Three-dimensional and non-destructive characterization of nerves inside conduits using laboratory-based micro computed tomography

(1)

Article

Reference

Three-dimensional and non-destructive characterization of nerves inside conduits using laboratory-based micro computed tomography

BIKIS, Christos, et al.

Abstract

Histological assessment of peripheral nerve regeneration in animals is tedious, time-consuming and challenging for three-dimensional analysis.

BIKIS, Christos, et al . Three-dimensional and non-destructive characterization of nerves inside conduits using laboratory-based micro computed tomography. Journal of Neuroscience Methods , 2018, vol. 294, p. 59-66

PMID : 29129635

DOI : 10.1016/j.jneumeth.2017.11.005

Available at:

http://archive-ouverte.unige.ch/unige:154945

Disclaimer: layout of this document may differ from the published version.

1 / 1

(2)

ContentslistsavailableatScienceDirect

Journal of Neuroscience Methods

j ou rn a l h om epa g e : w w w . e l s e v i e r . c o m / l o c a t e / j n e u m e t h

Three-dimensional and non-destructive characterization of nerves inside conduits using laboratory-based micro computed tomography

Christos Bikis

^a

, Peter Thalmann

^a

, Lucas Degrugillier

^b

, Georg Schulz

^a

, Bert Müller

^a,∗

, Daniel F. Kalbermatten

^c,d

, Srinivas Madduri

^b,c,e,∗∗

, Simone E. Hieber

^a,∗^∗^∗

aBiomaterialsScienceCenter,DepartmentofBiomedicalEngineering,UniversityofBasel,Switzerland

bCenterforBioengineeringandRegenerativeMedicine,DepartmentofBiomedicalEngineering,UniversityofBasel,Switzerland

cDepartmentofPlastic,Reconstructive,AestheticandHandSurgery,BaselUniversityHospital,Switzerland

dDepartmentofPathology,UniversityHospital,Switzerland

eDepartmentofBiomedicine,UniversityofBasel,Switzerland

h i g h l i g h t s

•3D imaging of parafﬁn-embedded peripheral nerves without contrast agentachieved.

•Micro-anatomicalfeaturesofnerves resolvedeveninsideconduit.

•Automatic nerve segmentation despiteofcharacteristicartifacts.

•Automaticextractionof anatomical parametersin afew minutes fora 100mm³volume.

g r a p h i c a l a b s t r a c t

a r t i c l e i n f o

Articlehistory:

Received26August2017

Receivedinrevisedform7November2017 Accepted7November2017

Availableonline10November2017

Keywords:

␮CT X-ray 3Dimaging

Computedtomography Nervesegmentation Vesselassessment

a b s t r a c t

Background: Histologicalassessment of peripheralnerve regeneration inanimals is tedious,time- consumingandchallengingforthree-dimensionalanalysis.

Newmethod:Thepresentstudyreportsonhowandtowhatextentmicrocomputedtomographyof parafﬁn-embeddedsamplescanprovideareliablethree-dimensionalapproachforquantitativeanalysis ofperipheralnerves.

Results: Rat sciatic nerves were harvested, formalin-fixated, positioned intonerve conduits (NC), paraffin-embedded, and imaged using a laboratory-based X-ray microtomography system with an isotropic voxel length of 4␮m. Suitable quantitative measures were identified and auto- matically evaluated, i.e. nerve length, cross-sectional area and volume, as well as vascular structures, to be used as an assessment and comparison indicator of regeneration quality.

Abbreviations:NC,nerveconduit;␮CT,X-raymicrotomography.

∗Correspondingauthorat:BiomaterialsScienceCenter,DepartmentofBiomedicalEngineering,UniversityofBasel,Gewerbestrasse14,4123Allschwil,Switzerland.

∗∗Correspondingauthorat:CenterforBioengineeringandRegenerativeMedicine,DepartmentofBiomedicalEngineering,UniversityHospitalBasel,Spitalstrasse21/Peters- graben4,4031Basel,Switzerland.

∗∗∗Correspondingauthorat:BiomaterialsScienceCenter,DepartmentofBiomedicalEngineering,UniversityofBasel,Gewerbestrasse14,4123Allschwil,Switzerland.

E-mailaddresses:[email protected](B.Müller),[email protected](S.Madduri),[email protected](S.E.Hieber).

https://doi.org/10.1016/j.jneumeth.2017.11.005

0/).

(3)

60 C.Bikisetal./JournalofNeuroscienceMethods294(2018)59–66

Comparisonwithexistingmethods:Comparedtoimagingusingcontrastagents,theinvestigatedspecimens cansubsequentlyundergotheconventionalhistologicalanalysiswithoutrequiringadditionalpreparation steps.Contrastandspatialresolutionarealsoincreasedsigniﬁcantly.

