Int J Legal Med (2007) 121: 917DOI 10.1007/s00414-005-0058-6ORIGINALARTI CLES. A. Blackwell . R. V. Taylor . I. Gordon .C. L. Ogleby . T. Tanijiri . M. Yoshino .M. R. Donald . J. G. Clement3-D imaging and quantitative comparison of humandentitions and simulated bite marksReceived:7 August 2005 / Accepted: 13 October 2005 / Published online:4 January2006# Springer-Verlag 2006Abstract Thisstudypresentsatechniquedevelopedfor3-D imaging and quantitative comparison of humandentitions and simulated bite marks. Asample of 42studymodelsandthecorrespondingbites, madebythesame subjects in acrylic dental wax, were digitised by laserscanning. Thistechniqueallowsimagecomparisonof a3-D dentition with a 3-D bite mark, eliminating distortiondue to perspective as experienced in conventional photog-raphy. Cartesian co-ordinates of a series of landmarks wereused to describe the dentitions and bite marks, and a matrixwas created to compare all possible combinations ofmatches andnon-matches usingcross-validationtechni-ques.Analgorithm, whichestimatedtheprobabilityofadentition matching its corresponding bite mark, wasdeveloped. A receiver operating characteristic graphillustratedtherelationshipbetweenvaluesfor specificityand sensitivity. This graph also showed for this sample that15%ofnon-matchescouldnotbedistinguishedfromthetrue match, translating to a 15%probability of falselyconvicting an innocent person.Keywords Forensic odontology . Bite mark .3-dimensional. 3-D quantification . AnimationIntroductionProblemswith human bitemark analysisBitemarkanalysisiscurrentlycontentious.Forasubjectwith such potentially serious outcomes for both suspect andvictim, littleresearchanalysingmethods andevaluatingoutcomes has reached peer reviewed journals [1]. Althoughadmissibilityofbitemarkevidencehasbeenestablishedand routinely accepted in the USA and other legal systemsfor a long time [2] some odontologists argue that bite markmethodologyhasneverundergonecriticalevaluationandlegitimately passed the Frye [3] test for admissibility. Thisproblemisalsorelevant for other areassuchasearprintidentification [4]. Other legal observers are concerned thatforensic odontologists are giving insufficient critical atten-tiontothequalityof bitemarkevidencepresentedtothecourts [5, 6]. Central tothe problemof analysis is thedifficulty of comparing 2-dimensional images of a bite markwith 3-dimensional replicas of dentitions which may havecaused them.Over time, a deficiency in quantitative bite mark researchhasresultedinuncertaintyinbitemarkevidenceinlegalsystemsworldwide, particularlyinAustralia. Thenaturaltendencytoseewhatonewantstosee,therebytemptingexaminers to over-interpret bite mark evidence, has led toserious difficulties when bringing such evidence before thecourts [7]. This area of forensic science requires standardi-sation to ensure consistency of expert opinions. Twonotorious Australian cases [8, 9] have seen bite markevidence rejected as unsafe, and convictions overturned onS. A. Blackwell(*) . R. V. Taylor . J. G. ClementOral Anatomy, Medicine and Surgery Unit,School of DentalScience, Faculty of Medicine,Dentistry and HealthSciences,The University of Melbourne,720 Swanston Street,3010, Victoria, Australia,e-mail:sherie@unimelb.edu.auFax: +61-3-93411594I. Gordon . M. R. DonaldStatisticalConsulting Centre,The University of Melbourne,3010, Victoria, Australia,C. L. OglebySchool of Geomatics,The University of Melbourne,3010, Victoria, Australia,T. TanijiriMedic Engineering,Kyoto, Japanhttp://www.rugle.co.jp/M. YoshinoNational Research Instituteof Police Science,Chiba, Japanhttp://www.nrips.go.jp/index-e.htmlappeal. Perhaps for suchreasons, bite markanalysis iscurrentlyundergoingreview.Ingeneral,courtsnowlookfor quantitative rather than simply descriptive analysis be-fore accepting scientific evidence. It can be anticipated thatfuture developments in the analysis of bite marks will needto follow this general trend if convictions are going to bemade with confidence.Importanceof the third dimensionMany studies have described and quantified bite patterns intwodimensions(photographs,overlaysetc.).Yet,despitethe fact that the dentition of the biter and the correspondingbite marks are both 3-D phenomena, there have been