• Aucun résultat trouvé

Implicit Representations of the Human Intestines for Surgery Simulation

N/A
N/A
Protected

Academic year: 2021

Partager "Implicit Representations of the Human Intestines for Surgery Simulation"

Copied!
7
0
0

Texte intégral

(1)

HAL Id: inria-00517243

https://hal.inria.fr/inria-00517243

Submitted on 14 Sep 2010

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Implicit Representations of the Human Intestines for Surgery Simulation

Laure France, Alexis Angelidis, Philippe Meseure, Marie-Paule Cani, Julien Lenoir, François Faure, Christophe Chaillou

To cite this version:

Laure France, Alexis Angelidis, Philippe Meseure, Marie-Paule Cani, Julien Lenoir, et al.. Implicit

Representations of the Human Intestines for Surgery Simulation. Modelling and Simulation for

Computer-aided Medicine and Surgery, MS4CMS, Nov 2002, Rocquencourt, France. �inria-00517243�

(2)

Simulations

Laure France 1

Alexis Angelidis 2

PhilippeMeseure 1

Marie-Paule Cani 2

Julien Lenoir 1

Francois Faure 2

Christophe Chaillou 1

1

LIFL,Universite LilleI, Villeneuve d'Ascq

[Laure.FrancejPhilippe.MeseurejJulien.LenoirjChristophe.Chaillou]@li.fr

2

iMAGIS-GRAVIR, INRIA Rh^one-Alpes, Montbonnot

[Alexis.AngelidisjMarie-Paule.CanijFrancois.Faure]@imag.fr

Abstract

Inthispaper,weproposeamodelingoftheintestinesbyimplicitsurfacesforabdominalsurgerysimula-

tion. ThediÆcultyofsuchasimulationcomesfromtheanimationoftheintestines. Asamatteroffact,the

intestines areaverylong tubethatis notisotropicallyelastic, andthat bendsoveritselfatvariousspots,

creatingmultipleself-contacts.

We usea multiple component modelfor the intestines: Therst component is amechanicalmodelof

their axis; the second component is aspecicsphere-based modelto managecollisions and self-collisions;

and the third component is a skinning model to dene their volume. This paper focuses on the better

representation for skinning the intestines. We compare two implicit models: Surfaces dened by point-

skeletons andconvolutionsurfaces.

A direct applicationof this simulation is the training of a typical surgical gesture to move apart the

intestinesinordertoreachcertainareasoftheabdomen.

Keywords: Surgicalsimulation,virtualreality,implicitsurfaces,real-timecomputation,applicationto

medicaldomain.

Resume

Danscetarticle,nousproposonsunemodelisationdesintestinsparsurfacesimplicitespourlasimulation

dechirurgieabdominale. La diÆculted'unetellesimulationprovientdel'animationdesintestins. Eneet,

lesintestinssontuntreslongtubequin'estpaselastiquedemaniereisotrope,etquiserepliesurlui-m^eme

endenombreuxendroits,creantdemultiplesauto-contacts.

Nousutilisonsunmodeleaplusieurscomposantspourlesintestins: lepremiercomposantestunmodele

mecaniquereduit aleur axe; le deuxiemecomposant estun modelespeciquebase surdes spheres pour

gererlescollisions etauto-collisions ;etletroisiemecomposantestunmodeled'habillagepourdenirleur

volume. Cet article sefocalisesurlameilleurerepresentationpourhabiller les intestins. Nous comparons

deuxmodelesimplicites:les surfacesdeniespardessquelettesponctuelsetlessurfacesdeconvolution.

Uneapplicationdirectedecettesimulationestl'apprentissaged'ungestechirurgicalcaracteristiquepour

rassemblerlesintestinsandedegagercertaineszonesdel'abdomen.

Mots cles: Simulationchirurgicale, realitevirtuelle, surfacesimplicites,calculstemps-reel,application

audomainemedical.

1 Introduction

Laparoscopicsurgeryisasurgicaltechniquethatusesacameraandsurgicaltoolsinsertedintothebodyvia

smallincisions. Thistechniqueleadstounusualworkingconditionsthatrequiremanyrehearsals. Thetraining

thencouldbeperformedonasimulator. Surgeonswouldlearnthesurgicalgesturesonphysicallybasedvirtual

organsthatreacttotheiractions.

Some endoscopic simulators exist, usually consisting in simulating the interaction of surgical tools with

a single organ. Dynamic virtual organs have already been modeled, such as the liver [5, 10, 12, 16] or the

(3)

ofverydeformableobjectsthatareincontactin variousspots andmayundergoverylargedisplacements.