Conclusions:Wedemonstratethepotentialofthemicrocomputedtomographyfornon-destructivemon- itoringofperipheralnervesinsidetheconduits.

1. Introduction

Histology and histopathology focus on the study of cellu- larandtissuemicroanatomyinhealthanddisease,respectively.

Theyplay aprominent role, bothin researchand clinicalprac- tice and are based on the main principle of examining tissue slicesunderanopticalmicroscope.Inseveralapplications,however,thisapproachisnotstraightforward.Invivoimagingmethods formonitoringaxongrowthusingﬂuorescencelackmicrometer resolutionandofferamaximalpenetrationdepthcloseto2mm (Kerschensteiner etal.,2005).Immunohistochemicialstainingis thegoldstandardfortheanalysisofregeneratednervetissue,but exhaustiveserial sectioningis demanding for theevaluation in threedimensions(Godinhoetal.,2013;Maddurietal.,2010b).Elec- tronmicrographsofferahighdegreeoflateralresolution,butthe approachisassociatedwithsamplesizeconstraints(Godinhoetal., 2013).

ItisknownthatX-raymicrotomography(␮CT)ofsofttissues andnerves scannedin aqueoussolutions providesimageswith limitedcontrast,duetotheweakdifferenceoftissueconstituents inX-rayabsorption.Therefore,iodineascontrastagent,hasbeen applied for the X-ray imaging of nerves in conduits (Hopkins etal.,2015).Veryrecentlyithasbeendemonstratedthatparaf- ﬁnembeddingofhumancerebellumallowsforthevisualization ofindividualPurkinje cells withoutstaining,using alaboratory

␮CT system(Khimchenko et al., 2016) and theirsegmentation usingfeature-based ﬁlteringof synchrotron radiation␮CT data (Hieberetal.,2016).Weherebyassumesuchalabel-freeapproach isalsosuitableforperipheralnerves,thathavea greatdemand forhigh-throughputregenerationmonitoring.Hence,wepropose toemployamulti-stepprocedurecombiningtissuepreparation, laboratorymicrocomputedtomographyandtailoredimageanal- ysistoquantitativelycharacterizeregeneratingnervesinsidean absorbablenerveconduit(NC)inatruethree-dimensionalmanner.

Thisnon-destructivemethodwillbeappliedbeforethepreparation ofhistologicalsections,alsoaidingintheselectionoftheappropri- atecuttingplanes(Stalderetal.,2014).Forthisfeasibilitystudy, healthyratsciaticnerveswereexplantedandembeddedinparafﬁn insideandoutsideacollagenNC(Maddurietal.,2010a).

Inordertoexhaustthefullcapacityofmicrocomputedtomogra- phy(Holmeetal.,2014),excessparaffinwasremovedasrequired, bymeansofapplyingaspecificpreparationprotocol.Tomography datawereacquired withhighcontrastand analyzed by asoft- waredevelopedforbothratnerveswithandwithoutNC.Identical anatomicaldetailswereidentifiedinbothcases.Additionally,the volume,areaandlengthoftheinvestigatednervesandtheirvas- culaturewerealsocalculatedinanautomatedapproach,allowing forhigh-throughputinvestigationsofnerveregeneration.

2. Materialsandmethods

Thestudywasconductedincompliancewiththeethicalinstruc- tionsoftheVeterinaryOfﬁceoftheKantonofBasel-Stadt(Basel, Switzerland),permissionnumber25212.Threesciaticnerveswere surgicallyextractedfromhealthySpragueDawleyrats.Collagen

NCswerefabricatedbyspinningmandreltechnology,asillustrated previously(Maddurietal.,2010a).Insolublecollagen(2.5%,w/w) wasswollenin1Maceticacidandhomogenizedwithahigh-speed mixerat10,000rpm(Polytron^®,Kinematica,Lucerne,Switzerland) for1min.Thehomogeneouscollagendispersionwasappliedviaa syringeontoaspinningAu-coatedmandrel(diameterof1.5mm), installedinasidewaysreciprocatingapparatus,andthesolventwas driedoffunderlaminarairﬂow.Theresultingtubeswereneutral- izedbyincubationin0.1Mdi-sodiumhydrogenphosphate(pHof 7.4)for1h.Thetubeswereﬁnallycutinto14mmlongspecimens, whichwerecross-linkedbyphysicalmeans,i.e.bysubjectingthe collagentubestoadehydro-thermaltreatment(DHT)atatemper- atureof110^◦Candapressureof20mbarfor5days.Theresulting tubesexhibitedanouterdiameterof2.5mmandinnerdiameterof 1.5mm.