few 3-Danalyses [1012]. This is surprising, but it may reflect thelack of access to methods of measuring in three dimensionsthat haverecentlybecomeavailable. Legal problemsin-volving bite mark evidence suggest that alternative methodsof analysis may be required, and the logical first step is toanalyse bite marks in 3-D.There are three factors of 3-dimensionality involvedwhen one person bites another: the curvature of the skin, theshapeofthebitingdentitionandthedepthoftheinjuryshouldthetooth/teethpuncturetheskintocreateade-pression, althoughthisisinfrequent. Theinjury, asit isbeing inflicted, is a 3-dimensional eventthe skin deformsto accommodate the shape of the teeth. However, once theteeth are withdrawn, the skin is restored near to its originalshape and the resultant mark is represented without depthinformation on the curved surface of the skin. If the force ofthe bite is great enough to leave an indentation in the skin,then the mark is also 3-dimensional. Injuries range from adefined mark/s, a diffuse bruise, complete removal of tissueand swelling due to inflammation.Aim of studyThis study develops a method for 3-Dimaging andquantitativecomparisonof human dentitionsandthe cor-responding simulated bite marks. This study also defines anoptimumthreshold for this sample, which segregatesmatches from non-matches.Materials and methodsStudy models and simulatedbite marksForty two third year students from the School of DentalScience, TheUniversityof Melbourne, Australia, con-tributed a maxillary and a mandibular dental-stone studymodel of their own teeth. Students were then asked to biteinto a wafer of double-thickness, pre-heated acrylicdental wax (Lordell Trading Pty Ltd, Wetherill Park,New South Wales, Australia) to create an upper and lowerbiteimpressionof their ownteeth. Theseimpressionswerecalledsimulatedbitemarksastheyweremadeinwaxandnot humanskin. Dental waxhas beenusedpreviously in bite mark research [13, 14] and during bitemarkinvestigations. Correspondingsetsofmodelsandwax bites were numerically labelled; all upper dentitionsandbitemodelswerelabelledUandlowerdentitionsand bite models L.Reversemodels of wax biteimpressionsMirroredor reversemodelsof thewaxbiteimpressionswere created to make the laser scanning process faster andmoreefficient.Imaging an indentation, i.e. a bite in wax,would be difficult. The alternative process of scanning anoutwardprotrudingsurfacewouldresult inobtainingthemaximum amount of data points. The substrate selected tocreatethereversemodelswastype3lowviscosityHy-droflex hydrophilic vinyl polysiloxane impression material(GC Corporation, Tokyo, Japan), the physical properties ofwhich are suitable for this purpose. It remains highly stableafter setting, can be stored long-term without shrinking ordecomposingandhas99.7%recoveryfromdeformation.Hydroflex has an accuracy of 204 m and meets the ISO4823International Standard[15]andADASpecificationNo.19 [16].Themodels were made bysqueezingtheimpressionmediumandcatalyst simultaneouslyfroma two-cham-bered, triggered dispenser onto the wax bite. Latex gloveswere not worn as research suggests that they causeretardation in the setting time of some vinyl polysiloxanematerials[17].Hydroflexwasappliedononesideofthewaxsheet andwas allowedtoflowunder theforceofgravity to force out air bubbles. Setting time was ap-proximately30mindependingonroomtemperatureandhumidity. Once set, the Hydroflex was lifted out of the waxresulting in a reverse bite model of around 5-mm thickness.The process was repeated for the opposite side of the wax.Labelling was achieved by engraving the wax with amirrorednumbersothelabel couldbereadsequentiallyonce the Hydroflex was lifted from the wax.Laser scanning equipmentThe FARO Gold Arm (2001 FARO Technologies Inc, LakeMary, FL, USA) and ModelMaker H40 [3-DScanners (UK)Ltd, Coventry, UK] (Fig. 1a) were used in conjunction tolaser-scanthestudymodelsandbitemodels(Fig. 1b)tocreate 3-D images.The FARO Arm (http://www.faro.com, 8 Aug 2001) is ahighly accurate measurementinstrument designed for useinengineeringandmanufacturingfor thecontrol of di-mensional qualityinmass production. The armcanbescrewedorfixedwithclampstoaflatsurface,whichre-quires a high degree of stability to achieve the best