The goal of our simulator is to clearan area to treat, by pushing/pulling and maintainingthe intestines

in another place of the abdominal cavity. Indeed, this clearing stage is needed before surgeries such as the

gallbladderremovalorcolontumors. However,thisactionrisesseveralchallenges: Largedisplacements,contacts

andself-collisions.

Wedevelopedarstmodeloftheintestinesin[13],composedofthreecomponents: Amechanicalmodelto

denetheirphysicalbehavior;amodelforself-collisionsbetweentheirparts andcollisionswithotherobjects;

andaskinningmodelto representand displaytheirshape. However,theskinningmodelbasedonparametric

generalizedcylinderswasnotsatisfying,sinceitcouldresultinsurfacediscontinuitiesinhighly curvedregions

andinvolumeloss.

In this paper, we addressthis problem byusing implicit surfaces,known for representingquite naturally

organicshapes[2]. Indeed, theyfollowaskeletonin anintuitive wayregardlessofitsdeformation. Moreover,

this representation allows thegeneration of contact between surfaces [6]. Lately, implicitsurfaces were even

adaptedtolevelsofdetail(LOD)bytaking intoaccountmulti-resolutionaspects[1].

Thepaperisorganizedasfollows: Section2reviewsourthreecomponentmodelfortheintestines. Sections

3and4presentthetwoalternativeskinningmodelsanddiscusstheirabilitytomodeltheintestines. Section5

concludesanddiscussessomefuture work.

2 A three component model for the intestines

Wewantto simulatetheevolutionof thebehaviorand thevisual aspectof theintestinesunder theaction

ofsurgicaltools,andto returnanappropriatehapticfeedbackviatheuserinterface. Tosimulatethephysical

behavior of the intestines, we use a mechanical model that animates their virtual skeleton. The intestines

skeleton can be represented asa curve with mechanical properties. Next, since the intestines may undergo

self-collisionsand collisions with their environment and some surgical tools, we use a detectionmodel based

onthemechanicalmodelto computethese collisionsand theirresponse. Finally,weprovideaskinningmodel

in orderto representanddisplaytheshapeoftheintestines (butwith nomechanicsat this level). Webriey

presentour rstmodel (furtherdetails canbefound in[13])and discussthe awsof theparametricskinning

modelweused.

2.1 MechanicalModel

WeuseacubicCatmull-Rom[9]segmentedsplinetodenetheintestinesaxis(theirskeleton). Thisdenition

allowstheusertocontroltheshapeofthesplinejust bymodifyingthecontrolpoints. Inorder toanimatethe

spline,weuseaLagrangianformalisminspiredfrom[17]. This formalismtakesinto accountthecontinuity of

theobjectandthusenablesacontinuousmassdistributionalongthecurve. Moreover,itallowsexternalactions

and/orconstraintstooccuranywherealongthespline.

The formulation of the equations leadsto the resolution of a systemof the form MA = B, namely the

resolutionof A=M 1

B, withM thegeneralizedmassmatrix, Athe accelerationof thedegreesof freedom

(the control points), and B the forces. It waspointedout that M is a diagonalblock matrixwith identical

diagonalblocksM. M is asymmetric, time-independent matrix,allowingthe pre-computationof theinverse

matrix, yielding in a faster resolution of the equation system. Besides, we take benetof the locality of the

Catmull-Romspline: For anycubic spline, the matrixM is band. Finally, the system resolution complexity

becomes O(n). Once thissystem solved,weget theaccelerationofthe degreesof freedom. Weuse aRunge-

Kuttaintegrationschemetoestimatethenewvelocitiesandnewpositions ofthecontrolpoints.

2.2 Collision/Self-collisionModel

Ourcollisiondetectionmethodusesanapproximationoftheobjectshapebyboundingspheres. A3Dgrid

thatdiscretizesthespaceisusedtospeeduptheprocess: Collisionsonlyneedtobeperformedbetweenspheres

belongingtoasamevoxel. Thisprocessisiteratedforeachsphereaddedtothegrid. Sincethepenaltymethods

arewellsuitedtodeformableobjects,theyareusedforcollisionresponse(see[11]formoredetails).

(4)

uniformlyaccordingtotheparametricabscissa(seeFigure1). Whenasurgicaltoolinteractswiththeintestines,

the intestines spheres ensure the detection of the collision and provide penalty forces as reaction response,

yieldinginthemovementand/ordeformationoftheintestinesandinforcesreturnedbythefeedbackdevice.

