Page 203 - Demo

Generaldiscussionandfutureperspectives20112allshapeswerescaledtothesamesizetocorrectforgrowthInadditionwedidnotdifferentiatebetweenmalesandfemalesSincewediscoveredinChapter2thatbothageandsexinfluencethecranialshapeitwouldbeveryinterestingifwecouldtrainthePCAmodelwiththislargerandmorestringently-selecteddataset(seealsoAppendixChapter2) TrainingthePCAmodelwithmorecranialvariations(byincludingabnormalcranialshapes)mayalsoimprovethemodel’srobustnessFigure1Variousvisualizationandanalysisoptionswhichareusedbythetreatmentteamandshowntopatientsduringclinicalconsultationandfollow-uptoobjectivelyevaluatecranialshapechangesInarecentstudybyourgroupamachinelearningtechnique ‘deeplearning’,wasusedtoclassifyasetof3DphotosasnormalscaphocephalytrigonocephalyorplagiocephalyOutof196photos195subjects(995%)werecorrectlyclassifiedandonlyoneplagiocephalypatientwasmisclassifiedasnormal3Thisprovestheenormouspotentialofdeeplearningmethodsfortheclassificationofcraniosynostosis45UnfortunatelyaconsequenceofdeeplearningisthatthefeaturesthatareusedtoclassifyapatientareunknownThisissatisfactoryforclassificationbutisnotideal during follow-upFeatures that describe a deformation(suchastheprincipalcomponentsdiscussedinChapter3)couldalsobeusedtoquantify the severity of the deformation and could thereforebeused during followupbycheckingwhetherornotacertaindeformationimprovesAlternatively