results.The Gold Arm, with an accuracy of 84 m, was used inthis project. The arm portion of the instrument is composedof two shafts of lightweight aircraft-grade aluminium,internally counterbalanced for ease of use. Rotation of the10shafts about seven pivot points provides a sphericalworking space 3 m in diameter.TheModelMaker H40(http://www.3dscanners.com, 8Aug 2001) is a hand-held, non-contact reverse engineeringandinspectionsensor whichmounts tothe endof theFARO Arm. The scanning system works on the principle oflaser stripe triangulation. A laser diode and stripe generatorprojects a laser line onto the object to be scanned. The lineis viewed at an angle by a camera, and height variations inthe object are seen as changes in the shape of the line. Theresulting captured image of the stripe is a profile thatcontains the shape of the object. The accompanying surfaboard uses digital signal processing to convert video data todigital datatocapturesurfaceshapeinreal timeat over14,000 points per second. Either keyboard and mouse, or afoot pedal, drives the system.Pre-scan preparationThe matt, beige surface of the dental stone study modelswas ideal for scanning; however the shiny surface of theHydroflex models was too reflective. Each Hydroflexmodel was sprayed with Flawcheck visible inspectionsystem[DY-MARK(Aust) Pty Ltd, Laverton North,Victoria, Australia], a fine white powder whichgave thesurface a matt finishassistingthe camera indetectingmaximumlasersignal.Thethickness ofsprayed powderwas sufficient to reduce the surface shine of the Hy-droflex. Onepassof thesprayresultsinalayer of ap-proximate thickness 1030 m according to J. Morgan ofDY-MARK(Aust) PtyLtd(personal communication, 2June, 2005). Therefore, the effect of this additionalcoating is minimal.Laser scanning processThesystemwascalibratedagainstamachinedalignmentcube of precisely known dimensions. The calibration cubewas also used as a pedestal to elevate the models for easierscanning. It wasessential that thepositionofthemodelremained undisturbed during scanning, otherwise thesystem would reassignthe location of the object in spaceandadouble-imagewouldresult. If themodel wasdis-placed, the process would need to be repeated fromthe start.A pre-scan was initially taken by passing the laser lineover themodel inseveral directions sothesystemreg-isteredthesizeandlocationof theobject inspace. Thecolour of the resultant 3-D image was white and composedof thousands of interlinked polygons (Fig. 2). A systematicmethodfor scanningeachstudymodel was developed:first, the facial aspect, followed by the occlusal and then thelingual surfaces. Being systematic was important to avoidoverlappingpasses of thelaser line, as patches of dataaccumulated on top of one another creating a ruffled effectonce the polygons had been merged. Using the lasso tool toselect small areasreducedoverlapping, asnewdatawasadded to the selected areas only. Scanning time wasapproximately 30 min for a complete set of study modelsand15minfor a set of the less undulatingHydroflexmodels.ba Fig. 1 a The FARO GoldArmand ModelMakerH40 laserscanner usedto digitise thestudy dentitions and Hydroflexbite models and ba laser linegeneratedby the ModelMakerpassing over the surface of a bitemodel. The image is reflectedback into the ModelMakerscamera to create the resultant3-D data set11Data processingImages were processed using ModelMaker software (Ver-sion3.3.3.3199619983DScannersUKLtd, Coventry,UK)byfollowingasequenceofsteps. Superfluousdatawere cleaned or deletedfromaroundeachimage, forexamplethedatafromscanningthecubeunderneaththemodel. Groupsof datapointsor polygonsthat composeeach image were merged or linked. Small holes or areasvoid of data between scan patches were filled to produce acontinuous surface. Hole size ranged from 50 to 500 points,and considering that a scan of each half of a study modelwas approximately140,000points, theeffect of extrap-olating the data for these holes was not detrimental to theanatomical accuracy of the surface. A minimum amount ofsmoothing wasdonefor eachimagetocounteract anyrufflingeffect causedbyoverlappingpassesofthelaser.Images weredecimatedtobetween15and25MBandconverted to stereolithography interface format (stl) files toenable