Manual
ComputationalAnatomyToolbox-CAT12
QUICKSTARTGUIDE ......................................................................................................................... 3
INTRODUCTIONANDOVERVIEW ...................................................................................................... 5
GETTINGSTARTED ............................................................................................................................ 5
DownloadandInstallation
5
StartingtheToolbox
6
BasicVBManalysis(overview)
6
BASICVBMANALYSIS(DETAILEDDESCRIPTION) ............................................................................... 9
9
PreprocessingData
FirstModule:SegmentData
SecondModule:Displayonesliceforallimages
9
11
ThirdModule:Checksamplehomogeneity
FourthModule:Smooth
FifthModule:EstimateTotalIntracranialVolume(TIV)
11
13
13
14
15
BuildingtheStatisticalModel
Two-sampleT-Test
FullFactorialModel(fora2x2Anova)
MultipleRegression(Linear)
16
17
MultipleRegression(Polynomial)
FullFactorialModel(Interaction)
FullFactorialModel(PolynomialInteraction)
18
19
20
EstimatingtheStatisticalModel
CheckingforDesignOrthogonality
DefiningContrasts
21
21
24
SPECIALCASES................................................................................................................................ 28
1
28
CAT12forlongitudinaldata
ChangeSettingsforPreprocessing
PreprocessingofLongitudinalData
StatisticalAnalysisofLongitudinalDatainOneGroup
29
29
30
StatisticalAnalysisofLongitudinalDatainTwoGroups
31
AlteredWorkflowsforVBM-analyses
35
Adaptingtheworkflows
36
36
37
CustomizedTissueProbabilityMaps
CustomizedDARTEL-template
OTHERVARIANTSOFCOMPUTATIONALMORPHOMETRY .............................................................. 39
Deformation-basedmorphometry(DBM)
39
Surface-basedmorphometry(SBM)
40
Regionofinterest(ROI)analysis
43
ADDITIONALINFORMATIONONNATIVE,NORMALIZEDANDMODULATEDVOLUMES.................... 45
NAMINGCONVENTIONOFOUTPUTFILES....................................................................................... 47
CALLINGCATFROMTHEUNIXCOMMANDLINE ............................................................................. 49
TECHNICALINFORMATION ............................................................................................................. 49
2
Quickstartguide
Errorsduringpreprocessing
Please use theReport Errorfunction if any errors during preprocessing occurred. You have to first
selectthe"err"directorythatcanbefoundinthefolderofthefaileddatasetandfinallytheindicated
zip-fileinthemailshouldbeattachedmanually.
VBMdata
• Segmentdatausingdefaults(forlongitudinaldatauselongitudinalpipeline)
• CheckdataqualityusingsamplehomogeneityforVBMdata
• Smoothdata(suggestedstartingvalue8mm)
• Estimatetotalintracranialvolume(TIV)inordertocorrectfordifferentheadsizeandvolume
• Build2nd-levelmodel:
o Use"Fullfactorial"forcross-sectionaldata
o Use"Flexiblefactorial"forlongitudinaldata
o UseTIVascovariate(confound)tocorrectfordifferentbrainsizesandselectcentering
withoverallmean
o Selectthresholdmaskingwithanabsolutevalueof0.1.Thisthresholdcanbeincreased
inthefinalanalysisto0.2oreven0.25.
• Estimatemodel
• Checkdesignorthogonalityusingthe“Review”functionintheSPMGUI.Ifyoufindaconsiderable
correlation between TIV and any other parameter of interest it is recommended to rather use
globalscalingwithTIV.Checkthesection“Buildthestatisticalmodel”formoredetails
• OptionallytransformandthresholdSPM-mapsto(log-scaled)p-mapsorcorrelationmaps
• Optionally estimate results for ROI analysis using“Analyze ROIs”. Here, the SPM.mat file of an
alreadyestimatedstatisticaldesignwillbeused.Pleasechecktheonlinehelp“Atlascreationand
ROIbasedanalysis”formoreinformation.
3
Additionalsurfacedata
• Segmentdataandadditionallyselect"Surfaceandthicknessestimation"in"Writingoptions"
• Optionally extract additional surface parameters (e.g. suclus depth, gyrification index, cortical
complexity)
• Resample and smooth surface data (suggested starting value 15mm for cortical thickness and
20mmforfoldingmeasures)
• Checkdataqualityusingsamplehomogeneityforsurfacedata
• Build 2nd-level model: Use "Full factorial" for cross-sectional data and "Flexible factorial" for
longitudinaldata
• Estimatesurfacemodelforeachhemisphere
• OptionallyDisplaysurfaceresultsforbothhemispheres.Selectresults(preferablysavedaslog-p
maps using“Transform and threshold SPM-surfaces”) for left and right hemisphere in order to
showrenderviewsofyourresults
• OptionallyExtractROI-basedSurfaceValuessuchasthickness,gyrificationorfractaldimensionto
provideROIanalysis.
• Optionally estimate results for ROI analysis using“Analyze ROIs”. Here, the SPM.mat file of an
alreadyestimatedstatisticaldesignwillbeused.Pleasechecktheonlinehelp“Atlascreationand
ROIbasedanalysis”formoreinformation.
