Page 219 - Demo

Generaldiscussionandfutureperspectives21712Figure11Avirtualimpressionofthepatient’sdentitionwascapturedwithanintra-oralscannerIncombinationwith3DprintingasplintwithanuniquefitonthelowerdentitionwascreatedandusedtoholdaQRmarker(left) TheQRcouldbetrackedbytheHoloLensduringahighcondylectomy(right)ARTIFICIALINTELLIGENCE-BASEDREGISTRATIONFutureprocedurescoulduseacombinationofArtificialIntelligence(AI)andcomputervisiontechniquestoimproveandautomatetheregistration Thiscouldeliminateusererrorsinceindicatingmarkersmakinga3Dscanor fabricateasplintwillnolongerberequiredVarioustoolkitsincludingtheVuforiatoolkitthatwasusedinthisthesisformarkertrackingarecurrentlybeingexploredaspossibleAIbasedregistrationmethods4243 ThesemethodsuseaDeepNeuralNetwork(DNN)andartificiallycreated2DimagesasaninputTheseimagesareautomaticallyconvertedfromthe3Dvirtualdata(egacranialshape)Byartificiallyadjustingvariousparameterssuchasscalelightningcolor,viewingangleandnoise(tosimulateblood stains orreflections) a dataset ofmillions of different 2Dimages canbegenerated43NextaDNNcouldbetrainedwiththisdatasetandwhenanew2DimagecapturedintheoperationtheatreisgiventotheDNNsystemthebestfittingregistrationparameterscouldbeestimatedThisAIsystemcorrelatesthe3Dplanningdatadirectlytoasetof2Dimagesofthesurgicalsitebyforexampleusing thebuilt-inHoloLenscamera Thismethodis verypromisingbecauseitdoesnotrequireuserinputisrelativelyfastrequiresonlya2DcamerafeedandcanbeperformedmultipletimesifnecessaryInitialresearchincollaborationwiththeElisabeth-TweeStedenhospitaliscurrentlybeingperformedbyourgroupHowever,futurestudiesarerequiredtoevaluatetheusabilityandaccuracyofthismethodfordifferentclinicalapplicationsThismethodcouldbeused forregistrationbutifitcanbeoptimizedandmade toworkreally fastit couldbeused for tracking as well Thismethod couldpossiblybeappliedtoregistertargetsthatarepronetodeformationsuchasmovingorganspatientswithswellingandsofttissue