Neighbor spheres

Self-collision

Figure 1: Spheredistributionalongtheintestinesaxis.

Self-collisionsbetweenintestinesfoldsaremanagedbyusingthesamemethod. Nevertheless,inthedetected

collisions,wemustdiscardthespheresthatareneighborsalongtheintestinesaxis(seeFigure1).

2.3 Skinning Model

Themechanicalmodelrunsat100Hzormore,independentlyfromtheskinningmodel,displayedat25Hz.

Ateachdisplaystep(thatmaybelargerthanthesimulationstep),askinningoftheintestinesmustbegenerated.

Itmustdependupontheposition andtheplacementoftheirskeleton. Inotherwords,theskeletondescribing

thegeneralmotionoftheobject,theskinninghastofollowit,providingtheresultingshapeoftheobject.

Dierent methodsexist to deneavolume, someofthem aremoreadapted toatubular volume basedon

alinearskeleton. Theideais to createa volume delimited byasurfacearoundthe skeleton. Inourprevious

model[13], wedenedtheintestines surfaceasaparametricsurfacearoundthediscreteskeletonbasedonthe

splinepoints,atadistancerepresentingtheintestinesradius. Thisrepresentationallowedaninteractivedisplay.

Nevertheless,it suersfrom twomajordrawbacks: First,theresultingshapeisimperfect forhigh curvatures,

andsuersfromtangentandvolumediscontinuities;second,there isnodirectmethodtodeformitinorderto

modelexactcontactbetweensurfaces.

On theother hand, implicit surfaces particularly suit organicshapes, since theycontrol a volume around

theskeleton,enablingrealisticcurvedshapesandgeometriccontactbetweensurfaces[7]. Theremainderofthe

articlepresentstwomethodsbasedonimplicitsurfacesforthedisplayoftheintestines.

3 Implicit Intestines using Point-Skeletons

3.1 Method

An implicit surfaceisdened by theset of pointssuchthat the valueofa giveneld function f forthese

pointsP equalsto a given valuee, namelythe iso-value,for which the iso-surfaceis drawn: f(P)=e. The

evaluation of theeld function for anypointallowsastraightforwarddetectionof theinterior/exteriorof the

object. Thischeckingsimpliesthecollisiondetectionbetweenobjects,andthustheblendingcontrol.

One possibility to generate an implicit surface from a skeleton is to use distance surfaces [4]. Distance

surfacescomputetheeld valueat agivenpoint P fromits distanceto itsclosestpointontheskeleton. The

blendingof severalskeletonscontributionsisdirectly performedbysummingtheireld values. Thistechnique

thensuppressessurfacefoldsatthejointofskeletons,but maycreatebulges[3].

Wedenetheintestinessurfaceasanimplicitsurfacegeneratedbydiscretepoint-skeletonspositionedalong

thespline. Thestrengthoftheireldvalueistunedaccordingtotheradiusoftheintestines(seeFigure2(a)).

Duringtheanimation,theskeletonshapevariesasthesplinepointsmove. However,iftwopointsoftheelastic

spline are too far from each other, there is no moreblending between them, thus separatingthe objectinto

twoparts. In order to avoidthese topology changes, the numberof sampling points mustchange during the

simulation. This is done by adaptivelyplacing spheresalongthe curve at intervalswhich rangeis afunction

of the spline segment length and the radius of the spheres. Next, we use areal-time implementation of the

marchingcubesalgorithmproposedby[19]forthevisualizationoftheimplicitsurface.

(5)

Theblendingpropertyoftheimplicitsurfaceensuresacontinuousshapefortheintestinesmodel(seeFigure

2(b)). However,itisparticularlydiÆculttoprovideaconstantradius. Indeed,thevariationofthenumberof

skeletonsleadstouctuationsofthe geometry(thiscouldrepresentthespasmsofthe intestines,but wewant

tocontrolthem). Moreover,caremust betaken toavoidblendingbetweennon-consecutiveparts. Itrequires

theuseoftheblendinggraph. Unfortunately,inourimplementation,theblendingcontrolcannotbedoneata

suÆcientratedueto themarchingcubemethodto displaythesurface.

qi

qj

(a)Model. (b)Interactivedisplay.

Figure2: Implicitintestinesusingpoint-skeletons.

