The author(s) shown below used Federal funds provided by the U.S.

Department of Justice and prepared the following final report:

Document Title:

Computerized Reconstruction of Fragmentary

Skeletal Remains for Purposes of Extracting
Osteometric Measurements and Estimating MNI

Author(s):

Mohamed R. Mahfouz, Natalie R. Langley,

Nicholas Herrmann, Emam ElHak Abdel Fatah

Document No.:

249948

Date Received:

June 2016

Award Number:

2011-DN-BX-K537

This report has not been published by the U.S. Department of Justice.
To provide better customer service, NCJRS has made this federally
funded grant report available electronically.

Opinions or points of view expressed are those

of the author(s) and do not necessarily reflect
the official position or policies of the U.S.
Department of Justice.


FinalReport
NationalInstituteofJustice
GrantNumber:2011DNBXK537
Title:ComputerizedReconstructionofFragmentarySkeletalRemainsforPurposesofExtracting
OsteometricMeasurementsandEstimatingMNI

PrincipalInvestigator:MohamedR.Mahfouz
Professor,DepartmentofMechanical,AerospaceandBiomedicalEngineering
CoCenterDirector,CenterofMusculoskeletalResearch
UniversityofTennessee
307PerkinsHall
Knoxville,TN37996
Telephone:(865)9747668
Fax:(865)9746394
Email:mmahfouz@utk.edu

CoPrincipalInvestigator:NatalieR.Langley
LincolnMemorialUniversity
Email:Natalie.Langley@lmunet.edu

CoPrincipalInvestigator:NicholasHerrmann
MississippiStateUniversity
Email:Herrmann@anthro.msstate.edu

PostDoctoralResearcher:EmamElHakAbdelFatah
UniversityofTennessee
Email:ealiabde@utk.edu

GrantStartDate:January1,2012
GrantEndDate:June31,2015

RecipientOrganization:UniversityofTennesseeOfficeofResearch,1534WhiteAve,Knoxville,TN37996
DUNSNumber:003387891
EmployerIdentificationNumber(EIN):626001636

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

1. Abstract
Thepurposeofthisresearchgrantistoimproveforensicanthropologypracticeandpolicybyfacilitating
morecompleteandaccurateanalysesoffragmentaryhumanskeletalremains.Throughthetermofthis
grant we developed a userfriendly software that enables forensic anthropologists to quantify and
reconstruct fragmentary human skeletal remains (crania, pelves, humeri, and femora) from three
dimensionalsurfacefilesgeneratedbycomputedtomographyorlaserscans.First,thesystemservesas
an osteological case or scene management tool. All scanned skeletal remains from each scene can be
reviewedwithintheapplication.Thesoftwareallowstheuserabilitytovisualizefragmentaryelements,
rejectoracceptelements.Thisisthenfollowedbyinitialalignmentoffragmentaryelementswiththree
dimensional statistical template for each bone. Upon accepting the alignment of the elements, the
software merges the elements and provides a fully reconstructed bone; measurements can then be
conductedbytheusertoapplytoregressionequations,discriminantfunctions,ortousewithsoftware
suchasFordisc3.0.
Duringthecourseofthisgrant,atotalof24,569fragmentsweresortedfromtheMortonShellMound
fromthegulfcoastofLouisiana,18,373fragmentswerecoded,1232fragmentswerewasdigitizedand
2061fragmentswereCTScanned..Thedevelopedsoftwarewasvalidatedusingbothsimulateddata,and
realdatafromMortoncollection.SoftwarewascomparedtoGIS,samefragmentIDswerematchedusing
both systems and fragments was placed in comparable location on the template. The reconstruction
resultsonsimulateddatawaslessthan2mmmeanRMSerrorforpelvis,skull,humerusandfemur.The
developedsystem generatedsimilarresultswhen compared to GISlookingatboth theIDofmatched
fragmentsandlocationofmatchedfragmentsonthetemplate.

2. Introduction
Withinthelargermedicolegalsystem,forensicanthropologistsperformtheessentialtaskofcreatinga
biologicalprofiletoaidlawenforcementinidentifyingunknownhumanremainsanimportantfirststep
in the criminal apprehension and conviction process. The four primary components of the biological
profileareage,sex,ancestry,andstature.Theproposedprojectfocusesonthenumberofindividuals,sex
andancestryestimation,andmeasurementextractionforuseinstatureestimation.Theparametersof
thebiologicalprofileareintricatelyinterwoveninthat,frequently,onecomponentisnecessaryinorder
to make precise determinations about other component(s). In cases of mass disasters or commingled
remains,thedeterminationofindividualbiologicalprofileparametersiscomplicatedbythepresenceof
multipleunassociatedelements.Theabilitytomakebiologicalprofileassessmentsonisolatedbonesor
bonefragmentsiscritical.Theproposedapplicationprovidesasystemtoperformsuchanalysesandto
managecomplexmassdisasterorcommingledbonecases.
Commingled human remains represent a logistical and methodological challenge to forensic
anthropologists.Incaseswherethematerialishighlyfragmentedthechallengeincreasesexponentially.
RecentworkoncommingledhumanremainsbyAdamsandByrd(2008)[1]attemptstoprovideaseries
ofcasesexamplesandstudiesofmethodstoaddresscommingledremainsinforensiccontexts.In2011,
the Scientific Working Group for Forensic Anthropology (SWGANTH) approved draft guidelines for
resolving commingled human remains (http://swganth.startlogic.com/Commingling%20Rev0.pdf). Key
pointsoftheSWGANTHguidelinestargetthedeterminationoftheminimumnumberofindividuals(MNI),
the assessment of the most likely number of individuals (MLNI or the Lincoln Index), and the use of