viewing in 3D Rugle3 software (Version 1.0 19982001, Medic Engineering Corporation, Kyoto, Japan.Release 24.09.2001 Australia).3D Rugle3 software3DRugle3 (http://www.rugle.co.jp, 5 Jun 2001) is a 3-Ddataviewingandmeasurement softwareprogramusedpredo-minantlyinresearchon3-Dfacialanalysis[18,19].Theprogramiscomposedoffivemodulessuperimposition,basicmeasurement,intersurfacedistance,fittingand sim-ulation.Theintersurfacedistanceandbasicmeasurementmodules were used in this study.The3-Dimagesofthedentitionsandbitemarkswereimported as stl files into the intersurface distance module.Four pointsthe midpoint of the buccal cusp of each of thesecond pre-molars and the midpoint of the occlusal surfaceof each of the anterior incisorswere designated on eachimageusingthecorrect tiltfunctiontoaligntheimagewith the xy plane, the x-axis and the midline. This resultedinastandardorientationof eachimage. Theimagewasexported into the basic measurement module as a materialsand geometry format (mgf) file. Mgf files are smaller thanstl files, therefore, they are easier and faster to manipulateusing the software. Small voids or holes in the data, whichare often caused by the process of decimation, were filledusingthefilter, fillholefunction. Holesinthedataarefilledusingheight orzvaluesofthesurroundingpixels,and the original data set is not affected.MorphometriclandmarksThe first five teeth of each quadrant (incisors, canines andpre-molars) were used for landmark placement. Molar teethare less likely to make contact with the skin during a bitedue to their posterior location within the oral cavity. Severaldentitions in the sample had pre-molar teeth missing and themolars had moved anteriorly to fill the gap. The molar wasincludedinthesecases. Landmarks wereplacedonthebuccal cusps of pre-molars and molars.Atotal of 42 landmarks were placed on each image, 30 onthe teeth and the remainder along the midline and referenceline (Fig. 3). The reference line joins the peak points of thesecond pre-molars and the midline joins the midpoint of thereference line and the midpoint between the anteriorincisors. Landmarks 130 were placed on the occlusalsurface of each tooth and consisted of the peak point, themesial-mostpointandthe distal-mostpoint.3D Rugle3smaxpointfunctionallowedobjectiveplacement ofthepeakpoint oneachtoothbylocatingthehighest zvaluewithin a defined area. The operator located the mesial-mostand distal-most points. Landmark 31 was the midpoint ofthe reference line, landmark 32 was the midpoint betweenthe anterior incisors andlandmarks 3342were placedalong the midline. Lines were drawn from the distal-mostpoint of eachtoothtothemidline, andlandmarkswereplaced at the intersection each line made with the midline.Statistical analysisAnumber of variables comprising curves, angles anddistances were created using the x, y, z co-ordinates of eachFig. 2 3-Dimageofamaxillarystudymodel asaresult oflaserscanning. Thousands of interlinked polygons represent the mor-phology of the dentitionFig. 3 Forty-two landmarks were placed on the 3-D images of eacha study model and its b corresponding Hydroflex bite model12landmark providing a numerical description of thedentitions and bite marks. Variables underwent transforma-tionssuchassquareroot andlogarithm, toapproximatesymmetryandnormality,sincediscriminationislikelytobe more effective using variables with these properties. Thevariables are as follows:1) Curve of the arch defined by landmarks 130:yi a1x2i a2x4iyi = transformed y-value; xi = transformed x-value; a1,a2=coefficients of the curve; i=1, 2,..., 302) Fromthiscurve, thepoint withthesmallest residual(distancefromthecurveintheydirection) andthepoint with the largest residual generated the variables:minres and maxres.3) Length of each tooth (mesial-most point to distal-mostpoint)lj x3j2 x3j_ _2 y3j2 y3j_ _2_for j=1,,10 together with the total length=