4
IntroductionandOverview
This manual is intended to help any user to perform a computational anatomy analysis using the
CAT12 Toolbox. Although it will mainly focus on voxel-based morphometry (VBM) other variants of
computationalanalysissuchasdeformation-basedmorphometry(DBM),surface-basedmorphometry
(SBM),andregionofinterest(ROI)morphometricanalysiswillbealsointroducedandcanbeapplied
withafewchanges.
Basicallythemanualmaybedividedintofourmainsections:
• Naturally, a quick guide of how to get started is given at the beginning. This section provides
informationhowtodownloadandinstallthesoftwareandstarttheToolbox.Furthermore,ashort
overviewonthestepsofaVBManalysisisgiven.
• AdetaileddescriptionofabasicVBManalysisissubsequentlygiven,whichwillguidetheuserstep
by step through the whole process – from preprocessing to the selection of contrasts. This
descriptionshouldprovideallnecessaryinformationtoanalyzemoststudiessuccessfully.
• There are a few special cases of VBM analyses, for which the basic analysis workflow has to be
adapted.Thesecasesarelongitudinalstudiesandstudiesinchildrenorspecialpatientpopulations.
Relevant changes to a basic VBM analysis are described here and a description of how to apply
these changes is provided. Importantly, only the changes are described – steps like for example
qualitycontrolorsmoothingarethesameasinthebasicanalysisandnotdescribedasecondtime.
• The manual closes with information on native, normalized and modulated volumes, which
determines how the results may be interpreted. Furthermore an overview of the naming
conventionsusedaswellastechnicalinformationisgiven.
GettingStarted
DOWNLOADANDINSTALLATION
• TheCAT12ToolboxrunswithinSPM12.Thatis,SPM12needstobeinstalledandaddedtoyour
Matlab search path before the CAT12 Toolbox can be installed (see
http://www.fil.ion.ucl.ac.uk/spm/andhttp://en.wikibooks.org/wiki/SPM).
• Download (http://dbm.neuro.uni-jena.de/cat12/) and unzip the CAT12 Toolbox. You will get a
foldernamed“cat12”,whichcontainsvariousmatlabfilesandcompiledscripts.Copythefolder
“cat12”intotheSPM12“toolbox”folder.
5
STARTINGTHETOOLBOX
• StartMatlab
• StartSPM12(i.e.,type“spmfmri”)
• Select“cat12”fromtheSPMmenu(seeFigure1).Youwillfindthedrop-downmenubetweenthe
“Display”andthe“Help”button(youcanalsocalltheToolboxdirectlybytyping“cat12”onthe
Matlabcommandline).ThiswillopentheCAT12Toolboxasadditionalwindow(Fig.2).
Figure1:SPMmenu
Figure2:CAT12Window
BASICVBMANALYSIS(OVERVIEW)
TheCAT12Toolboxcomeswithdifferentmodules,whichmaybeusedforananalysis.Usually,aVBM
analysiscomprisesthefollowingsteps
(a)Preprocessing:
1. T1imagesarenormalizedtoatemplatespaceandsegmentedintograymatter(GM),white
matter(WM)andcerebrospinalfluid(CSF).Thepreprocessingparameterscanbeadjustedvia
themodule“SegmentData”.
2. After the preprocessing is finished, a quality check is highly recommended. This can be
achievedviathemodules“Displayonesliceforallimages”and“Checksamplehomogeneity”.
6
Both options are located in the CAT12 window under “Check Data Quality”. Furthermore,
quality parameters are estimated and saved in xml-files for each data set during
preprocessing.ThesequalityparametersarealsoprintedonthereportPDF-pageandcanbe
additionallyusedinthemodule“Checksamplehomogeneity”.
3. BeforeenteringtheGMimagesintoastatisticalmodel,imagedataneedtobesmoothed.Of
note,thisstepisnotimplementedintotheCAT12ToolboxbutachievedviathestandardSPM
module“Smooth”.
(b)Statisticalanalysis:
4. The smoothed GM images are entered into a statistical analysis. This requires building a
statistical model (e.g., T-Tests, ANOVAs, multiple regressions). This is done by the standard
SPMmodules“Specify2ndLevel”or“BasicModels”intheCAT12windowcoveringthesame
function.
5. The statistical model is estimated. This is done by the standard SPM module “Estimate”
(exceptforsurface-baseddatawherethefunction“EstimateSurfaceModels”shouldbeused
instead.
6. If you have used total intracranial volume (TIV) as confound in your model to correct for
different brain sizes it is necessary to check whether TIV reveals a considerable correlation
withanyotherparameterofinterestandratheruseglobalscalingasalternativeapproach.
7. After estimating the statistical model, contrasts will be defined to get the results of the
analysis.ThisisdonebythestandardSPMmodule“Results”.
Thesequenceof“preprocessing!qualitycheck!smoothing!statisticalanalysis”remainsthe
sameforeveryVBManalysis,evenwhendifferentstepsareadapted(see“specialcases”).