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

scientificmethodstoachievethesegoals.Theguidelinesrecommendelementreconstruction,visualpair
matching,osteometricevaluation,andtaphonomiccomparisonsintheassessmentofMNIandMLNI.The
applicationandmethodproposedinthisprojectwouldaddressallthesepointsandprovideasystematic
andquantifiableevaluationofeachbonefragment.
Althoughdevelopedindependently,thethreedimensionalapproachtothequantificationofcommingled
remainsisalogicalextensionofcodingandtwodimensionalmethodsdevelopedinzooarchaeology[2]
and bioarchaeology [3][4]. Recent work by Herrmann and Devlin (2008)[4] quantify small fragmented
remains into an Osteological Information System (OIS) using Geographic Information System (GIS)
softwaretoderiveminimumnumberofelements(MNE)valuesandMNIestimates.Thesesystemsare
time consuming and are dependent on the observer to manually digitize each fragment into the OIS
application.TheresultingimageprovidesanMNEestimatefortheelementunderinvestigation.
By and large, stature estimation is the only element of the biological profile that has been given due
attention with respect to fragmentary remains [5][6][7]. Most of these studies developed population
specificdiscriminantfunctionsfromaseriesoflongbonemeasurements.However,somestudiesusedan
indirect method of first reconstructing long bone length and then estimating stature [5][8][9]. In
paleoanthropology,however,theusefulnessofcomputermodelingforfossilreconstructionshasbeen
recognizedasamathematicallyrigorousprocess[10][11][12].ReconstructionsoftheAL2881(Lucy)
femurbymembersofourresearchteamprovedtobecloseapproximationsoforiginalreconstructions
(Sylvesteretal.2008)[12].Furthermore,validationtestsonfemoraofknownlengthfromtheWilliamBass
DonatedSkeletalCollection(WBDSC)resultedinreconstructionswitherrorsoflessthan1mm(Sylvester
etal.2008)[12].Thesereconstructionsweredoneusingpreviouslydevelopedstatisticalboneatlasesto
reconstructthefragmentaryremains[13][14][15].

3. Methods
Thepurposeofthisresearchisdesignedtoworkonfragmentsoffourskeletalelements,femur,humerus,
pelvisandskullbones.Thefirststepinvolvesgenerationofsurfacesmodelsforthefragmentaryelements.
ThiscanbedoneusingCTorlaserscanner.InthecaseofCTscanning,asegmentationstepisperformed
togeneratethesurfacemodels.Uponcreationofthefragmentaryelementsmodels,thenextstepisto
extract features from each bone fragment by measuring surface roughness. Surfaces points with
roughnessabovecertainspecifiedthresholdvaluesaredenotedasfeaturepoints.Amultistagetechnique
isthenusedtoidentify,matchandregisterthesebonefragmentstotheircorrespondingtemplatebone
model.Foreachbonefragment,thefeaturesareusedtobematchedwithtemplatebonefeaturesusing
iterativeclosestpoint(ICP)algorithm.AslongasithasbeenproventhatICPalgorithmconvergetoalocal
minimumwithrespecttothemeansquaredistancemetric,whichmeansthatitguaranteesthecorrect
registrationgiventhatthetwoshapesaresomehowcloseintermsofpositionandorientation,weused
differentcombinationsofpositionandorientationofbonefragmentrelativetothetemplatebone.For
each combination, we applied ICP to try to match and register each fragment features with the four
templatebonesfeatures.Thebestmatchscore,intermsofminimumrootmeansquareerror,isused
along with the position/orientation and transformation to register the bone fragment with the
correspondingtemplatemodel.
Werefertotemplatemodelhereasanaveragemoldthatcapturestheprimaryshapeinformationofa
skeletalelement.Threedimensionalstatisticalboneatlasesareconstructedfromlargedatasetsof
bones[13][14][15].Theatlasguaranteespointtopointandsurfacecorrespondenceacrosstheentire
sample,whichallowsforanaccuratecalculationofanaveragetemplate.Thistemplatebonecapturesthe

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

globalshapecharacteristicsforeachboneacrossanentirepopulationwhichcanthenbeusedtoguide
the placement of the fragmentary pieces into anatomically correct space and to estimate missing
fragments.
Thepresentedworkisembeddedina3Dvirtualenvironmentwheretheuserloadssurfacemodels,starts
theprocedureandviewsthefinalresults.Theusercanacceptorrejectthematchingandregistration
resultsafterrunningtheprocedure,aswellasshe/hecanmanipulateeachmodelseparatelyeitherby
handorthroughGUIcontrolsandstarttheprocedureagain.
Figure 1 highlights the proposed framework where fragmentary pieces are first CT scanned and
segmented to generate surface models. The resultant surface models are then used to extract
fragmentarybonesfeatures,followedbymatchingandregistration.Thevisualfeedbackblockpresents
userinteractionsthroughdeveloped3Dvirtualenvironment.