Thenerveswerefixedinformalinsolutionandsubsequently embeddedin paraffin blocks according tostandard histological preparation. In order to optimize the specimen’s diameter for thetomographyandtoavoidentrappedairbubblestheparaffin embeddingprocedurewasoptimized.Thelongnervewascutin twopartsforeaseofplacingthenervesegmentsinsidethecon- duitandfurthertoavoidanytissuedeformationoverthecourse oftissuealignmentalongthelumenoftheNC.Afterinsertingthe twosegmentsofnerve,eachfromtheoppositeopeningofNC,the NC-nervecomplexwasembeddedinparaffin.

The ␮CT system used for this study was the nanotom^®m (phoenix |X-ray, GE Sensing & Inspection Technologies GmbH, Wunstorf,Germany).Itisequippedwitha180kVnanofocustrans- missionsourceanda3072×2400pixelsGEDXRdetector.Forall thepresentedmeasurements,theaccelerationvoltagewassetto 60kVand thebeamcurrentto280␮A.Source-to-specimenand source-to-detectordistanceswere9mmand225mm,respectively, resultinginaneffectivepixelsizeof4␮m.Forallspecimens,1800 projectionswereacquiredequiangularlyovertherange of360^◦. Inordertoincreasephotonstatistics,nineimageswithanexpo- suretimeof0.5seachwereaveragedateachangularposition.This resultedinatotalscanningtimeof150minfor oneheightstep coveringanervesection9.6mmlong.Forspecimenslongerthan 9.6mm,twoormoreheightstepswereacquiredandstitchedafter reconstruction(Mülleretal.,2012).

Toreducetheeffectof faultydetectorpixelsappearing con- stantlydarkorbrightintheprojections,theacquiredradiographs werefilteredwitha 2Dmedian filterwitha3×3 kernelusing theopen-source software ImageJ(Schneider etal., 2012), prior toreconstruction.Therebythecontrastconsiderablyincreased,as foundintherelatedabsorptionhistograms,attheexpenseofneg- ligibleblurring.Thefilteredradiographswerereconstructedusing the nanotom^®m manufacturer’s software phoenix datosx 2.0.1 –RTM(GESensing&InspectionTechnologiesGmbH,Wunstorf, Germany),whichemploysamodifiedFeldkampconebeamrecon- structionalgorithm.Subsequently,thethree-dimensionaldatasets wereimportedintheVGStudioMAX2.1(VolumeGraphicsGmbH, Heidelberg,Germany)softwareandscaledtothesamegray-scale rangenecessaryforimageanalysis.These16-bitdatasetshavea sizeofapproximately2GBconsidering1cmofnerve.TheVGStudio

(4)

MAXwasusedforthethree-dimensionalrenderingoftheacquired tomograms.

Forthemorphologicalanalysis,theparaffin-embeddednervous tissue, partlywithin theconduit,hastobe segmented.Related stepsof this procedure are illustratedin Fig. 1. Image (a)rep- resents theraw data.Image (b) showsthat thecontrast inthe raw data wassufficient to successfully apply theOtsumethod (Otsu,1979),inordertodigitallyremovethescaffold.Adistance transform served for removing a thin part (3–6 voxels) of the remainingspecimen’s surface,which wasfollowed byselecting thelargestconnected component(c).Thisstepwasparticularly neededtoseparatethespecimenfromheavystreakingandarti- factscaused bytheremainingairbubbles.Astheartifactswere removedfromthehistograms,asecondthresholdingstepusing Otsu’smethodwasapplied(d).GiventhattheGaussiandistribu- tionsofthe bareparaffin andtheparaffin-embedded specimen inthe absorptionhistogramswereinsufficientlyseparated, this secondthresholdingstepdidnotresultinperfect virtualparaf- finremovalandcreatedsomevoids insidethetissue.Such kind of noise was removed using a 2D median filter with a kernel of size 3×3 (e). The segmentation was finalized by selecting thelargestconnectedcomponentandapplyinganadaptivevoid- filling filter(f). Aftersegmentation, thenerve wasskeletonized by calculating thegeometrical center for each slice. The entire segmentation and analysis procedure was performed in Mat- lab(MATLAB2016a,TheMathWorks,Inc.,Natick,Massachusetts, UnitedStates).

SincetheX-rayabsorptionvaluesofthevesselsoverlapwiththe onesofotherstructures,afeature-basedsegmentationisapplied tosegmentthem.Theprobabilityofavoxeltobelongtoaves- selis determinedbytheestablishedFrangi-Filterthatexamines theeigenvaluesoftheHessianmatrix(Frangietal.,1998).Theﬁl- terparameterswerechosento˛=0.1,ˇ=0.1,=0.2intheMatlab implementationbyKroon(2009).Thevoxelrangewasbetweenone andthreeandthethresholdofvesselness0.4forthenormalized dataset.