10j1 lj4) Nine distances (point 2 to each peak point):rk x2x3k2 2y2y3k2_ _2_; k 1; . . . ; 95) Nineangles(formedbypoint 2toeachpeakpointand the midline)ak tan1x2x3k2y2y3k2_ _; k 1; . . . ; 9A matrix was created to compare all possible combina-tionsofmatchesandnon-matchesofdentitionsandbitemodels. For eachcombination, the absolute differencesbetweenthevariablesinthedentitionandthesamevar-iables inthebite model were recorded. These absolutedifferencesreflectthequantitativeproximitybetweenthedentition and the bite model with which it is compared. Thematrix consisted of 1,722 combinations: 42 dentitions 41bitemodels. Thesamplewas composedas follows: 40complete sets of dentitions and their corresponding Hydro-flexbitemodels, resultingin4040=1,600combinationsfor each set, 40of which were true matches. That is,dentition1matchesbitemodel1butdoesnotmatchtheremaining bite models 240; dentition 2 matches bitemodel 2but doesmatchbitemodel 1or340, etc. Twoadditional dentitions were included for which the bitemodel data was excluded, and one bite model for which thedentitiondata was excluded. Hence, the total of 1,722combinations came from 42 dentitions and 41 bite models(40 matches, 1,682 non-matches).Logistic regression was used to obtain a predictivemodel, or algorithm, for a match (available from author onrequest). Cross-validationwas alsoimplementedonthedata; in turn, each bite model was removed fromthe data setand a logistic regression model was fitted using the rest ofthe data. The fitted statistical model was then used to makepredictions for the 41 combinations in the omitted data. Thisprocess was repeated for each of the 42 dentitions. In thisway, the data used to make predictions was separate fromthe data used to estimate the performance of the algorithminall cases, which gives unbiased estimates of sensitivity andspecificity. Alternatively, when cross-validation is not used,the data used to test the predictive power of the algorithmare the same as those used to generate it, which tends to giveover-estimates of the algorithms predictive performance.The probabilities generated were expressed as values forsensitivity and specificity: Sensitivity=P(TP)=1P(FN) Specificity=P(TN)=1P(FP)where P(TP)=Probability of a true positive=probability ofobtaining a match for dentitions and bites that do in factmatch; P(TN)=Probability of a true negative=probabilityof obtaining a non-match for dentitions and bites that donot in fact match; P(FP)=Probability of a false positive=probability of obtaining a match fordentitions and bitesthat donot infact match; P(FN)=Probabilityofafalsenegative=probabilityof anon-matchfor dentitionsandbites that do in fact match.These probabilities were expressed in an ROC curve, agraphical representationof thecompromisebetweenthetrue positive (TP) and false positive (FP) probabilities forevery possible threshold value [20]. The ROC graph allowsa continuous assessment of the relationship between P(TP)andP(FP) for eachthresholdvalue andincorporates adegreeofuncertaintyinthedecision-making, ratherthansimplymakingadichotomous, yesor nodecision[21].ROC curves are typically used in medicine and are usefulaidsinthediagnosisof disease,suchashumanimmuno-deficiencyvirusor cancer. ROCanalysishasbeenusedpreviously in bite mark analysis [22, 23].Anideal ROCcurve(Fig. 4a)wouldbealinewhichfollowedthey-axisandmadearightanglefollowingtheline extendinghorizontallyfrom(0, 100). The optimalpoint on this line would be at (0, 100) where the probabilityof obtaining a true positive was 100% and a false positive0%. Other points alongthe curve represent a range ofvaluesfortrueandfalsepositiverates.Itisunlikelythatany realistic situation would result in an ideal ROC curve,so a compromise between TP and FP rates must beachieved.Figure 4b depicts a partial ROC curve for the dentitionsand bite marks used in this study and Table 1 shows howtheTP and FP rates vary for several threshold values on eitherside of the optimum threshold for this curve. For example,at threshold0.18, the TPrate or the probabilitythat adentitionandbite model were correctlyidentifiedas amatch, is only 55% which results in a FP rate of 3%. Sucha lowfalse positive rate is almost ideal, however, thesensitivity should be higher. At the other end of the scale at13threshold 0.006, the TP rate is much higher which is good,but at the expense of the FP rate which is now 26%. Thebest point on the graph is a compromise between TP and FPrates.Theoptimalpoint(arrowed)inFig.4bisonewiththeleast distance from the top left corner where the thresholdvalue is 0.018. At this point the TP rate is 78% and