4 Implicit Intestines generated by a Convolution Surface

4.1 Method

In order to avoidthe aws of avarying numberof point-skeletons, we must use thewhole skeleton curve

orat least alinearapproximationof this curveto generatethe intestines geometry. Onepossibility is to use

convolutionsurfaces [3], andto representtheshapeasaset ofconnectedconvolutionsegment-skeletons. This

representation avoids bulges on the surface that coats the segments, and allows the number of segments to

changeateachtimestepwithoutcreatingjumps ontheimplicitsurfacegeometry,aslongasnewlyintroduced

verticesaresmoothlytranslated.

Forasingleconvolutionsegment-skeleton, theeldvalueatapointP isthesumofthecontribution ofall

thepoint-skeletonsalongthesegment. Thisintegralhasaclosed-formsolutionforvariouspoint-skeletonkernel

functions[8,15,18]. Thesolutionproposedin[8]isthefastest,anditsexpressionis:

f(P)= sin

1 sin

2

d 2

(P;H)

(4.1)

withd(P;H)thedistancebetweenP anditsprojectionH ontheskeletonsegment,

1 and

2

thesignedangles

betweenthis axisofprojectionandthelineslinkingP totheextremitiesoftheskeletonsegment.

A straightforwardwayof modeling asmoothcomplexsurfacewouldthen consist in summingallthe indi-

vidual eld contributionsof the simple segment-skeletons. This would nothowever preventthe surfacefrom

blendingbetweennon-consecutiveparts. Therefore, weusethecontrolledblendingtechniquedevelopedin [1],

thatrstneedstodescribethesamplingoftheimplicitsurface.

Todisplaythesurfacesatinteractiverate,weusetheseed-basedmethodof[8]whichtakesbenetfromthe

temporalcoherence. Seedsmigratetowardsthesurfaceaccordingtotheiso-surface. Initially,theystartfromthe

skeleton,andintheremainderofthesimulation,theystartfromtheirpreviouspositions. Thediscretizationof

thesurfacetriangulationisadaptive,allowingthesurfacetoberenderedatdierentlevelsofdetail. Toproduce

theadaptivesampling, aroughpolygonizationisinitially attachedto afew samplingpoints,andthen rened

recursivelybyauniform subdivision untilacertaincriterionissatised.

(6)

assigns to every point P

i 2 IR

3

on the surface a parameter u

i

along the skeleton, which corresponds to the

attachedpointof theseedaxis. A localportion ofthe skeleton aroundu

i

is usedto computef(P

i

),which is

aneighborhood of constant size D around u

i

. The eld value is then computed exactlyas before, with the

exception that only the useful partof the skeleton is considered in this case. A constrain on D is D < 2R

(Ris theintestinesradius),sincetwopointsonthesurfaceseparatedbyadistancegreaterthan2R mustnot

inuenceeachother.

4.2 Results

With this model, thevisual rendering of theintestines is satisfying: There is no blending, and no bulges

created(seeFigure 3(b)). However,the computation timeis stilltooslowifwe wantane discretization of

thesurfaceobject.

j i

j i

j i

D u

u P

P D

(a)Model. (b)Displayofintestinesonlyanimatedwithgravity.

Figure3: Implicitintestinesgeneratedbyasubdivisioncurve.

5 Conclusion

Inthispaper,wepresentedamethodtosimulate atinteractiveratetheintestinesin theabdominal cavity,

for surgical training purposes. The intestines are very deformable objects, and during a surgery, they may

undergolargedisplacementsand multiple interactions. Theintestinesmodelisbasedonaskeletondened by

aspline tocomputeitsmechanicalmotionfromtheLagrangeformalism. Collisionspheresarethenusedfora

fastcollisiondetection,providingreal-timefeedbackviathehapticdevice. Thepaperfocusedontheskinningof

thismodel. Itcomparedtwoimplicitsolutionsthatoerquitegoodresultsintermsofvisualeectsconcerning

the movements and deformations of the intestines. We showed that this skinning by implicit surfaces may

be attractivein the context of a surgical simulator. Theintroduction of adaptive implicit surfaces based on

convolutionisinterestingsincetheanimationcouldbedisplayedatdierentlevelsofdetailtoobtaininteractive

ratewhatevertheworkstation.

Currently, only the discretization of the surface triangulation is adaptive. A future work could consist

in dynamically adapting the discretization of the skeleton according to the varying curvature. It could be

interestingto dealwith bothaspects in order to tune theparameters to obtaincomputations anddisplay in

real-time. Moreover, it could be possible to add the contact surfaces presented in [1] so that the contact is

betterhandled. Ontheother hand,self-collisionscouldbedetectedmoreprecisely bytakingintoaccountthe

informationprovidedbytheimplicitsurface.