3. Results

3.1. Descriptiveevaluation

Fig.2comparesrepresentativeslicesoftwoCTdatasetsshow- ingthesamenerve.Inthepresentedcrosssectionsonecanreadily recognize the microanatomy of the selected nerve. In particu- lar,thecomponentsincludingtheepineuriumand perineurium membranessurroundingthemainaxonalmasscanbeidentiﬁed.

Thecharacteristicmicroanatomyofthenervefasciclebundlesis alsovisible.Atsomepositions,thevasanervorum(white-colored arrowheads)insidethenervebulkexhibitaconsiderablyhigher X-rayabsorptionthantheirsurroundingnervoustissue.

Theleftsectionalsocontainstheporousconduit,whichcanbe quantitativelycharacterizedaswell.Boththesurfacemorphology aswellasthelocationsandsizesoftheindividualporesinsidethe scaffoldmaterialareaccessible.

Ingeneral,thenervevisualizationinsidetheconduitischalleng- ingbecauseofitsrelativelyhighX-rayabsorption.Artifactssuch asstreaksimpedeadetaileddescription.Theimagesdemonstrate, however,thatthepresentexperimentalsetupenablesustoobtain resultswithoutsigniﬁcantdifferencesbetweenthebarenerveand thenervewithintheconduit.Theremainingdifferencesarelimited deformations,aresultofnervehandlingbetweentheacquisitions ofthetomograms.Streakartifactsarerare.

Contrary to histology, tomography not only allows two- dimensional imaging in predeﬁned directions but also the three-dimensionalrenderingwithisotropicvoxelsize.Fig.3shows

Table1

Measuredgeometricalparametersofthreeselectednerves.

Samplenumber Volume(mm³) Length(mm) Averagecrosssection(mm²) 1 3.54±0.39 9.59±0.23 0.45±0.05

2 1.81±0.20 11.00±0.28 0.34±0.04 3 2.81±0.31 11.92±0.38 0.47±0.05

anexemplaryrendering,whereaselectedspecimenisvirtuallycut intotwopartsalongthesymmetryaxis.Theinternalstructureof thetworatsciaticnervesegments,thegapbetweenthem,aswell asthesurroundingconduitcanbesimultaneouslyvisualized.

3.2. Nervesegmentation

Thequantiﬁcationofnerveregenerationrequirestheidentiﬁ- cationofthenervoustissuewithinthethree-dimensionaldataset.

The contrast,density resolution, of theCT dataenables a clear distinction between the air, the paraffin, the scaffold, and the nervoustissuebasedontheirspecificX-rayattenuation.Thus,a thresholdingapproach(Mülleretal.,2002)performswellforthe segmentationasillustratedinFig.1,cp.especiallyimages(b)and (d).Imagingartifactsatthesurfaceandnoiseduetophotonstatis- ticscanbehandledappropriatelybyfilteringtechniquesseeimages (c), (e) and (f). As a result, thesegmentation mask in Fig. 1(f) matcheswellwiththenervoustissueshowninthenon-treated dataset(Fig.1(a)).Eventhinlayersofnervoustissueincludingthe epineuriumandperineuriummembranesarerepresentedclearly.

3.3. Nervecharacterizationbygeometricalparameters

Thecontrastinthetomographydatawassufﬁcienttoautomati- callysegmentthenerve.Thesuccessoftheprocedurewasmanually validated.Forthispurpose,twoexpertscomparedtheoriginaland relatedsegmenteddatasection-wise.

In a subsequent step, the segmented nerve was charted to extractthemorphologicaldata,i.e.volume,length,andcrosssec- tion. Table 1 lists these geometrical parameters for the three selectednerves.

Although␮CTprovidestruemicrometerresolution,theerror bars arerather large. Inaccuracyis caused bydiscontinuitiesin thecurveprogression,asindicatedinFig.4.Thesediscontinuities arecausedbystreakartifacts,duetohighX-rayabsorbingspecies.

Therefore,thestreaksaremorefrequentlyfoundatconduitsand theirinterfaces(Fig.4).

From theanatomical point of view,the cross-sectional area shouldbeconsistentfromsectiontosection.Therefore,thesmooth- nessofthecurvesinthediagramsofFig.4isanindicatorforthe reasonableperformanceoftheautomaticsegmentation.

3.4. Vesselvisualization

Fig.5shows thesegmentedvessels afterﬁlteringthethree- dimensional dataset for tubular structures. At some positions thesegmentsareapparentlyconnectedtostreakartifacts,which impedestheirsegmentation.Ingeneral,onerecognizesthecourse ofthevessels,althoughtheyarefrequentlyinterruptedwithinthe acquiredthree-dimensionaldata.Therelatedvirtualsectionsprove thesigniﬁcantlyhigherX-rayabsorptionofthesegmentedpartsof thevisualizedbloodvesselsystem.