FP rate15%. At this point, theTPrateis as highas it canbewithoutconsequentlycausingtheFPratetobetoohigh.Therefore, at the optimal point on the ROC curve for thissample, 15% of non-matching dentitions and bite modelscouldnotbedistinguishedfromthetruematch(i.e.a15%chanceof wronglyconvictinganinnocent person).Whilst 78%ofmatchingdentitionsandbitemodelswerecorrectly identified as a match by the algorithm.Anideal situationfor this sample wouldbe 42truematches (TP rate=100%) and the remaining combinationswouldnot match. However, theactual situationfor thissample was: Only four matched with self (i.e. true match) and no other Twenty-seven matched with self and at least one other Seven did not match with self but matched with others100-specificity (P(FP))%sensitivity (P(TP)) %100 80 60 40 20 0100806040200100-specificity (P(FP))%sensitivity (P(TP)) %100 80 60 40 20 0100806040200a b Fig. 4 A receiveroperatingcharacteristic(ROC)graph wasused to illustrate the data.a Shows an ideal ROCcurveand b a partialROC curve forthis sample using cross-vali-dated data. The optimalpointsfor each graphare arrowedTable1 Therelationshipbetweenthreshold, theprobabilityof atruepositiveP(TP)andafalsepositiveP(FP)forthepartialROCcurve (Fig. 4b). The optimal values arein boldfaceThreshold Sensitivity P(TP)% 1-Specificity P(FP)%0.006 82.5 25.60.01 80.0 21.40.011 77.5 20.20.012 77.5 19.60.016 77.5 16.50.018 77.5 15.40.02 75.0 14.70.18 55.0 3.014 Four did not match with anything, including selfWhen the matrix was sorted according to threshold valueandthecombinationsrankedaccordingtothisvalue, thefirst 296combinationshadathresholdvalue0.018, sowere considered a match. The first ten are listed in Table 2asanexample. Thedentitionmodel labelled43anditscorrespondingbite model, havingthe highest thresholdvalue, were the closest match as determined by thealgorithm.Image visualisationand animationImages were imported into 3D Rugle3 software forvisualisation. During scanning, each data point wasrepresentedbyanx, yandz Cartesianco-ordinate andimages could consequently be rotated and viewed from anyangle. Aselectionoftheimageswereimportedinto3dsMax[Discreet (Autodesk) release4.0commercial 2001]animation software and rendered in artificial colours.Animations werecreatedfrom(1) apositivelymatcheddentition and bite mark showing near perfect registration,(2)anon-matchingdentitionandbitemark,(3)acuttingplane to illustrate topography of biting surfaces of teeth, (4)amaxillarydentitionregisteringwithaninkbiteprintedonto skin. These animations can be viewed on http://www.dent.unimelb.edu.au/3dbitemarks/.DiscussionAlthoughthemorphologyofeachdentitioninthisstudymaybe unique, we hypothesisedthat verysimilar andindeed indistinguishable bite marks may be produced by anumberofdifferent dentitions, despitetheuniquenessofthesedentitions. Bitemarksproducedbydifferent denti-tions in a firm substrate, cheese for example, may be moreunique with respect to each other, and more similar to theircorrespondingdentitions,thanbitemarksinflictedbythesameset ofdentitionsonahighlydeformablesubstrate,likeskin. Dynamic, tissueandpostural distortion, asex-plainedbySheasbyandMacDonald[24], haveasignifi-cant impact on the quality of a bite mark on skin.Theresults of this studyindicatedthat 15%of com-binations of dentitions and bite models in this sample werecategorised as a match when they were in fact a non-match,i.e. 15% of non-matching combinations were indistinguish-able from the true match. This translates to six out of the 42peopleinthissampleat riskofbeingfalsepositives, orwronglyconvicted. Thisfigureisonlyindicativeofthisparticular sample and may be lower in actual casework, butthis is not certain.Study sampleSomemaycriticisethesampleusedinthisstudy. Studymodels of thedentitionwereobtainedfromagroupofyoung,universitystudentsand mayprovidean unusuallyhomogenous sample. Anumber of subjects may haveundergone orthodontic treatment, a convergent processresulting in different people having teeth similarly ar-ranged. It isconsequentlypossiblethat thereisahigherdegree of similarity amongst dentitions in this samplecompared with a sample of high forensic significance, i.e.the sample may not be typically representative of dentitionscommonly examined