Figure1Softwareframework

3.1. TemplateModel
Astatisticalshapeatlas,SSA,isanaveragemold,oratemplatemesh,thatcapturestheprimaryshape
variationofabone(Figure2)[13],alongwithitsstatisticalmodesofvariations.Aproposedmethodfor
creating a SSA and generating dense correspondence across populations for complex anatomical
structureswasdevelopedin[13].Figure3showstheprocessofaddinganewbonetothestatisticalatlas.
Inthepresentstudy,weusedfemur,humerus,pelvis,andskulltemplatebonesurfacemodels.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure2Pelvisstatisticalshapeatlas

Figure3Processofaddinganewbonetostatisticalatlas

3.2. SurfaceModelGeneration
Figure4showsthegenerationprocessofbonessurfacemodels.ThebonefragmentsarefirstCTscanned
voxelresolution.TheDICOMimagesacquiredfromtheCTscansare
using0.625 0.625 0.625
thenmanuallysegmented,andsurfacemodelsaregeneratedthough3Dreconstruction.Figure5shows
anexampleofthesurfacemodelsgenerationprocess,wheretheCTimageissegmentedtocreateabinary
regionofinside/outsidethebone,whichisusedforgeneratingthe3Dsurfacemodel.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure4Surfacemodelsgenerationprocess

Figure5Exampleofsurfacemodelsgenerationprocess

3.3. FeatureExtraction
Figure6highlightsthestepsoffeatureextractionprocess.

Figure6Featureextractionprocess

3.3.1. MeshDifferentialProperties
Different methods were proposed for approximation of the surface differential properties. A valuable
evaluationofdifferentcurvatureestimationcanbefoundin[16].TheevaluationshowedthattheGauss
Bonett scheme gives the best results for estimation of overall curvature. Consequently, GaussBonett
schemewaschosenasthemethodforestimationofsurfacedifferentialproperties.Belowarethedetails
of the calculation of differentials properties using a discrete GaussBonett scheme. Figure 7 shows the
neighborhoodnotationusedinGaussBonettscheme.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure7NeighborhoodnotationofGaussBonettscheme
ThefollowingstepsareusedforcalculatingmeshdifferentialpropertiesusingGaussBonettscheme.
1.
2.
3.
4.

Facenormal,Eq.1.
Weightedvertexnormal,Eq.2.
Gaussianandmeancurvature,Eq.3,Eq.4.
Maximumandminimumcurvature,Eq.5,Eq.6.

.
2

/3

Eq.1

Eq.2

,
0.25

/3

Eq.3

Eq.4

max 0,

Eq.5

max 0,

Eq.6

where isthedeviationofnormal .

3.3.2. SurfaceRoughness
Forthepresentapplication,curvednessshowedthebestresultsforpresentinglocalshapevariations.We
haveproposedadifferentmultiscalesurfaceroughnessmeasurebasedonweightingthevarianceofthe
pointsneighborscurvature.Thenewlyproposedmeasure,Eq.7,isbasedonthecurvedness,anditforms
theweightsbasedonthepointtoneighborsedgelengths.

1
| |

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Eq.7

1
| |

Eq.8

Where

Roughnessatpoint
Averagecurvednessoftheneighborsofpoint
Curvednessatneighbor
thelengthoftheedgebetween ,
totaledgelengthwhichisthesummationofthe
edgelengthsofthepoint sneighbors

Theproposedmeasurecanbecomputedatmultiplescalesbyincorporating levelneighborhoodpoints
intheequation, 1,2,Figure8Figure12showroughnessmapsoffemur,humerus,pelvisandskull
templatebonescomputedat1st,2nd,3rd,4thand5thneighborhoodlevels.