4. Discussion

Themicroanatomyofratperipheralnerveshasbeenstudiedin detail,seee.g.Bertellietal.(1995).ThequalityofthepresentCT slicesisalmostcomparabletotheslidesofdigitizedhistologywith

(5)

Fig.1. Across-sectionalandalongitudinalsectionofaselecteddatasetbeforeandaftersegmentation:(a)original,(b)airandscaffoldvirtuallyremoved,(c)boundaryeroded andlargestobjectselected,(d)paraffinvirtuallyremovedanddatasetbinarized,(e)medianfilteredand(f)largestobjectselectedandvoidsfilled.

medialmagniﬁcation.Previousimagingdataweremainlyobtained opticallyandrestrictedintwodimensionsshowingthenerve’ssur- faceorselectedcrosssections,orthethirddirectionofthedata setwashardlyresolvedbetterthan50␮m.Theperformance of imagingtechniquesincludingmicrocomputedtomographycanbe characterizedbythespatialanddensityresolution(Thurneretal., 2004).Whereastheaccessiblespatialresolutiononlydependson

theexperimentalsetupandthecomponentsincluded,thedensity resolution (contrast) cruciallydepends onthe specimenprepa- ration,i.e.dehydration,embedding,andstaining.CT-systemsfor animalexperimentswitharotatinggantry,asusedbyHopkinsetal.

(2015)areusuallynotstableenoughtoreachaspatialresolution below20␮m.Usingadvancedmicrocomputedtomographysys- temsbasedonnanofocusX-raytubes,suchasnanotom^®m,one

(6)

Fig.2. (a)Across-sectionalviewofaratsciaticnerve(green-coloredasterisk)insidethecollagenscaffold(magenta-coloredasterisk).Thenerve-scaffoldcomplexisembedded inparafﬁn(orange-coloredasterisk).Thelowestabsorbingelementinthepictureistheair(red-coloredasterisk).(b)Relatedsectionofthesamenervesubsequentlyscanned withouttheconduit.HigherX-rayabsorptionvaluesarerepresentedbylightergrayshades.

Fig.3.Athree-dimensionalviewofaselectednervespecimencomprisedoftwoparts(red-coloredasterisks)insideacollagenscaffold.Thespecimenwasdigitallysectioned, andtheparafﬁnwasmadetransparent.Thetwocomponentsvisiblearetheratsciaticnervetissue(gray)andthecollagenscaffold(white).TheNChasalengthof14mm andaninnerdiameterof1.5mm.

canachieveaspatialresolutionbelow1␮m.Theeffectivespatial resolution,however,alsodependsonthemaximaldiameterofthe specimenandthedetectorpixels(Holmeetal.,2014).Therefore,

theeffectivepixelsizeinthepresentstudywasrestrictedto2␮m basedonthespecimens’dimensions.Inordertoimproveacquisi- tiontimeandimagequalityandtoavoidsampledeformationasa

(7)

Fig.4.Thecross-sectionalareaalongthelongitudinaldirectionofthespecimen;(a)barenerveand(b)nerveinsidetheconduit.Discontinuitiesasindicatedbythearrows relatetostreakartifactsandaremorefrequentlyfoundattheNC.

Fig.5. Thethree-dimensionalimages(a)and(b)elucidatethatthevesseltreeisonlypartlyrevealedthroughaseriesofsegments.Thelargestconnectedcomponentin(a)has avolumeof1.9×10⁵␮m³andanestimatedmeandiameterandlengthof28and311␮mrespectively.Thefourselectedvirtualsections(i–iv),assignedtothepositionwithin thethree-dimensionalrepresentationofthesegmenteddata(b),showthatthevesselsegmentsarecharacterizedbyhighX-rayabsorbingspeciesgiveninwhite,asindicated forexamplebythearrowin(i).Thestreakartifacts,alsoappearingwhiteinthevirtualsections,canleadtofalsesegmentationandweredigitallymadesemi-transparentin thethree-dimensionalrepresentation.

resultofthermalirradiationfromtheX-raysource,themagniﬁca- tionwassetto25×,i.e.,aneffectivepixelsizeof4␮m.Thespatial resolutionwasestimatedcloseto6␮m(Thurneretal.,2004).This resultisanimprovementofanorderofmagnitudewithrespectto

thestudyofHopkinsetal.(2015)ineachofthethreeorthogonal directions.

