Onkol1872011555560.StrahlentherOnkol 2011;187:555–60.
10. BarrosoEM,TenHoveI,SchutTCBakker, Mast H,vanLanschot CGF,Smits RWH, CaspersPJ,VerdijkR,NoordhoekHegtV, BaatenburgdeJongRJ,WolviusEB, Pup-pelsGJ,KoljenovicS.Ramanspectroscopy forassessmentofboneresectionmarginsin mandibulectomy for oralcavity squamous cell carcinomaEur J Cancer9220187787. EurJCancer2018;92:77–87.
11. DillonJK,BrownCB,McDonaldTM, Lud-wigDC,ClarkPJ,LerouxBG,FutranND. Howdoestheclosesurgicalmarginimpact recurrence andsurvivalwhentreatingoral squamouscellcarcinoma?JOralMaxillofac Surg73201511821188. J Oral Maxillofac Surg2015;73:1182–8.
12. VarvaresMA,PotiS,KenyonB,Christopher K,WalkerRJ.Surgicalmarginsandprimary siteresectionin achievinglocalcontrolin oral cancer resections. Laryngoscope 2015;125:2298–307.
13. TarsitanoA,CioccaL,CiprianiR,ScottiR, Marchetti C. Mandibular reconstruction usingfibulafreeflapharvestedusinga cus-tomisedcuttingguide:howwedoit.Acta OtorhinolaryngolItal2015;35:198.
14. CorneliusCP,GiesslerGA,WildeF, Metz-ger MC,MastG,ProbstFA. Iterations of computer-andtemplateassistedmandibular ormaxillaryreconstructionwithfreeflaps containingthelateralscapular border—evo-lutionofabiplanarplug-oncuttingguide.J CraniomaxillofacSurg2016;44:229–41.
15. LiC,MenY,YangW,PanJ,SunJ,LiL. Computedtomographyforthediagnosisof mandibular invasion caused by head and neckcancer:asystematicreviewcomparing contrast-enhanced and plain computed to-mography. J Oral Maxillofac Surg 2014;72:1601–15.
16. LiC,YangW,MenY,WuF,PanJ,LiL. Magneticresonanceimagingfordiagnosisof mandibularinvolvementfromheadandneck cancers:asystematicreviewand meta-anal-ysis.PLoSOne2014;9:e112267.
17. Blatt S, ZiebartT, Kru¨ger M,PabstAM. Diagnosingoralsquamouscellcarcinoma: how muchimagingdowereally need?A review ofthecurrentliterature. J Cranio-maxillofacSurg2016;44:538–49.
18. DongY,DongY,HuG,XuQ. Three-dimen-sional reconstruction of extremity tumor regionsbyCTandMRIimagedatafusion forsubject-specificpreoperativeassessment and planning. Comput Aided Surg 2011;16:220–33.
19. KraeimaJ,DorgeloB,GulbittiHA, Steen-bakkersRJHM,SchepmanKP,Roodenburg JLN,SpijkervetFKL,SchepersRH,Witjes MJH.Multi-modality3Dmandibular resec-tionplanninginheadandneckcancerusing CTandMRIdatafusion:aclinical serie-sOral Oncol8120182228. Oral Oncol 2018;81:22–8.
20. KleinS,StaringM,Murphy K,Viergever MA,PluimJP.Elastix:atoolboxfor intensi-ty-basedmedicalimage registration. IEEE TransMedImaging2010;29:196–205.
21. The´venazP,UnserM.Optimizationof mu-tual informationfor multiresolutionimage registration. IEEE Trans Image Process 2000;9:2083–99.
22. KleinS,PluimJP,StaringM,ViergeverMA. Adaptivestochasticgradientdescent optimi-sationforimageregistration.IntJComput Vis2009;81:227–39.
23. YushkevichPA,PivenJ,HazlettHC,Smith RG,HoS,GeeJC,GerigG.User-guided3D active contoursegmentationofanatomical structures:significantlyimprovedefficiency andreliabilityNeuroimage31200611161128. Neuroimage2006;31:1116–28.
24. Chang H,FitzpatrickJM.Atechniquefor accuratemagneticresonanceimaginginthe presence of field inhomogeneities. IEEE TransMedImaging1992;11:319–29.
25. RueckertD,SonodaLI,HayesC,HillDL, LeachMO,HawkesDJ.Nonrigid registra-tionusingfree-formdeformations: applica-tiontobreastMRimages.IEEETransMed Imaging1999;18:712–21.
26. Bhatia KK, HajnalJV,Puri BK,Edwards AD,RueckertD.Consistentgroupwise non-rigidregistrationforatlasconstruction. Bio-medical Imaging: Nano to Macro, 2004. IEEEInternationalSymposiumon Biomed-icalImaging.IEEE;2004:908–11.
27. AganjI,IglesiasJE,ReuterM,SabuncuMR, Fischl B. Mid-space-independent deform-able image registration. Neuroimage 2017;152:158–70.
28. LorenziM,AyacheN,FrisoniGB,Pennec X.Alzheimer’sDiseaseNeuroimaging Ini-tiative(ADNI).LCC-Demons:arobustand accuratesymmetricdiffeomorphic registra-tion algorithm. Neuroimage 2013;81:470– 83.
29. Rivest-He´nault D, Dowson N, Greer PB, FrippJ,DowlingJA.Robust inverse-consis-tent affine CT–MR registration in MRI-assisted and MRI-alone prostate radiation therapy.MedImageAnal2015;23:56–69.