Figure8Roughnessmapsoffemurtemplateboneat1st,2nd,3rd,4thand5thneighborhoodlevels

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure9Roughnessmapsofhumerustemplateboneat1st,2nd,3rd,4thand5thneighborhoodlevels

Figure10Roughnessmapsofpelvistemplateboneat1st,2nd,3rd,4thand5thneighborhoodlevels

Figure11:Roughnessmapsofskulltemplateboneat1st,2nd,3rd,4thand5thneighborhoodlevels

3.3.3. GaussianMixtureModel
GaussianmixturemodelisaparametricmodelusedforclusteringdatainwhichanalyticGaussiandensity
functionsarefittedtothedataassumingthedataweredrawnfromanumberofGaussiandistributions.
Fortheprobleminhand,wereassumingthecalculatedsurfaceroughnessofeachtemplatemodelisa
mixtureoftwoGaussiandistributions,thatrepresentsdistributionofsurfacepointswithlowroughness
values(smoothsurfacepoints)andhighroughnessvalues(featurepoints).Figure12showthehistogram

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

and the two distributions of Gaussian components of femur and skull template bone models surface
roughnesscalculatedusingtheproposedroughnessmeasure,Eq.7,Eq.8.

Figure12Surfaceroughnessoffemur(left),skull(right)templatebonemodel,histogramand
distributionsofGaussiancomponents
Basedontheideathatfracturedsurfacesintroducehighervaluesofroughness,comparedwithnative
surfaces,weassumedthatthecalculatedsurfaceroughnessofeachbonefragmentisamixtureofthree
Gaussiandistributions,representingnativesurfacesmoothpoints,nativesurfacefeaturesandfracture
surfacespoints.Figure13showsthesurfaceroughnessoffemurfragmentbonemodelhistogramand
distributionsofGaussiancomponentsofitssurfaceroughnessvalues.Itisworthtonoteherethatthe
featuresoftheoriginalbonestillcanbedetectedbytheproposedsurfaceroughnessmeasure,inthe
presenceofhighcurvature,orhighrough,pointsattheedgeofthefracturesurface.

Figure13Surfaceroughnessoffemurfragmentbonemodel
Weve selected the mean of the second Gaussian distribution , resulting from GMM analysis as the
thresholdthatfilteroutthefeaturesfromtemplatebonesurfacemodels.Toextractsimilarpointsfrom
thefragmentbone,weveselectedtherangebetweenthesecond, ,andthethird, ,meansasthe
rangeofroughnessthatincludethefeaturesoftheoriginalbone.Figure14showtheextractedfeatures
fromtemplateandfragmentfemurbonemodelsusingGMMextractedthresholds.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Figure14Featuresdetectedonfemurtemplatemodel(left)andfemurfragment(right)

3.3.4. MatchingandRegistration
TheprocessofmatchingandregistrationishighlightedinFigure15andFigure16.Basicallytheprocess
isdividedintotwostages.Thefirststageinvolvesmatchingoftheextractedfragmentfeatureswiththe

.Fragmentfeaturesarepositionedatdifferentlocations
featuresofthefourtemplatemodels
withrespecttotheboundingboxofthetemplatemodel,andateachposition,theyarerotatedaround
.Foreachpositionandorientation,fragmentfeaturesarematched
differentcoordinatesystemaxes,
withtemplatefeaturesusingiterativeclosestpoint(ICP).

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure15MatchingandregistrationprocessA
Fromallmatchings,weselectthebestmatchingscore
, ,
basedontheminimumrootmean
andextractitsrigidtransformation, .Thesecondstageinvolvesregistrationofthe
squareerror,
fragmentsurfacemodelwiththetemplatesurfacemodelbasedonthebestmatchingscorefromthefirst
stage. The fragment model is first positioned and oriented using ( ,
, transformed by and then
registeredwiththetemplatemodel usingICP.Iftheoutputerrorfromtheregistrationprocess
is
,thenwearefacingafalsepositive
greaterthantheoutputerrorfromthefeaturematchingprocess
matching.Tosolvethisissue,wediscardthetemplatemodelwiththemaxfeaturematchingfromthe
matchingprocessandusedthesecondbestfeaturematchingscoretemplatemodeltoregisterwiththe
fragment.Iftheregistrationerrorissmallerthanthefeaturematchingerror,thenwearedonewiththe
matching,registrationandidentificationtoo.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure16MatchingandregistrationprocessB

3.3.5. VirtualEnvironment
Amajorcomponentofthisgrantwastodevelopavirtual3DenvironmentbasedonOpen Graphics
Library (OpenGL) 3D visualization library [https://www.opengl.org/] and QT graphical user interface
library.Thevirtualenviromentencapsulateboththeautomaticalignmentfunctionalitiesandthemanual
registration and manipulation functionalities. The developed environment enables the user, through
graphicaluserinterface(GUI),toloadscannedmodelsineithersurfacemodelfileformat(*.surf)oropen
inventor file format (*.iv). The GUI has the capability of loading one or multiple files at once each
containingoneormoremodels.Oncethemodelsloadedtothesceneviewer,theuserhastheabilityto:
1. Dorotation,panningandzoomingthroughinteractiveviewerwidgetaswellaswheelcontrols.
2. Returntothehomeview,setthehomeview,viewalltheloadedmodels.
3. Switchbetweenparallelandperspectiveprojectionmodes.
4. Changethelightingdirection.
5. Changebackgroundcolor.
6. Selectspecificmodelforfurtheroperation.

