Automated 3D nonlinear deformation procedure for determination of gross morphometric variability in human brain

摘要: Abstract We describe an automated method to register MRI volumetric datasets a digital human brain model. The technique employs3D non-linear warping based on the estimation of local deformation fields using cross-correlation invariant intensity featuresderived from image data. Results registration simple phantom, complex phantom and real MRIdata are presented. Anatomical variability is expressed with respect Talairach-like standardized brain-based coordinatesystem show that reduces inter-subject homologouspoints in space by 15% over linear methods. A 3D map shown. 1 INTRODUCTION New imaging modalities techniques, e.g., PET, functional (fMRI), SPECT, magnetoencephalography (MEG), andEEG have made it possible areas anatomy. Two aspects this workrequire integration data different individuals: 1) low signal associated cognitive activation (e.g., subtlechange cerebral blood flow (CBF) as measured PET) requires averaging across subjects improve statistical significanceofmeasured CBF changes'4"°. 2) Although high resolution techniques such fMRI now make measureactivation within single subject, will still be necessary compare results individuals order fully understandthe relationship between underlying gross morphology gyral For both situations weideally wish remove all morphological differences individual brains before considering distribution functionalinformation superimposed anatomical substrate. This for deforming one match another atall points, has typically been accomplished mapping into coordinatesystem24. Until recently, most centers used transformations only13'15'19'25. However, previous work24'23 shownthat even after mapping, there up 1.5 cm position cortical structures, which may representa significant source error when foci. shown9, average points throughout brain(cortical sub-cortical), 6-7mm not accounted registration.The objective paper present establishing morphometric ina population normal MRI. Non-linear homologous landmark matching2'9 beenpractical routine use deformation/warping model because subjectivity involved selecting precise locationand number define deformation. lead our group others (e.g.,1"6'20) consider fullyautomated, techniques. Our uses set itwith another, deformations derived neighbourhood correlation featurescalculated data3'4.To properly assess first essential well-defined coordinate where linearcomponent variation removed application affine transformation. Without priori knowledgeof variability, best minimum variance frame cannot defined since wholly dependent former.Therefore, we selected system very similar proposed Talairach24. implementationuses global transformation whereas Talairach employs 12 piece-wise (as implemented