AfewwordsabouttheBatchEditor…
− As soon as you select a module from the CAT12 Toolbox menu, a new window (the Batch
Editor)willopen.TheBatchEditoristheenvironmentwhereyouwillsetupyouranalysis(see
Figure3).Forexample,an“<-X”indicateswhereyouneedtoselectfiles(e.g.,yourimagefiles,
thetemplate,etc.).Otherparametershaveeitherdefaultsettings(whichcanbemodified)or
requireinput(e.g.,choosingbetweendifferentoptions,providingtextornumericvalues,etc.).
− Onceallmissingparametersareset,agreenarrowwillappearonthetopofthewindow(the
currentsnapshotsinFigure3showthearrowstillingray).Clickthisarrowtorunthemodule
or select “File ! Run Batch”. It is very useful to save the settings before you run the batch
(clickonthedisksymbolorselect“File!SaveBatch”).
7
− Of note, you can always find helpful information and parameter-specific explanations at the
bottomoftheBatchEditorwindow.1
− Allsettingscanbesavedeitheras.matfileoras.mscriptfileandreloadedforlateruse.The
.mscriptfilehastheadvantagetobeeditablewithatexteditor.
Figure 3: The Batch Editor is the environment where the analysis is set up. Left: For all settings
marked with “<-X”, files have to be selected (“Select Files”). Right: Parameters can be edited and
adapted(“EditValue”).
1
AdditionalCAT12-relatedinformationcanbefoundbyselecting“VBMWebsite“intheCAT12window(Tools→Internet
→VBMWebsite”).Thiswillopenawebsite.Here,lookfor“VBMsubpages”ontheright.
8
BasicVBMAnalysis(detaileddescription)
PreprocessingData
FIRSTMODULE:SEGMENTDATA
Please note that additional parameters for expert users will be displayed in the GUI if you set the
optioncat.extopts.expertguito“1”incat_defaults.morcallcat12by:
cat12(‘expert’)
CAT12!Preprocessing!SegmentData
Parameters:
o Volumes<-X!SelectFiles![selectthenewfiles]!Done
-
-
Selectonevolumeforeachsubject.AstheToolboxdoesnotsupportmultispectral
data(i.e.,differentimagingmethodsforthesamebrain,suchasT1-,T2-,diffusionweightedorCTimages),itisrecommendedtochooseaT1-weightedimage.
Importantly,theimagesneedtobeinthesameorientationasthepriors;youcan
double-check and correct via using “Display” in the SPM menu. The priors are
locatedinyourSPMfolder“SPM12!tpm!TPM.nii”)
o Splitjobintoseparateprocesses![usedefaultsormodify]
- Inordertousemulti-threadingtheCAT12segmentationjobwithmultiplesubjects
can be split into separate processes that run in the background. You can even
closeMatlab,whichwillnotaffecttheprocessesthatwillruninthebackground
without GUI. If you don’t want to run processes in the background then set this
valueto0.
- Keep in mind that each process needs about 1.5..2GB of RAM, which should be
consideredtochoosetheappropriatenumberofprocesses.
- Please further note that no additional modules in the batch can be run except
CAT12segmentation.Anydependencieswillbebrokenforsubsequentmodules.
o OptionsforinitialSPM12affineregistration![usedefaultsormodify]
- The defaults provide a solid starting point. The SPM12 tissue probability maps
(TPMs)areusedfortheinitialspatialregistrationandsegmentation.Alternatively,
customizedTPMscanbechosen(e.g.forchildrendata)thatwerecreatedwiththe
Template-O-Matic(TOM)Toolbox.
o ExtendedoptionsforCAT12segmentation![usedefaultsormodify]
9
Again,thedefaultsprovideasolidstartingpoint.Usingtheextendedoptionsyou
canadjustspecialparametersorthestrengthofdifferentcorrections("0"means
nocorrectionand"0.5"isthedefaultvaluethatworksbestforalargevarietyof
data).
- CAT12 provides a template for the high-dimensional DARTEL registration that
should work for most data. However, a customized DARTEL template can be
selected (e.g. for children data) that was created using the DARTEL toolbox. See
the section “Customized DARTEL-template” for more information about the
necessarysteps.
o Writingoptions![usedefaultsormodify]
-
-
-
-
-
-
For GM, and WM image volumes see at the end of the document: “Additional
Informationonnative,normalizedandmodulatednormalizedvolumes”.Note:The
default option “Modulated normalized” will result in an analysis of relative
differences in regional GM volume, that have to be corrected for individual brain
sizeinthestatisticalanalysisusingtotalintracranialvolume(TIV).
A Bias, noise and globally intensity corrected T1 image, in which MRI
inhomogeneities and noise are removed and intensities are globally normalized,
can be written in normalized space. This is useful for quality control and also to
createanaverageimageofallnormalizedT1imagesinordertodisplay/overlay
theresults.Note:ForabasicVBManalysisusethedefaults.
A partial volume effect (PVE) label image volume can also be written in
normalized or native space or as a DARTEL export file. This is useful for quality
control and also for future applications using this image to reconstruct surfaces.
Note:ForabasicVBManalysisusethedefaults.
TheJacobiandeterminantforeachvoxelcanbewritteninnormalizedspace.This
informationcanbeusedtodoaDeformation-BasedMorphometry(DBM)analysis.