Theloadedmodelsareassignedrandomkeysaswellasrandomcolorswhichcanbeupdatedthrough
material editboxes. The user has the ability to apply transformation (translation, rotation and scaling)
throughusercontrolsaswellasinteractiveviewer.Thetransformationscanbeapplied,resetorconfirmed
fortheselectedmodel(s).Theinterfaceenablestheusertosavethemodifiedmodelsseparately.Figure
17andFigure18showexamplesoftwoscreensfromthevirtualenviroment,AppendixAcontainsamore
indepthscenarioofusingthedevelopedapplication.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure17Fragmentmodelsloadedinthedevelopedvirtualenvironment

Figure18Alignfragmentswithtemplatemodel
The GUI also provides the user the control to generate and export surface roughness values for the
selectedmodel(s)aswellasallloadedmodelsusingvariableneighborhoodsize,differentlocalcurvature
measures (mean curvature,shapeindex, curvednessandsigned curvedness)and differentscalespace
calculationscheme(differencebased,variancebased,edgeweightedbased).Forourapplication,after
loadingthefragmentmodelsinthescenetotheviewer,theusercanselectdifferentfragmentsthrough
selecttool,rightclickandselectalignselectedmodelswithtemplate.Thiswillpopupasmalldialogto
choosewhichtemplatemodeltoalignwith,ortheusercancheckAlltotestalignmentwithalltemplate
bones.Onceaccepted,aseparatealignmentdialogwillappearshowingtheselectedfragmentsandthe
templatemodel(s)loaded.
In the alignment dialog, the user has the ability to apply any transformation to the loaded fragment
models as well as to the template model before starting the algorithm. Default values for maximum
iterationsandminimumrelativeerrorofICPalgorithmareloadedandcanbechangedbytheuser.The
dialogenablestheusertoeitherstartthematching/registrationprocedurefortheselectedmodel(s)or
forall.Oncedone,theGUIwilldisplaythefragmentmodelsregisteredwiththepropertemplatebone.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

The user can accept/reject matching(s)/alignment(s) by right clicking at the fragment and selecting
accept/rejectforselectedorallfragments.Ifthealignmenthasbeenaccepted,itwillbereflectedtothe
fragment model in the main viewer, and it will be neglected if rejected. Again, the user can apply
transformationstothefragment(s)ofrejectedmatching(s)/alignment(s)andstarttheprocedureagain.

4. Results
ResultsofthefeaturedetectionontemplatemodelsforthefourbonesisshowninFigure19.Asubsetof
Mortoncollectionfragmentsaswellasasubsetofsimulatedfragmentswereusedtoverifythedeveloped
matchingalgorithmasshowninFigure20andFigure21.Foreachfragment,matchingscoresagainstall
thetemplatebonesatdifferentpositionsandorientations,werelisted,basedontherootmeansquare
error,wheretherootmeansquareerroriscalculatedincentimetersastheaverageofdistancesbetween
correspondingpoints.Thetemplatebonewiththeminimumnonfalsepositiverootmeansquareerror
wasconsideredthebestmatch,andthefragmentwasidentifiedasbeingpartofit.

Figure19Templatemodels,theirroughnessmapsanddetectedfeatures

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure20Mortoncollectionsampleusedforvalidatingbonematchingalgorithm

Figure21Exampleofsimulatedpartialfragmentsusedforvalidatingbonematchingalgorithm
Table1showsthematchingandregistrationrootmeansquareerrors(RMSE)witheachtemplatemodel
atdifferentorientationsandpositionsandtheautomaticallymatchedbone.Redcolorshighlightfalse
positive matchings which are detected by comparing feature matching errors and registration errors.
Greencolorshighlightcorrecttruepositivematchingswhichareverifiedbycomparingfeaturematching
errors and registration errors. Figure 22 shows results of matching fragmentary pieces from Morton
collectionusingdevelopedsoftware.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table1MatchingalgorithmvalidationandminimumRMSE(cm)greencolorhighlightsacorrect
matching,whileredcolorhighlightsanincorrectmatching
Fragment