in the wider community, which mayexhibit more uniquely identifying features such as missingor fractured teeth. Dentitions in this sample exhibited a lownumber of uniquely identifying characteristics, influencedperhaps by the young age (early 20s) of the subjects, thegoodconditionof their teethandtheir socio-economicbackground. However, even with a more varied sample, thepossibilitystill existsthat anumberofpeoplemayhavedentitions similar enough to produce indistinguishable bitemarks.Wax as an impression mediumWax was chosen as a bite impression substrate because it isaslightlyimperfectrecorderoftoothshape, asishumanskin. However, weacknowledgethat noalternativesub-strate can accurately mimic the complex physical mechan-ics of human skin.ErrorThe process of producing reverse bite models from the waxbites and creating a digital image by laser scanning wouldhave introduced a certain amount of error. However,Hydroflex, withanaccuracyof 204m, wasthebestmaterial to use to minimise this error. Acrylic wax is easilydeformedwithheatandforce, soitwasadvantageoustoreplace the wax bites with more robust Hydroflex models.Also, laser scanningwithminimal overlappingof scanpatches was the best method to keep error to a minimum.Table2 Thefirstten(outof296)combinationsofdentitionsandbite models with a threshold value 0.018, and therefore considereda matchDentition Bite model Threshold43 43 0.9338 1 0.8854 4 0.8724 44 0.84785 8 0.79330 30 0.7881 1 0.74752 52 0.7374 52 0.7153 3 0.688Labelling of models was not consecutivefor a number of reasons,hence model numbers such as 43, 85 and 52. However, be assuredthat they formed part of the 42 models used in this study15Laser scanningUse of the laser scanner and associated equipment requireda certain degree of training, and technique improved withpractice. Before startingtoscanthestudysamples, theoperators practised using the equipment on a dental modelthat was not included in the study.Best scanning results were achieved by using minimumpasses of the laser to reduce the ruffling effect fromoverlappingof scanpatches. Thescanner haddifficultydetecting sharp edges, in particular the incisal edges of theanterior incisors, so it was important not to over-scan theseedges as the data may provide a false representation of themorphology of the tooth.Image processingThereis a conflict betweendecimatingimagesenough toproduce files sufficientlysmall toopenandmanage inviewing software; however, if one over-decimates toomuchdatacanbelost.Itisimportanttogetthecrispest,cleanestoriginallaserscannedimagepossibleinthefirstinstance. If there is too much noise or excess data presentintheoriginal scan, thiswill giveaninaccuratesurfacerepresentation and it is wiser to re-scan the model. For thisreason, the corresponding images of two dentitions and onebitemodelusedinthisstudywereexcluded.Theimageswere of insufficient quality to accurately represent the truemorphology and would require re-scanning.Landmark placementThepeakpointoftheocclusalsurface,orthepartofthetooth with the greatest distance fromthe gingiva, waschosen rather than the midpoint, as this is the first point ofcontact with the skin during a bite.AnimationThe use of 3ds Max requires some expertise and the processofcreatingananimationistime-consuming,howevertheresults can be impressive. Such a pictorial display may be ofbenefit in a courtroomsituation to assist the jury inunderstandingevidence,notonlywithbitemarkanalysisbut in many other fields. 3-D crime scene reconstruction iscurrentlyusedfortrainingandincourtintwoAustralianstates, QueenslandandWesternAustralia, andhasbeenbeneficial (http://www.qmisolutions.com.au/article.asp?aid=77&pfid=5, 15 Nov 2005).3-D in the futureIt isexpectedthat 3-Dimagingequipment will becomemore affordable and accessible in the future. If 3-Dimagingandquantificationtechniques weredevelopedandwereproventobesuccessful,policeandforensicinvestigationinstitutionswouldbemoreencouragedtoutilisefundsinthis area.SummaryTheideasandmethodsdevelopedinthisstudyfor 3-Dimaging and quantitative comparison of human dentitionsand their corresponding bite marks are by no means a finalsolution to the complex problems bite mark analysispresents. Wehopethisresearchwill increaseawarenessabout the