Note:ForabasicVBManalysisthisisnotneeded.
Finally, deformation fields can be written. This option is useful to re-apply
normalization parameters to other co-registered images (e.g. fMRI or DTI data).
Note:ForabasicVBManalysisthisisnotneeded.
Note:Ifsegmentationfailsthisoftenoccursduetoanunsuccessfulinitialspatialregistration.Inthis
caseyoucantrytosettheorigin(anteriorcommissure)intheDisplaytool.Roughlysetthecursorto
theanteriorcommissureandpress“SetOrigin”Thenowdisplayedcorrectioninthecoordinatescan
beappliedtotheimagebyusingthebutton“Reorient”.Thisprocedurehastoberepeatedforeach
datasetseparately.
10
SECONDMODULE:DISPLAYONESLICEFORALLIMAGES
CAT12!Checkdataquality!Displayonesliceforallimages
Parameters:
o Sampledata<-X!SelectFiles![selectthenewfiles]!Done
-
Selectthenewlywrittendata[e.g.the“wm*”files,whicharethenormalizedbias
correctedvolumes].Thistoolwilldisplayonehorizontalsliceforeachsubject,thus
givingagoodoverviewifthesegmentationandnormalizationproceduresyielded
reasonableresults.Forexample,ifthenativevolumehadartifactsorifthenative
volumeshadawrongorientation,theresultsmaylookodd.Solutions:Use“Check
Reg”fromtheSPMmainmenutomakesurethatthenativeimageshavethesame
orientationliketheMNITemplate(“SPM!templates!T1”).Adjustifnecessary
using“Display”fromtheSPMmainmenu.
o Proportionalscaling![usedefaultsormodify]
- Check“yes”,ifyoudisplayT1volumes.
o Spatialorientation
o Showsliceinmm![usedefaultsormodify]
-
This module displays horizontal slices. This default setting provides a good
overview.
THIRDMODULE:CHECKSAMPLEHOMOGENEITY
CAT12!Checkdataquality!Checksamplehomogeneity!VBMdata
Parameters:
o Data!New:Sampledata<-X!SelectFiles![selectgraymattervolumes]!Done
- Selectthenewlywrittendata[e.g.the“mwp1*”files,whicharethemodulated(m)
normalized (w) GM segments (p1)]. It is recommended to use the unsmoothed
segmentations that provide more anatomical details. This tool visualizes the
correlationbetweenthevolumesusingaboxplotandcorrelationmatrices.Thus,it
willhelpidentifyingoutliers.Anyoutliershouldbecarefullyinspectedforartifacts
or pre-processing errors using “Check worst data” in the GUI. If you specify
differentsamplesthemeancorrelationisdisplayedinseparateboxplotsforeach
sample.
11
o Loadqualitymeasures(optional)![optionallyselectxml-fileswithqualitymeasures]
- Optionallyselectthexml-filesthataresavedforeachdataset.Thesefilescontain
usefulinformationaboutsomeestimatedqualitymeasuresthatcanbealsoused
forcheckingsamplehomogeneity.Pleasenote,thattheorderofthexml-filesmust
bethesameastheotherdatafiles.
o Separationinmm![usedefaultsormodify]
- Tospeedupcalculationsyoucandefinethatcorrelationisestimatedonlyeveryx
voxel. Smaller values give slightly more accurate correlation, but will be much
slower.
o Nuisance![enternuisancevariablesifapplicable]
- For each nuisance variable which you want to remove from the data prior to
calculating the correlation, select “New: Nuisance” and enter a vector with the
respective variable for each subject (e.g. age in years). All variables have to be
entered in the same order as the respective volumes. You can also type
“spm_load” to upload a *txt file with the covariates in the same order as the
volumes.ApotentialnuisanceparametercanbeTIVifyouchecksegmenteddata
withthedefaultmodulation.
Awindowwithacorrelationmatrixwillopen,whichdepictthecorrelationbetweenthevolumes.The
correlationmatrixshowsthecorrelationbetweenallvolumes.Highcorrelationvaluesmeanthatyour
dataaremoresimilartoeachother.Ifyouclickinthecorrelationmatrixthecorrespondingdatapairs
willbedisplayedattherightbottomcornerandallowamorecarefulinspection.Thesliderbelowthe
image changes the displayed slice. The popup menus at the right top corner provide more options.
Hereyoucanselectothermeasuresthataredisplayedintheboxplot(e.g.optionallyqualitymeasures
ifloadedsuchasnoise,bias,weightedoverallimagequality),canchangetheorderofthecorrelation
matrix(byfilenameormeancorrelation).Finally,theworstdatacanbeshownintheSPMgraphics
windowtocheckthedatamorecarefully.
TheboxplotintheSPMgraphicswindowaveragesallcorrelationvaluesforeachsubjectandshows
thehomogeneityofyoursample.Asmalloverallcorrelationintheboxplotnotalwaysmeansthatthis
volume is an outlier or contains an artifact. If there are no artifacts in the image and if the image
qualityisreasonableyoudon’thavetoexcludethisvolumefromthesample.Thistoolisintendedto
utilitizetheprocessofqualitycheckingandthereisnoclearcriteriadefinedtoexcludeavolumeonly
basedontheoverallcorrelationvalue.However,volumeswithanoticeableloweroverallcorrelation
(e.g.belowtwostandarddeviations)areindicatedandshouldbecheckedmorecarefully.