Itshouldbenotedthatparafﬁnembeddinghasadvantageswith respecttoemployingcontrast-enhancingspeciessuchasLugol’s

(8)

iodine.TheemploymentofhighlyX-ray-absorbingspeciesneces- sitateshigheracceleratingvoltagesandgenerallyimpairstheimage quality.

Thecontrastofthe␮CTdatacollectedwithinthecurrentstudy issurprisingly high,assoft-tissuecomponentsgenerallyexhibit much lesscontrastthanbonytissues.Thisis theresultoflocal densityandcompositiondistributions.Here,theembeddingofthe nerveintoparaffin,amixtureofhydrocarbonswithadensityof about0.9g/cm³,isadvantageouscomparedtotheuseoffixation fluidwithadensityof1.0g/cm³.Itis,however,importanttoavoid confinedbubbles,whichgiverisetosevereartifacts.Therefore,sev- eraltrialstoimprovetheparaffinembeddingweretested,beforea reasonableresultcouldbefound.

Animportantachievementofthepresentstudyis thenerve visualizationinsidetheNC,whichisamajorchallenge,sincethe NCmaterialexhibitssigniﬁcantlyhigherX-rayabsorptionthanthe nervetissue.Theobtained␮CTdata,however,provethatboththe microanatomyofthenerveinsidetheconduitandtheporousstruc- tureofthescaffoldareaccessibleinareasonablemanner.

Giventhat ␮CT is anon-destructivetechnique, conventional histology can beperformed subsequently. In addition, contrar- ilytoimagingapproachesbasedoncontrastagents,thatrequire additional,time-consumingtissueprocessingsteps,thepresented approachrequireshardlyanyadaptationofthestandardhistolog- icalparafﬁn-embeddingprotocol.Eveninthebestcasescenario, wheretheiodinecontrastagentcanberemovedwithoutaffect- inghistology,stainingand de-stainingthesamplespriortoand afterimaging,respectively,needsatleastfouradditionaldaysof time.Thistimedelaycanbeconsiderablyincreased,sinceincuba- tiontimesareincreasingwithsampledimensions(Hopkinsetal., 2015).Here,the␮CTdatacanreadilysupporttheappropriateselec- tionofthedirectionofthehistologicalsectionsvirtuallycuttingthe tomographydata(Stalderetal.,2014)andsavingprecioustime.

Anatomicallandmarksguidetheorientationandallowforthepre- cisechoiceofthedesiredplaneofinterest.

Athree-dimensionaldatasetwithminimizedartifactsisapre- requisite for the successful application of segmentation tools.

Currentlythesegmentationisoftencarriedoutmanually,whichis time-consuminganderror-proneduetothethousandsofsections andthelargesizeofdataattheorderofseveralgigabytes.There- fore,anincreasingnumber ofresearchteamsemploycomputer softwareforsegmentationandanalysistasks,seee.g.,Longetal.

(2014)andIrshadetal.(2014).Theautomaticprocedureproposed inthepresentworkissufﬁcientlyaccuratetoextractgeometrical characteristicsofthenervemicroanatomyeveninsidethehigher X-rayabsorbingconduit.

Tounderlinethereliabilityofthesegmentation,fourselected sectionsweresegmentedmanuallyandcompared tothecorre- spondingautomaticresult.Themanualsegmentationincludedthe nervousandconnectivetissuesandwasbasedonvisualinspec- tion.Inclusionsofthenervoustissuewereconsideredinthemanual segmentationonlyifthickerthan50␮m.Thevoxelsegmentation resultedin6%falsepositiveand17%falsenegative.Thesegmented areainthecrosssectionshowsanerrorof11%.Thus,weassume anerrorof11%inthevolumemeasurement,aswell.

Thelengthofthecomputedcenterlinecanbeaffectedbythe uncertaintyofthecrosssectionsegmentationandtherelatederror barsaregiveninTable1.Theminimallengthwasdeterminedafter smoothingwithamovingaverageﬁlterwitha kernelsizeof10 voxelsandthemaximallengthwasdeterminedwithoutsmooth- ing.

Itiswellknown,thata higherelectronbeamcurrentwithin thesourcegivesrisetoahigherphotonﬂuxallowingforashorter acquisitiontimeduetoincreasedphotonstatistics,seee.g.Holme etal.(2014).Therefore,onetriestoworknearlythemaximallypos- siblebeamcurrent.Reducingtheelectronbeamcurrent,however,

onenotonlyextendsthelifetimeoftheﬁlament,butalsoobtains higherspatialresolutionbecauseofthesmallercross-over(beam spot).Thechoiceofthebeamcurrentis,therefore,anoptimiza- tiontask.Forourstudy,wehavefoundthat90%ofthemaximally possiblebeamcurrentguaranteesthenecessaryspatialresolution andphotonstatistics(contrast)atreasonableacquisitiontime.The acquisitiontimecanbeconsiderablyshortenedbymeansofaliq- uidjetanode(Bartelsetal.,2013).Suchanexperimentalsetupalso allowsphaseX-rayimaging,beneﬁcialforthethree-dimensional nervevisualizationwithincreasedcontrast.