Femur

Humerus

Pelvis

Skull

MatchedBone

0.73328

0.514663

0.699231

0.385311

Humerus

0.667225

0.557928

0.783834

0.455202

Humerus

0.850877

0.640472

0.898699

0.698427

Humerus

0.634426

0.664907

0.699361

0.753343

Humerus

0.653042

0.563865

0.781884

0.840516

Humerus

0.733875

0.619259

0.776236

0.519864

Femur

0.79863

0.608945

0.823335

0.625949

Femur

0.716187

0.573982

0.829414

0.549071

Humerus

0.845232

1.296698

4.058093

2.949372

Femur

10

0.932372

1.024249

1.250394

0.883769

Femur

11

0.486868

0.499618

0.602452

0.472178

Femur

12

0.739734

0.755775

0.954680

0.414900

Femur

Figure22MatchingoffragmentarypiecesfromMortoncollectionusingdevelopedsoftware

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Validationofthebonereconstructionwasconductedonsimulateddatatoassesstheabilityofstatistical
atlastoreconstructmissinginformationfromexistingfragments.Usingcorrespondenceofstatistical
atlases,differentfragmentsofthepelvisweredeletedfrom120samplesrepresenting10,20and30%of
theoverallsurfacearea.Foreachsamplethestatisticalshapemodel(SSM)wasconstructedusingthe
leaveoneoutmethod.Foreachoutputmodelfromtheatlas,thismodelwasexcludedfromthetraining
setandtheremainingboneswereusedtocreatetheSSM.Figure23showsthesimulatedfragmentsand
theaveragesurfacedistanceerrormapofreconstructionforthe120pelvisbones.Table2showthe
mean,standarddeviationandmaximumRMSerrorforthereconstructionforthe120pelvis.Figure24
showstheaveragedistanceerrorofreconstructionfor90skullsusingleaveoneoutexperiment,
Table3showsthemean,standarddeviationandmaximumRMSerrorforthereconstructionusingleave
onoutexperimentfor90skull.

Figure23Fullbonereconstructionvalidationforpelvis

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table2Resultsoffullbonereconstructionvalidationforpelvis

%Missing
10%

RMS(mm)
0.54

STD(mm)
0.35

Max(mm)
2.91

20%

0.6

0.4

3.2

30%

0.61

0.41

3.29

Figure24Fullbonereconstructionvalidationforskull

Table3Resultsforfullbonereconstructionvalidationforskull

%Missing

RMS(mm) STD(mm)

Max(mm)

30%

0.45

4.8

0.35

Figure25showstheresultsofreconstructionofsimulatedskullfragmentusingdevelopedsoftwarewith
meanRMSerrorof0.5mmandmaximumerrorof1.1mm.Figure26toFigure29showtheresultsof
comparisonoffragmentsmatchingandfullbonereconstructionusingdevelopedsoftwareandGIS
system.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure25Distancemapoftheerrorinthereconstructedskullusingthesoftware(simulated
fragmentaryskull)

Figure26ValidationofreconstructedbonefromMortoncollectionusingdevelopedsoftware(left),
distancemapbetweenreconstructed3DmodelandfragmentarypiecesandGIS(right)

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure27ValidationofreconstructedbonefromMortoncollectionusingdevelopedsoftware(left),
distancemapbetweenreconstructed3DmodelandfragmentarypiecesandGIS(right)

Figure28ValidationofreconstructedbonefromMortoncollectionusingdevelopedsoftware(left),
distancemapbetweenreconstructed3DmodelandfragmentarypiecesandGIS(right)

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


Figure29ValidationofreconstructedbonefromMortoncollectionusingdevelopedsoftware(left),
distancemapbetweenreconstructed3DmodelandfragmentarypiecesandGIS(right)

Table4showspercentagecoverageofMortoncollectionelementsusedinthevalidationstudy.Table5
showsthenumberoffragmentssorted,coded,digitizedandCTscannedfromtheMortoncollection
duringthecourseofthegrant.
Table4ElementsreconstructedfromMortoncollectioninthevalidationexperimentandtheirelement
coverage%value
FEM
FEM
FEM
FEM
FEM
FEM
FEM
FEM
FEM
FEM
HUM
HUM
HUM
HUM
HUM
HUM
HUM
HUM
HUM
HUM

0.442519627
0.369916495
0.199869862
0.421780743
0.037546688
0.173381003
0.078867841
0.337983823
0.250040714
0.218311589
0.340868503
0.140539778
0.719475515
0.268089708
0.119372121
0.251810742
0.43019809
0.753070155
0.160353832
0.048746059

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table5NumberofMortoncollectionfragmentssorted,coded,digitizedandscannedduringthecourse
ofthegrant
FragmentSorted
FragmentsCoded
FragmentsDigitized
FragmentsCTscanned