If you have loaded quality measures you can also display the Mahalanobis distance between two
measures:meancorrelationandweightedoverallimagequality.Thesetwoarethemostimportant
measurestoevaluateimagequality.Meancorrelationmeasuresthehomogeneityofyourdatathat
will be used for statistical analysis and is therefore a measure about image quality after preprocessing.Datathatdeviatefromyoursamplewillincreasevarianceandthereforeminimizeeffect
sizeandstatisticalpower.Weightedoverallimagequalityincontrastcombinesmeasuresofnoiseand
spatial resolution of the images before pre-processing. Although CAT12 uses effective de-noising
12
approaches (e.g. spatial adaptive non-local means filter) pre-processed images will be also affected
andshouldbechecked.
TheMahalanobisdistanceallowstocombinethesetwomeasuresofimagequalitybeforeandafter
pre-processing.IntheMahalanobisplotthedistanceiscolor-codedandeachpointcanbeselectedto
obtainthefilenameanddisplaytheselectedslicetocheckdatamorecarefully.
FOURTHMODULE:SMOOTH
SPMmenu!Smooth
Parameters:
o ImagestoSmooth<-X!SelectFiles![selectgreymattervolumes]!Done
- Select the newly written data [e.g. the “mwp1” files, which are the
modulated(m)normalized(w)greymattersegments(p1)].
o FWHM![usedefaultsormodify]
- 8-12mmkernelsarewidelyusedforVBM.Tousethissettingselect“edit
value”andtype“888”(or“121212”,respectively)forakernelwith8mm
(with12mm)FWHM.
o DataType![usedefaultsormodify]
o FilenamePrefix![usedefaultsormodify]
FIFTHMODULE:ESTIMATETOTALINTRACRANIALVOLUME(TIV)
CAT12!StatisticalAnalysis!EstimateTIV
Parameters:
o XMLfiles<-X!SelectFiles![selectxml-files]!Done
- Selectthexml-filesinthereport-folder[e.g.the“cat_*.xml”].
o Savevalues!TIVonly
- ThisoptionwillsavetheTIVvaluesforeachdatasetinthesameorderas
the selected xml-files. Optionally you can also save the global values for
eachtissueclass,whichmightbeinterestingforfurtheranalysis,butisnot
recommendedifyouareinterestedinonlyusingTIVascovariate.
o Outputfile![usedefaultsormodify]
13
PleasenotethatTIVisstronglyrecommendedtouseascovariateforallVBManalysisinorderto
correctfordifferentbrainsizes.Fordeformation-orsurface-baseddatathisstepisnotnecessary.
PleasealsocheckthatTIVisnotcorrelatingtoomuchwithyourparametersofinterest(pleasetake
carethatyouuse“Centering”with“Overallmean”,otherwisethecheckfororthogonalityinSPMis
sometimesnotworkingcorrectly).InthatcaseyoushouldratheruseglobalscalingwithTIV.
BuildingtheStatisticalModel
Althoughtherearemanypotentialdesignsofferedinthe2nd-levelanalysisIrecommendtousethe
“Full factorial” design because it covers most statistical designs. For cross-sectional VBM data you
haveusually1..nsamplesandoptionallycovariatesandnuisanceparameters:
Numberoffactorlevels
Numberofcovariates StatisticalModel
1
0
one-samplet-test
1
1
singleregression
1
>1
multipleregression
2
0
two-samplet-test
>2
0
Anova
>1
>0
AnCova (for nuisance parameters) or
Interaction(forcovariates)
14
TWO-SAMPLET-TEST
CAT12→StatisticalAnalysis→BasicModels
*
Parameters:
o
o
Directory <-X ! Select Files ! [select the
workingdirectoryforyouranalysis]!Done
Design!“Two-samplet-test”
"
Group 1 scans ! Select Files !
[select the smoothed grey matter
volumes for group 1; following this
scriptthesewillbethe“smwp1”files]
!Done
"
Group 2 scans ! Select Files !
[select the smoothed grey matter
volumesforgroup2]!Done
"
Independence!Yes
"
Variance!EqualorUnequal
"
Grandmeanscaling!No
"
ANCOVA!No
o
Covariates*(seetextbox)
o
Masking
It is strongly recommended to always use total
intracranial volume (TIV) as covariate if you use
modulated data in VBM in order to correct for
differentbrainsizes.ForsurfaceanalysisandDBM
thisisnotnecessary.
•
•
Covariates!NewCovariate
•
•
•
Name<-X!SpecifyText(e.g.,“age”)
Vector <-X ! enter the values of the
covariates (e.g., TIV and optionally age in
years)inthesameorderastherespectivefile
names or type “spm_load” to upload a *.txt
file with the covariates in the same order as
thevolumes
Interactions!None
Centering!Nocentering
"
ThresholdMasking!Absolute![specifyvalue(e.g.,“0.1”)]
"
ImplicitMask!Yes
"
ExplicitMask!<None>
o
GlobalCalculation!Omit
o
GlobalNormalization
" Overallgrandmeanscaling!No
o
Normalization!None
You could specify one or many covariates (i.e.,
partial out the variance of specific factors when
lookingatgroupdifferences).