possibilitythat the number of false positivesresultingfrombitemarkevidencegivenincourtmaybehigherthanwerealise,aproblemwhichisnotuniquetobite mark analysis but affects many other areas of forensicscience such as earprint identification (http://www.forensic-evidence.com/site/ID/DNAdisputesEarlID.html, 15 Sept2005). We hope researchers will be inspired to continue toinvestigate the 3-dimensionality of bite marks and toimprove the science of bite mark analysis.Acknowledgements Wearegratefulfortheadviceandgenerousassistance of the participants of the study; Scanning and InspectionPtyLtd, Melbourne; theVictorianInstituteof ForensicMedicine,Melbourne; Mr DavidThomas andMr RonnTaylor, School ofDental Science, The Universityof Melbourne, Australia andMrChris Scott. We are grateful for the financial support of the AustralianAcademy of Forensic Sciences, the Australian Dental ResearchFoundation and The University of Melbourne.References1. Pretty IA, DSweet (2001) The scientific basis for humanbitemark analysesa criticalreview. Sci Justice 41:85922. Peoplev. Marx, in54Cal. App. 3d100, 126Cal Rptr. 350.19753. Frye v. United States,in 293 F.1013 (D.C. Circ). 19234. Rutty GN, Abbas A, Crossling D (2005) Could earprintidentification be computerized? An illustrated proof of conceptpaper. Int J Legal Med, on-linepublication10.1007/s00414-005-0527-y5. Rothwell BR (1995) Bite marks in forensic dentistry: a reviewof legal, scientificissues. J Am DentAssoc 126:2232326. Gundelach A (1989) Lawyers reasoning andscientific proof:acautionarytaleinforensicodontology. JForensicOdontos-tomatol 7:11167. Wells D(1998) Bitemarks (Teaching resource material forForensic Diploma of Clinical Forensic Medicine). MonashUniversity, Victoria,Australia8. Raymond JohnCarroll,inAustralianCriminalReports(1985)Court of CriminalAppeal, Queensland,p 4109. LewisvTheQueen, inFederal LawReports(1987)CourtofAppeal of the NorthernTerritory, p 10410. Forrest A, Davies I (2001) Bitemarks ontrialtheCarrollcase. Aust Soc ForensicDentistry 18:6811. Thali MJ, Braun M, Markwalder ThH, Brueschweiler W,ZollingerU,MalikNJ,YenK,DirnhoferR(2003)Bitemarkdocumentationandanalysis: theforensic3D/CADsupportedphotogrammetry approach. ForensicSci Int 135:11512112. Martin-delas Heras S, ValenzuelaA, Ogayar C, ValverdeAJ, Torres JC(2005) Computer-basedproductionof com-parison overlays from 3D-scanned dental casts for bite markanalysis.JForensicSci50:1271331613. Nambiar P, Bridges TE, Brown KA (1995) Quantitativeforensic evaluation of bite marks with the aid of a shapeanalysis computer program: Part 1; The development of SCIPand the similarityindex. J ForensicOdontostomatol 13:182514. Sweet D, Bowers CM (1998) Accuracy of bite mark overlays: acomparisonof five commonmethods toproduce exemplarsfrom a suspects dentition. J ForensicSci 43:36236715. International StandardOrganisation(ISO). Dental elastomericimpression materials, ISO 4823, 2nd edn. 1992-03-15, Section5.11 Detailed reproduction, p 216. Revised American Dental Association specification No.19 fornon-aqueous, elastomericdentalimpressionmaterials(1977)JAmDentAssoc94:73374117. BaumannMA(1995) Theinfluenceof dental gloves onthesettingof impressionmaterials. BrDent J 179:13013518. Fraser N, Yoshino M, Imaizumi K, Blackwell SA, Thomas CD,Clement JG(2003)AJapanesecomputer-assistedfacial iden-tification systemsuccessfully identifies non-Japanese faces.Forensic Sci Int 135:12212819. Yip E, Smith A, Yoshino M (2004) Volumetric evaluation of facialswelling utilizing a 3-D range camera. Int J Oral Maxillofac Surg33:17918220. Swets JA, Dawes RM, Monahan J (2000) Psychologicalscience can improve diagnostic decisions. AmPsychol Soc1:12621. PrettyIA(2005)Reliabilityofbitemarkevidence. In: DorionRBJ (ed) Bitemark evidence. Marcel Dekker, New York,pp53154522. Whittaker DK, Brickley MR, Evans L (1998) A comparison oftheabilityofexpertsandnon-expertstodifferentiatebetweenadult and child human bite marks using receiver operatingcharacteristic(ROC)analysis. ForensicSci Int 92:112023. Pretty IA, Sweet D(2001) Digital bite mark overlaysananalysis of effectiveness.J Forensic Sci 46:1385139124. Sheasby DR, MacDonald DG (2001) A forensic classificationof distortion in human bite marks. Forensic Sci Int 122(1):757817