Nonetheless, theobserved streakartifactsincrease the error barstoabout15%,avaluetobefurtheroptimized.Giventhatwe havealreadyoptimizedtheselectionoftheappropriatescanning parameters,samplepreparationitselfneedstobefurtherimproved towardsthatgoal.Avoidinghigh-absorbingspeciesandcracksthat canbepresentintheparafﬁnblock,aswellastrappedairbubbles, mainlyonthenervesurface,willgreatlyreducethepresenceof streakartifacts.Furthereffortsforreﬁningourestablishedproce- dureareunderprogress.

Thevascularnetworkisanessentialcomponentofhealthytis- sue,thereforetheconsiderationofvesselsinnervesisanimportant aspectoftheirevaluation.Inthisstudy,thefeature-basedfilterfor bloodvesselswassetuptodetecttubularmicrostructureswith adiameterstartingfrom8␮m, whichcorrespondstotwovoxel lengths.Theapplicationofthefilterenabledustodetectbloodves- selswitha diameterlargerthan12␮m.Nevertheless,theblood vesselsoftheformalin-fixatedandparaffin-embeddedtissuetend tocollapseandare,therefore,lessappropriatetoestimatethespa- tialresolutionreachedwithinthepresentstudy.Thevesseltreeis onlypartlyvisible.Probably,onlytheremainingbloodclotswere madevisible,whichshowrelativelyhighX-rayabsorptionbecause oftheiron present.Thesedatasetsalreadyprovideareasonable estimateofthevesseldensity.

5. Conclusions

Advancedconventional␮CTallowsthethree-dimensionalchar- acterizationofparafﬁn-embeddednerveseveninsideaconduitof signiﬁcantlyhigherX-rayabsorptionthanthetissue.Inspiteof occurringstreakartifacts,ourapproachemploysfullyautomatic toolsthatyieldreliableresults.Consequently,reproduciblenon- destructiveevaluationsofspecimenserieshavebecomepossiblein quantitativemannerwithinhours(ref.publication2).Themethod isfullycompatiblewithestablishedhistologytobeperformedsub- sequenttothetomographymeasurements.

Conﬂictsofinterest

Theauthorsdeclarenoconﬂictofinterest.

Acknowledgements

TheauthorsthankStephanFrankandJürgenHenchfromthe NeuropathologyDepartment,BaselUniversityHospital,forallow- ingtheuseofhistologicalequipment,aswellasSaschaMartinand StefanGentschfromtheMechanicsShop,PhysicsDepartment,Uni- versityofBaselforprovidingthesampleholders.Theprojectwas supportedbySNSFprojects144535and133802.

References

Bartels,M.,Hernandez,V.H.,Krenkel,M.,Moser,T.,Salditt,T.,2013.Phasecontrast tomographyofthemousecochleaatmicrofocusX-raysources.Appl.Phys.

Lett.103,083703,http://dx.doi.org/10.1063/1.4818737.

Bertelli,J.A.,Taleb,M.,Saadi,A.,Mira,J.C.,Pecot-Dechavassine,M.,1995.Therat brachialplexusanditsterminalbranches:anexperimentalmodelforthe studyofperipheralnerveregeneration.Microsurgery16,77–85.

(9)

Frangi,A.F.,Niessen,W.J.,Vincken,K.L.,Viergever,M.A.,1998.MultiscaleVessel EnhancementFiltering.SpringerBerlinHeidelberg,Berlin,Heidelberg,pp.

130–137,http://dx.doi.org/10.1007/BFb0056195.

Godinho,M.J.,Teh,L.,Pollett,M.A.,Goodman,D.,Hodgetts,S.I.,Sweetman,I., Walters,M.,Verhaagen,J.,Plant,G.W.,Harvey,A.R.,2013.

Immunohistochemical,ultrastructuralandfunctionalanalysisofaxonal regenerationthroughperipheralnervegraftscontainingSchwanncells expressingBDNF,CNTForNT3.PLOSONE8,e69987,http://dx.doi.org/10.

1371/journal.pone.0069987.

Hieber,S.E.,Bikis,C.,Khimchenko,A.,Schweighauser,G.,Hench,J.,Chicherova,N., Schulz,G.,Müller,B.,2016.Tomographicbrainimagingwithnucleolardetail andautomaticcellcounting.Sci.Rep.6,32156.