15
FULLFACTORIALMODEL(FORA2X2ANOVA)
CAT12→StatisticalAnalysis→BasicModels
Parameters:
o
o
Directory<-X!SelectFiles![selecttheworkingdirectoryforyouranalysis]!Done
Design!“FullFactorial”
"
Factors!“New:Factor;New:Factor”
Factor
• Name![specifytext(e.g.,”sex”)]
• Levels!2
• Independence!Yes
• Variance!EqualorUnequal
• Grandmeanscaling!No
• ANCOVA!No
Factor
• Name![specifytext(e.g.,“handedness”)]
• Levels!2
• Independence!Yes
• Variance!EqualorUnequal
• Grandmeanscaling!No
• ANCOVA!No
"
SpecifyCells!“New:Cell;New:Cell;New:Cell;New:Cell”
Cell
• Levels![specifytext(e.g.,“11”)]
• Scans![selectfiles(e.g.,thesmoothedGMvolumesoftheleft-handedmales)]
Cell
• Levels![specifytext(e.g.,“12”)]
• Scans![selectfiles(e.g.,thesmoothedGMvolumesoftheright-handedmales)]
Cell
• Levels![specifytext(e.g.,“21”)]
• Scans![selectfiles(e.g.,thesmoothedGMvolumesoftheleft-handedfemales)]
Cell
• Levels![specifytext(e.g.,“22”)]
• Scans![selectfilese.g.,thesmoothedGMvolumesoftheright-handedfemales)]
o
Covariates*(seetextboxinexamplefortwo-sampleT-test)
o
Masking
"
ThresholdMasking!Absolute![specifyvalue(e.g.,“0.1”)]
"
ImplicitMask!Yes
"
ExplicitMask!<None>
o
GlobalCalculation!Omit
o
GlobalNormalization
" Overallgrandmeanscaling!No
o
Normalization!None
16
MULTIPLEREGRESSION(LINEAR)
CAT12→StatisticalAnalysis→BasicModels
Parameters:
o
Directory<-X!SelectFiles![selectthedirectoryforyouranalysis]!Done
o
Design!“MultipleRegression”
"
Scans![selectfiles(e.g.,thesmoothedGMvolumesofallsubjects)]!Done
"
Covariates!“New:Covariate”
Covariate
• Vector ! [enter the values in the same order as the respective file names of the
smoothedGMimages]
• Name![specifytest(e.g,“age”)]
• Centering!Overallmean
"
o
Covariates*(seetextboxinexamplefortwo-sampleT-test)
o
Masking
"
ThresholdMasking!Absolute![specifyvalue(e.g.,“0.1”)]
"
ImplicitMask!Yes
"
ExplicitMask!<None>
o
GlobalCalculation!Omit
o
GlobalNormalization
"
o
Intercept!IncludeIntercept
Overallgrandmeanscaling!No
Normalization!Non
17
MULTIPLEREGRESSION(POLYNOMIAL)
Inordertouseapolynomialmodelyouhavetoestimatethepolynomialfunctionofyourparameterpriorto
theanalysis.Usethefunctioncat_stat_polynomial(providedwithCAT12>r1140)forthatpurpose:
y=cat_stat_polynomial(x,order)
where“x”isyourparameterand“order”isthepolynomialorder(e.g.2forquadratic).
Exampleforpolynomialorder2(quadratic):
CAT12→StatisticalAnalysis→BasicModels
Parameters:
o
Directory<-X!SelectFiles![selectthedirectoryforyouranalysis]!Done
o
Design!“MultipleRegression”
o
o
"
Scans![selectfiles(e.g.,thesmoothedGMvolumesofallsubjects)]!Done
"
Covariates!“New:Covariate”
Covariate
• Vector![specifylinearterm(e.g.“y(:,1)”)]
• Name![specifytest(e.g,“agelinear”)]
• Centering!Overallmean
"
Covariates!“New:Covariate”
Covariate
• Vector![specifyquadraticterm(e.g.“y(:21)”)]
• Name![specifytest(e.g,“agequadratic”)]
• Centering!Overallmean
" Intercept!IncludeIntercept
Covariates*(seetextboxinexamplefortwo-sampleT-test)
Masking
"
ThresholdMasking!Absolute![specifyvalue(e.g.,“0.1”)]
"
ImplicitMask!Yes
"
ExplicitMask!<None>
o
GlobalCalculation!Omit
o
GlobalNormalization
"
o
Overallgrandmeanscaling!No
Normalization!None
18
FULLFACTORIALMODEL(INTERACTION)
CAT12→StatisticalAnalysis→BasicModels
Parameters:
o
o
Directory<-X!SelectFiles![selecttheworkingdirectoryforyouranalysis]!Done
Design!“FullFactorial”
"
Factors!“New:Factor”
Factor
• Name![specifytext(e.g.,”sex”)]
• Levels!2
• Independence!Yes
• Variance!EqualorUnequal
• Grandmeanscaling!No
• ANCOVA!No
o
SpecifyCells!“New:Cell;New:Cell”
Cell
• Levels![specifytext(e.g.,“1”)]
• Scans![selectfiles(e.g.,thesmoothedGMvolumesofthemales)]
Cell
• Levels![specifytext(e.g.,“2”)]
• Scans![selectfiles(e.g.,thesmoothedGMvolumesofthefemales)]
Covariates!“New:Covariate”
o
o
Covariate
• Vector ! [enter the values in the same order as the respective file names of the
smoothedGMimages]
• Name![specifytest(e.g,“age”)]
• Interactions!WithFactor1
• Centering!Nocentering
Covariates*(seetextboxinexamplefortwo-sampleT-test)
Masking
"
"
"
ThresholdMasking!Absolute![specifyvalue(e.g.,“0.1”)]
"
ImplicitMask!Yes
"
ExplicitMask!<None>
o
GlobalCalculation!Omit
o
GlobalNormalization
" Overallgrandmeanscaling!No
o
Normalization!None
19
FULLFACTORIALMODEL(POLYNOMIALINTERACTION)
Inordertouseapolynomialmodelyouhavetoestimatethepolynomialfunctionofyourparameterpriorto
theanalysis.Usethefunctioncat_stat_polynomial(providedwithCAT12>r1140)forthatpurpose:
y=cat_stat_polynomial(x,order)
where“x”isyourparameterand“order”isthepolynomialorder(e.g.2forquadratic).