(7)

ThisworkissupportedbyINRIA(FrenchNationalInstituteforResearchinComputerScienceandControl)

aspartof the ARC SCI (research action forIntestines SurgerySimulator). The authors would like to thank

people ofIRCAD(French DigestiveCancerResearchInstitute)fortheirconstructivediscussions.

References

[1] A. Angelidis, P. Jepp, and M.-P. Cani. Implicit modeling with skeleton curves: Controlled blending in

contactsituations. ShapeModeling International,2002.

[2] J.Bloomenthal,C.Bajaj,J.Blinn,M.-P. Cani,A.Rockwood,B.Wyvill,and G.Wyvill. Introduction to

ImplicitSurfaces. MorganKaufmann,July1997.

[3] J.BloomenthalandK.Shoemake. Convolutionsurfaces. Computer Graphics,25(2):251{256, July1991.

[4] J.BloomenthalandB.Wyvill.Interactivetechniquesforimplicitmodeling.ComputerGraphics,24(2):109{

116,March1990.

[5] M.Bro-NielsenandS. Cotin. Real-time volumetricdeformablemodels forsurgerysimulationusingnite

elementsandcondensation. Computer Graphics Forum 15(3):57{66, 1996.

[6] M.-P. Cani. An implicit formulation for precise contact modeling between exible solids. In Computer

Graphics,pages313{320.ACM,aug1993. PublishedunderthenameM.-P.Gascuel.

[7] M.-P.CaniandM.Desbrun. Animationofdeformablemodelsusingimplicitsurfaces. IEEETransactions

onVisualizationandComputerGraphics,1(3),March1997.PublishedunderthenameM.-P.Cani-Gascuel.

[8] M.-P.CaniandS. Hornus. Subdivisioncurveprimitives: anewsolutionforinteractiveimplicitmodeling.

InShapeModellingInternational, May2001.

[9] R.CatmullandR.Rom.Aclassoflocalinterpolatingsplines.ComputerAidedGeometricDesign,Academic

Press,1974.

[10] S.Cotin, H. Delingette,and N. Ayache. Real timevolumetricdeformablemodelsforsurgery simulation.

VisualizationinBiomedical Computing, volume1131. Springer Verlag,1996.

[11] J. Davanne, P. Meseure, and C. Chaillou. Stable haptic interaction in a dynamic virtual environment.

SubmittedtoIROS'02,2002.

[12] G.Debunne, M. Desbrun,M.-P. Cani, and A.H.Barr. Dynamic real-time deformationsusing space and

timeadaptivesampling. Computer Graphics,August2001.

[13] L.France,J.Lenoir,P.Meseure,andC.Chaillou.Simulationofaminimallyinvasivesurgeryofintestines.

SubmittedtoVirtual RealityInternational Conference,2002.

[14] A. Haristis, D. Gillies, and C. Williams. Realistic generation and real time animation of images of the

humancolon. ComputerGraphics Forum1992, vol.II, 3:367{379,1992.

[15] S.Hornus,A.Angelidis,andM.-P.Cani.Implicitmodellingusingsubdivision-curves.TheVisualComputer,

2002.

[16] G. Picinbono, H. Delingette, and N. Ayache. Non-linear and anisotropic elastic soft tissue models for

medicalsimulation. IEEEInternational ConferenceRobotics andAutomation,May2001.

[17] Y.Remion,J.M.Nourrit,andD. Gillard. Dynamicanimationofspline likeobjects. WSCG,1999.

[18] A. Sherstyuk. Kernel functions in convolutionsurfaces: acomparativeanalysis. The Visual Computer,

15(4),1999.

[19] F. Triquet, P. Meseure, and C. Chaillou. Fast polygonization of implicit surfaces. WSCG (2):283{290,

February2001.

Références

Documents relatifs

7 By integrating the technology of the Internet (interconnection systems and integrated services) and the economics of interconnection agreements (pricing policies and

Researcherrecommended for the neccessarily of adopting normative levels for the pupelassessmentwith an objective ways.. 221 І - ةمدقم : نإ كلاتما

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

Instead, using the characterization of ridges through extremality coefficients, we show that ridges can be described by a singular curve in the two-dimensional parametric domain of

This paper presents the results of a collaborative investigation between the National Research Council of Canada (NRC), Queen's University through the Intelligent Sensing

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

In this study, we aim to evaluate the Weather Research and Forecasting (WRF) model for regional climate application over Tunisia, focusing in simulated

As mentioned in section 2, to compute the topology of a plane algebraic curve in generic position, we need to compute its critical fibers and one regular fiber between two critical