Holme,M.N.,Schulz,G.,Deyhle,H.,Weitkamp,T.,Beckmann,F.,Lobrinus,J.A., Rikhtegar,F.,Kurtcuoglu,V.,Zanette,I.,Saxer,T.,Müller,B.,2014.

ComplementaryX-raytomographytechniquesforhistology-validated3D imagingofsoftandhardtissuesusingplaque-containingbloodvesselsas examples.Nat.Protocols9,1401–1415,http://dx.doi.org/10.1038/nprot.2014.

091.

Hopkins,T.M.,Heilman,A.M.,Liggett,J.A.,LaSance,K.,Little,K.J.,Hom,D.B., Minteer,D.M.,Marra,K.G.,Pixley,S.K.,2015.Combiningmicro-computed tomographywithhistologytoanalyzebiomedicalimplantsforperipheral nerverepair.J.Neurosci.Methods255,122–130.

Irshad,H.,Veillard,A.,Roux,L.,Racoceanu,D.,2014.Methodsfornucleidetection, segmentation,andclassiﬁcationindigitalhistopathology:areview-current statusandfuturepotential.IEEERev.Biomed.Eng.7,97–114.

Kerschensteiner,M.,Schwab,M.E.,Lichtman,J.W.,Misgeld,T.,2005.Invivo imagingofaxonaldegenerationandregenerationintheinjuredspinalcord.

Nat.Med.11,572–577.

Khimchenko,A.,Deyhle,H.,Schulz,G.,Schweighauser,G.,Hench,J.,Chicherova,N., Bikis,C.,Hieber,S.E.,Müller,B.,2016.Extendingtwo-dimensionalhistology intothethirddimensionthroughconventionalmicrocomputedtomography.

NeuroImage139,26–36.

Kroon,D.J.,2009.HessianBasedFrangiVesselnessFilter.http://www.mathworks.

com/matlabcentral/ﬁleexchange/24409-hessian-based-frangi-vesselness- ﬁlter.

Long,H.Q.,Xie,W.H.,Chen,W.L.,Xie,W.L.,Xu,J.H.,Hu,Y.,2014.Valueofmicro-CT formonitoringspinalmicrovascularchangesafterchronicspinalcord compression.Int.J.Mol.Sci.15,12061–12073.

Madduri,S.,Feldman,K.,Tervoort,T.,Papaloïzos,M.,Gander,B.,2010a.Collagen nerveconduitsreleasingtheneurotrophicfactorsGDNFandNGF.J.Controlled Release143,168–174,

www.sciencedirect.com/science/article/pii/S0168365909008621.

Madduri,S.,diSumma,P.,Papaloïzos,M.,Kalbermatten,D.,Gander,B.,2010b.

Effectofcontrolledco-deliveryofsynergisticneurotrophicfactorsonearly nerveregenerationinrats.Biomaterials31,8402–8409,

www.sciencedirect.com/science/article/pii/S0142961210008951.

Müller,B.,Beckmann,F.,Huser,M.,Maspero,F.,Székely,G.,Rufﬁeux,K.,Thurner, P.,Wintermantel,E.,2002.Non-destructivethree-dimensionalevaluationofa polymerspongebymicro-tomographyusingsynchrotronradiation.Biomol.

Eng.19,73–78.

Müller,B.,Deyhle,H.,Lang,S.,Schulz,G.,Bormann,T.,Fierz,F.C.,Hieber,S.E.,2012.

Three-dimensionalregistrationoftomographydataforquantiﬁcationin biomaterialsscience.Int.J.Mater.Res.103,242–249,http://dx.doi.org/10.

3139/146.110663.

Otsu,N.,1979.Athresholdselectionmethodfromgray-levelhistograms.IEEE Trans.Syst.ManCybern.9,62–66.,http://dx.doi.org/10.1109/tsmc.1979.

4310076.

Schneider,C.A.,Rasband,W.S.,Eliceiri,K.W.,2012.NIHImagetoImageJ:25years ofimageanalysis.Nat.Methods9,671–675,http://dx.doi.org/10.1038/nmeth.

2089.

Stalder,A.K.,Ilgenstein,B.,Chicherova,N.,Deyhle,H.,Beckmann,F.,Müller,B., Hieber,S.E.,2014.Combineduseofmicrocomputedtomographyandhistology toevaluatetheregenerativecapacityofbonegraftingmaterials.Int.J.Mater.

Res.105,679–691,http://dx.doi.org/10.3139/146.111050.

Thurner,P.,Beckmann,F.,Müller,B.,2004.Anoptimizationprocedureforspatial anddensityresolutioninhardX-raymicro-computedtomography.Nuclear Instrum.MethodsPhys.Res.B225,599–603http://www.sciencedirect.com/

science/article/pii/S0168583X04007980.