Exampleforpolynomialorder2(quadratic):
CAT12→StatisticalAnalysis→BasicModels
Parameters:
o
o
Directory<-X!SelectFiles![selecttheworkingdirectoryforyouranalysis]!Done
Design!“FullFactorial”
"
Factors!“New:Factor”
Factor
• Name![specifytext(e.g.,”sex”)]
• Levels!2
• Independence!Yes
• Variance!EqualorUnequal
• Grandmeanscaling!No
• ANCOVA!No
o
SpecifyCells!“New:Cell;New:Cell”
Cell
• Levels![specifytext(e.g.,“1”)]
• Scans![selectfiles(e.g.,thesmoothedGMvolumesofthemales)]
Cell
• Levels![specifytext(e.g.,“2”)]
• Scans![selectfiles(e.g.,thesmoothedGMvolumesofthefemales)]
Covariates!“New:Covariate”
o
Covariate
• Vector![specifylinearterm(e.g.“y(:,1)”)]
• Name![specifytest(e.g,“agelinear”)]
• Interactions!WithFactor1
• Centering!Overallmean
Covariates!“New:Covariate”
o
o
Covariate
• Vector![specifyquadraticterm(e.g.“y(:,2)”)]
• Name![specifytest(e.g,“agequadratic”)]
• Interactions!WithFactor1
• Centering!Overallmean
Covariates*(seetextboxinexamplefortwo-sampleT-test)
Masking
"
"
"
"
ThresholdMasking!Absolute![specifyvalue(e.g.,“0.1”)]
"
ImplicitMask!Yes
"
ExplicitMask!<None>
20
o
GlobalCalculation!Omit
o
GlobalNormalization
" Overallgrandmeanscaling!No
o
Normalization!None
ESTIMATINGTHESTATISTICALMODEL
SPMmenu!Estimate
Parameters:
o SelectSPM.mat<-X!SelectFiles![selecttheSPM.matwhichyoujustbuilt]!Done
o Method!“Classical”
CHECKINGFORDESIGNORTHOGONALITY
IfyouhavemodeledTIVasconfoundingparameteritisnecessarytocheckthatTIVwillbeorthogonal
(inotherwordsindependent)toanyotherparameterofinterestinyouranalysis(e.g.parametersyou
are testing for). That means that TIV should not correlate with any other parameter of interest,
otherwisenotonlythevarianceexplainedbyTIVwillberemovedfromyourdata,butalsopartsofthe
varianceofyourparameterofinterest.
Please keep in mind to use “Overall mean” as “Centering” for the TIV covariate. Otherwise, the
orthogonality check sometimes even indicates a meaningful orthogonality only due to scaling
issues.
InordertocheckfordesignorthogonalityyoucanusetheReviewfunctionintheSPMGUI:
SPMmenu!Review
o SelectSPM.mat<-X!SelectFiles![selecttheSPM.matwhichyoujustbuilt]!Done
o Design!Designorthogonality
21
Figure4:Grayboxesbetweentheparameterspointtoacorrelation:thedarkertheboxthelargerthe
correlation(whichalsoholdsforinversecorrelations).Ifyouclickintheboxthecolinearitybetween
theparameterswillbedisplayed.
In the case of a considerable correlation an alternative approach is to use global scaling with TIV.
Applythefollowingsettingsforthisapproach:
o GlobalCalculation!User!GlobalValues<-X!DefineheretheTIVvalues
o GlobalNormalisation!Overallgrandmeanscaling!Yes!Grandmeanscaledvalue
!DefineherethemeanTIVofyoursampleor(orasapproximationavalueof1500
whichmightfitforthemajorityofdatafromadults)
o Normalization!Proportional
22
Please note that the global normalization will also affect the absolute threshold for the masking
because your images will be now scaled to the “Grand mean scaled value” from the Global
Normalisationoption.IfyouhavedefinedherethemeanTIVofyoursample(orasapproximationa
value of 1500) no change of the absolute threshold is needed. Otherwise, you have to correct the
absolute threshold because your values are now globally scaled to the “Grand mean scaled value”
fromtheGlobalNormalisationoption.