24,569
18,373
1232
2061

5. Conclusion
Duringthecourseofthisgrant,userfriendlysoftwarewasdevelopedthatenableforensic
anthropologiststoquantifyandreconstructfragmentaryhumanskeletalremains(crania,pelvis,humeri,
andfemoraarethefocusoftheinitialplatform;additionalelementswillbeaddedwithsubsequent
releasesofthesoftware)fromthreedimensionalsurfacefilesgeneratedbycomputedtomographyor
laserscans.First,thesystemservesasanosteologicalcaseorscenemanagementtool.Allscanned
skeletalremainsfromeachscenewillbereviewablewithintheapplication.Thesoftwareallowstheuser
abilitytovisualizefragmentaryelementsandrejectoracceptelements.Thisisthenfollowedbyinitial
alignmentoffragmentaryelementswiththreedimensionalstatisticaltemplatesforeachbone.Upon
acceptingthealignmentoftheelements,thesoftwaremergestheelementsandprovidesafully
reconstructedbone.Measurementscanthenbeconductedbytheusertoapplytoregressionequations,
discriminantfunctions,ortousewithsoftwaresuchasFordisc3.0.Thedevelopedsoftwarewas
validatedusingbothsimulateddata,andrealdatafromMortoncollection.Softwarewascomparedto
GIS,samefragmentIDswerematchedusingbothsystemsandfragmentswasplacedincomparable
locationonthetemplate.ItisworthnotingthatGISdoesntprovideareconstructedboneorthree
dimensionaldistancemapbetweenmatchedfragmentsandtemplateandthusnoquantitativenumbers
weregeneratedforthecomparison.
Thedevelopedprojectexpandstheusefulnessofthestatisticalatlasesintotheanalysisoffragmentary
andcommingledremains.Inaddition,forensicanthropologistsareprovidedwithameanstoquantifyand
reconstructremainsthataredamagedorfragmentary,therebyenhancinganalysesinchallengingcases.
Thedatamanagementaspectoftheapplicationwillallowforensicanthropologiststodigitallyinventory
complexcommingledscenes;ifgeospatialdataisintegratedwitheachfragmentthentherefittingprocess
can proceed geographically. Therefore, the developed application will significantly impact forensic
anthropologists and crime scene investigators ability to reconstruct mass disasters, commingled mass
graves,andhighlyfragmentaryindividualburialsorsurfacescatters.
InadditionduringthecourseofthegranttheresearchteamhasCTscanned,sorted,codedanddigitized
largepercentageofMortoncollection.

6. FutureWork
WiththeBetareleaseofthesoftwareanditsusebythecommunity,supportwillbeprovidedforbugfixes
anduserinterfaceenhancementsbasedonusersfeedback.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

7. References
[1]

[2]

[3]

[4]

[5]
[6]
[7]
[8]
[9]
[10]

[11]
[12]
[13]

[14]

[15]

[16]

Adams,BradleyJ.andByrd,JohnE.Recovery,Analysis,andIdentificationofCommingledHuman
Remains.s.l.:Springer,2008.
MareanC,AbeY,NilssenP,andStoneE.2001Estimatingtheminimumnumberofskeletal
elements(MNE)inZooarchaeology:AreviewandaNewImageanalysisGISApproach.
AmericanAntiquity66:333348.
BuikstraJE,andUbelakerDH(editors).1994.StandardsforDataCollectionfromHuman
SkeletalRemains,ProceedingsofaSeminaratTheFieldMuseumofNaturalHistory(Arkansas
ArcheologicalSurveyResearchSeriesNo.44).Fayetteville:ArkansasArcheologicalSurvey.
HerrmannNPandBennettDevlinJ.2008.Assessmentofcommingledhumanremainsusinga
GISbasedapproach.InBAdamsandJByrd(editors):Recovery,Analysis,andIdentificationof
CommingledHumanRemains,pp.257269.NewYork:HumanaPress.
BadkurP,andNathS.1990.Useofregressionanalysisinreconstructionofmaximumbone
lengthandlivingstaturefromfragmentarymeasuresoftheulna.ForensicSciInt45(12):15
25.
Bidmos MA. 2008. Stature reconstruction using fragmentary femora in South Africans of
European descent. Journal of Forensic Sciences 53(5):1044-1048.
BidmosMA.2009.Fragmentaryfemora:Evaluationoftheaccuracyofthedirectandindirect
methodsinstaturereconstruction.ForensicScienceInternational192(13).
MysorekarVL,VerrmaPK,andNandedkarAN.1980.Estimationofstaturefrompartsof
boneslowerendoffemurandupperendofradius.MedicalSciLaw20(2836).
SteeleG,andMcKernTA.1969.Amethodforassessmentofmaximumlongbonelengthand
livingstaturefromfragmentarylongbones.AmJPhysAnthropol21:215227.
GunzP,MitteroeckerP,BooksteinFL,andWeberGW.2004.Computeraidedreconstruction
ofincompletehumancraniausingstatisticalandgeometricalestimationmethods.Enterthe
Past:ComputerApplicationsandQuantitativeMethodsinArchaeology,BARInternational
Series1227.Oxford:Archaeopress.p9294.
NeubauerS,GunzP,andWeberGW.2005.DigitalreconstructionofP.boiseiOH5.AmJPhys
Anthropol40(Suppl):156.
SylvesterAD,MerklBC,andMahfouzMR.2008.AssessingAL2881femurlengthusing
computeraidedthreedimensionalreconstruction.JHumEvol55:665671.
AbdelFatah,EmamE.ThreeDimensionalNonlinearStatisticalModelingFrameworkfor
MorphologicalAnalysis.Mechanical,AerospaceandBiomedicalEngineering,TheUniversityof
Tennessee.Knoxville,TN:s.n.,2012.PhDDissertation.
AbdelFatah,EmamE.,etal.4,2012,AThreedimensionalAnalysisofBilateralDirectional
AsymmetryintheHumanClavicle.,AmericanJournalofPhysicalAnthropology,Vol.149,pp.547
559.
Mahfouz,MohamedR.,etal.,AutomaticMethodsforCharacterizingofSexualDimorphismof
AdultFemora:DistalFemur.2007,ComputerMethodsinBiomechanicsandBiomedical
Engineering,Vol.10,pp.447456.
Surazhsky,Tatiana,etal.,AcomparisonofGaussianandmeancurvaturesestimationmethodson
triangularmeshes.2003.IEEEInternationalConferenceonRoboticsandAutomation.Vol.1,pp.
10211026.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