30. FitzpatrickJM,WestJB.Thedistributionof targetregistrationerrorinrigid-body point-basedregistration.IEEETransMedImaging 2001;20:917–27.
31. LudlowJB,GublerM,CevidanesL,MolA. Precisionofcephalometriclandmark identi-fication: cone-beamcomputed tomography vsconventionalcephalometricviews.AmJ OrthodDentofacialOrthop2009;136.312. e1–10.
32. Lagrave`reMO,LowC,Flores-MirC,Chung R,CareyJP,HeoG,MajorPW. Intraexami-ner andinterexaminerreliabilitiesof land-mark identification on digitized lateral cephalogramsandformatted3-dimensional cone-beamcomputerizedtomography ima-gesAm J Orthod Dentofacial Orthop1372010598604.AmJOrthod Den-tofacialOrthop2010;137:598–604.
33. Juerchott A, Freudlsperger C, Zingler S, SaleemMA,JendeJM,LuxCJ,Bendszus M,HeilandS,HilgenfeldT.Invivo reliabil-ityof3Dcephalometriclandmark determi-nation on magnetic resonanceimaging: a feasibility study. Clin Oral Investig 2020;24:1339–49.
34. WilliamsFL,RichtsmeierJT.Comparisonof mandibular landmarks from computed to-mographyand3Ddigitizerdata.ClinAnat 2003;16:494–500.
35. MurphyK,vanGinnekenB,KleinS,Staring M,deHoopBJ,ViergeverMA,PluimJPW. Semi-automatic construction of reference standardsforevaluation ofimage registra-tionMedImageAnal1520117184.Med Im-ageAnal2011;15:71–84.
36. RohlfingT.Imagesimilarityandtissue over-laps as surrogates for image registration accuracy:widelyusedbutunreliable.IEEE TransMedImaging2011;31:153–63.
37. Kraeima J,SchepersRH,vanOoijenPM, Steenbakkers RJ, Roodenburg JL, Witjes MJ. Integration of oncologic margins in three-dimensionalvirtualplanningforhead andnecksurgery,includingavalidationof the software pathway. J Craniomaxillofac Surg2015;43:1374–9.
38. FortunatiV,VerhaartRF,AngeloniF,vander LugtA,NiessenWJ,VeenlandJF,Paulides MM, vanWalsum T. Feasibilityof multi-modaldeformableregistrationforheadand necktumortreatment planningIntJRadiat Oncol BiolPhys9020148593.IntJ Radiat OncolBiolPhys2014;90:85–93.
39. LeibfarthS,Mo¨nnichD,WelzS,SiegelC, SchwenzerN,SchmidtH,ZipsD,Thorwarth D. A strategy for multimodal deformable imageregistrationtointegratePET/MRinto radiotherapy treatment planningActa Oncol52201313531359. Acta Oncol 2013;52:1353–9.
40. duBoisd’AischeA,DeCraeneM,GeetsX, Gre´goireV,MacqB,WarfieldSK. Estima-tionofthedeformationsinducedby articu-lated bodies: registration of the spinal column. Biomed Signal Process Control 2007;2:16–24.
41. WebsterGJ,KilgallonJ,HoKF,Rowbottom CG,SlevinNJ,MackayRI.Anovelimaging techniqueforfusionofhigh-quality immo-bilisedMRimagesoftheheadandneckwith CTscansforradiotherapytargetdelineation. BrJRadiol2009;82:497–503.
42. FortunatiV,VerhaartRF,VerduijnGM,van derLugt A, Angeloni F,NiessenWJ,Veenland JF,PaulidesMM,vanWalsumT.MRI inte-grationintotreatmentplanningofheadand necktumors:canpatientimmobilizationbe avoided?Radiother Oncol1152015191194. RadiotherOncol2015;115:191–4.
43. EggerJ,PfarrkirchnerB,GsaxnerC, Lind-nerL,SchmalstiegD,WallnerJ.Fully con-volutionalmandiblesegmentationonavalid ground-truthdataset.ProceedingsIEEE En-gineeringinMedicineandBiology Confer-ence(EMBC).IEEE2018:656–60.
44. WallnerJ,SchwaigerM,HocheggerK, Gsax-ner C, Zemann W, Egger J.Areview on multiplatformevaluationsofsemi-automatic
open-source basedimage segmentationfor cranio-maxillofacialsurgery.Comput Meth-odsProgramsBiomed2019;182:105102.
45.ModatM,RidgwayGR,Taylor ZA, Leh-mann M,BarnesJ,HawkesDJ, FoxNC, Ourselin S. Fast free-form deformation using graphics processing unitsComput Methods ProgramsBiomed982010278284. Comput Methods Programs Biomed 2010;98:278–84.
46.AvantsBB,TustisonNJ,StaufferM,SongG, WuB,GeeJC.TheInsightToolKitimage registrationframework.FrontNeuroinform 2014;8:44.
47.van BaarGJ, ForouzanfarT,Liberton NP, WintersHA,LeusinkFK.Accuracyof com-puter-assistedsurgeryinmandibular recon-struction: asystematic review.OralOncol 2018;84:52–60.
Address: StefanKlein
BiomedicalImagingGroupRotterdam DepartmentofRadiologyandNuclear Medicine
ErasmusUniversityMedicalCenter POBox2040
3000CARotterdam TheNetherlands Tel.:+31107043442 E-mail:s.klein@erasmusmc.nl