8. AppendixA(softwareusageexample)
I.
1.
2.
3.

InstallationandStartUp
DoubleclickBoneMatchSetup.exe.
WhenpromptedbytheUserAccountControlwindow,selectYes(WindowsVista,7,or8only).
Whenpromptedtochooseaninstalllocation,youmaychoosethedefault(C:\ProgramFiles
(x86)\BoneMatch)orselectBrowsetochooseanotherfolder.SelectInstalltobegininstalling
theapplication.
4. TheinstallationpopupwindowwilldisplayInstallationCompletewhenSetuphascompleted
successfully.SelectClosetoclosetheinstaller.
5. Tolaunchthesoftware,doubleclickthenewBoneMatchiconlocatedonyourdesktop.

II.

UsingtheApplication
1. LoadfracturedbonesbyclickingtheLoadModelsbuttonatthelowerrightcorner

2. IntheOpenModel(s)dialog,selectmultiplemodelsusingSHIFT/CTRLkeysonyourkeyboardor
usingmouse,thenhitOpen

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

3. Youcanshow/hidetheloadedmodelsfromthecorrespondingcheckboxesintheSelectModel
box(redhighlighted).Also,youcanupdatethemodelmaterial(color)byselectingthemodeland
updatingthematerialvalues(bluehighlighted).Youcandosometransformationsoneachmodel
by selecting model, changing translation/rotation/scaling values (green highlighted). Note that
youneedtohitapplysothatthenonreversibletransformationapplies.

4. In the viewer, you can use the ESC key on your keyboard to switch between navigating and
selectingmodes.Innavigationmode,ifthereareselectedmodelsthenyoucanmanipulatethem
(bydraggingmodelsmanipulator).Ifthereisnomodelselectedinnavigationmodelthenyoucan
rotate(leftmousebutton+move),pan(middlemousebutton+move)andzoom(mousewheel)
thewholeview.Youcanclearselectionbyrightclicking,thenselectingClearselectedmodel(s).
Inselectionmode,youcanselectmultiplemodelsusingSHIFT/CTRLkeys.Selectedmodelshave
manipulators around them, as shown in the next figure. You can select multiple models
(fragments),rightclickwithmouseandselectAlignselectedmodel(s)withtemplate

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.


5. Youcanselectwhichtemplateyouarewillingtoaligntheselectedmodels(fragments)with,from
thenextdialog.

6. Inthenext(alignment)dialog,youcanswitchtoselectingmode(ESCkey),draggingthemodel
withmousebyhittingthesmallwhitesphereatthecenterofthemanipulator,oryoucanrotate
themodelusingthethreerings(rotationaroundthreemajoraxes).

7. Byselectingmodel,rightclickingandhttingSelectregion(s)fromthepopupmeu,youcanstart
selectingpointsbelongingtotheoriginalnonfracturedbone.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

8. You can draw polygon on the screen that will be intersected with the models and extract the
selectedpoints

9. After being done with points selection, you need to right click and confirm your selection by
clickingConfirmselectedregion(s).YoucanalsocleartheselectionbyclickingClearselected
region(s)andstartoveragain.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

10. For better results, you can select models and align them to the template model using Align
Selected button. You can change alignment parameters (number of alignment iterations,
minimumerrordistance).Youalsocanchangemodelstransparencybyselectingthemodeland
changingthetransparencyslider.

11. You can see the distance map between fragment and template points after alignment, by
selectingmodelandclickingShowbuttonbesidesDistanceMaplabel.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

12. Ifyoubelievethatmultiplefragmentsbelongtooneelement(twofracturesofthesamebonefor
thesameperson),thenyoucanmergethem,byselectingthefragments,rightclickingandhitting
Mergeselectedmodels.Youcangobackandseparatethemifyouchangedyourmind.After
mergingmultiplefractures,theyshouldappearasoneobject(onecolor).

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

13. You need then to accept alignment by selecting the models, right clicking and hitting Accept
alignment. Any time before accepting alignment, you can change the template by selecting
Templatebutton.

14. Youcanreconstructthecompleteboneforthefracturedmodelsbyselectingthemodels,right
clickingandhittingReconstructbone(s).Ittakessometimetoreconstructthecompletebone.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

15. YoucanhitOktoacceptthealignmentandmorphing.Youcanseethemorphedmodelinthe
main application window. You also can save the morphed model to the computers HDD by
selecting the morphed model, clicking Save Model button at the lower right corner of the
screen,andselectfoldertosavethemodelat.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.