23
DEFININGCONTRASTS
SPM menu ! Results ! [select the SPM.mat file] ! Done (this opens the Contrast Manager) !
Definenewcontrast(i.e.,choose“t-contrast”or“F-contrast”;typethecontrastnameandspecifythe
contrastbytypingtherespectivenumbers,asshownbelow):
Please not that all zeros at the end of the contrast don’t have to be defined, but are kept
sometimesfordidacticpurposes.
Contrasts:
Two-sampleT-test
T-test
1-1
-11
• ForGroupA>GroupB
• ForGroupA<GroupB: F-test
1-1
• ForanydifferencesbetweenGroupAandB:
2x2ANOVA
T-test
•
•
•
•
•
•
•
•
Forleft-handedmales>right-handedmales:
Forleft-handedfemales>right-handedfemales:
Forleft-handedmales>left-handedfemales: Forright-handedmales>right-handedfemales:
Formales>females(maineffectsex): Forleft-handers>right-handers(maineffecthandedness):
Forleft-handers>right-handers&males>females:
Forleft-handers>right-handers&males<females:
24
1-100
001-1
10-10
010-1
11-1-1
1-11-1
1-1-11
-111-1
F-test
•
•
•
•
•
•
•
Formaineffecthandednessinmales:
Formaineffecthandednessinfemales:
Formaineffectsexinleft-handers:
Formaineffectsexinright-handers:
Formaineffectsex:
Formaineffecthandedness:
Forinteractionsexbyhandedness:
1-100
001-1
10-10
010-1
11-1-1
1-11-1
1-1-11
MultipleRegression(Linear)
T-test
• Forpositivecorrelation:
• Fornegativecorrelation:
001
00-1
F-test
001
• Anycorrelation:
The two leading zeros in the contrast are indicating the constant (sample effect, 1st column in the
designmatrix)andTIV(2ndcolumninthedesignmatrix).Incasethatnoadditionalcovariatesuchas
TIVisdefinedyouhavetoskiponeoftheleadingzeros(e.g.“01”).
MultipleRegression(Polynomial)
T-test
•
•
•
•
Forpositivelineareffect:
Forpositivequadraticeffect:
Fornegativelineareffect:
Fornegativequadraticeffect:
0010
0001
00-10
000-1
F-test
• Foranylineareffect: • Foranyquadraticeffect:
• Foranylinearorquadraticeffect:
25
0010
0001
0010
0001
The two leading zeros in the contrast are indicating the constant (sample effect, 1st column in the
designmatrix)andTIV(2ndcolumninthedesignmatrix).Incasethatnoadditionalcovariatesuchas
TIVisdefinedyouhavetoskiponeoftheleadingzeros(e.g.“01”).
Interaction(Linear)
T-test
• ForregressionslopeGroupA>GroupB:
• ForregressionslopeGroupA<GroupB:
001-10
00-110
F-test
001-10
• Foranydifferenceinregressionslope:
Thetwoleadingzerosinthecontrastareindicatingthemaineffect“group”.
Interaction(Polynomial)
T-test
•
•
•
•
ForlinearregressionslopeGroupA>GroupB:
ForlinearregressionslopeGroupA<GroupB:
ForquadraticregressionGroupA>GroupB: ForquadraticregressionGroupA<GroupB: 001-100
00-1100
00001-1
0000-11
F-test
• Foranydifferenceinlinearregressionslope: • Foranydifferenceinquadraticregression: • Foranydifferenceinregression:
Thetwoleadingzerosinthecontrastareindicatingthemaineffect“group”.
26
001-100
00001-1
001-100
00001-1
F-contrasts(effectsofinterest):
IfyouwouldliketousetheoldSPM2F-contrast“Effectsofinterest”therespectivecontrastvectoris:
eye(n)-1/n
wherenisthenumberofcolumnsofinterest.
For interaction and regression effects you have to add leading zeros for the constant term or the
sampleeffects:
[zeros(n,m)eye(n)-1/n]
wheremisthenumberofcolumnsofnointerest.
This F-contrast is helpful for checking whether there are any effects in your model (covariates of
interest)andisrequiredforplottingparameterestimatesofeffectsofinterest.
GettingResults:
SPMmenu!Results![selectacontrastfromContrastManager]!Done
• Maskwithothercontrasts!No
• Titleforcomparison:[usethepre-definednamefromtheContrastManagerorchangeit]
• Pvalueadjustmentto:
o None(uncorrectedformultiplecomparisons),setthresholdto0.001
o FDR(falsediscoveryrate),setthresholdto0.05,etc.
o FWE(family-wiseerror),setthresholdto0.05,etc.
• Extentthreshold:(eitheruse“none”orspecifythenumberofvoxels2)
2
Inordertoempiricallydeterminetheextentthreshold(ratherthansaying100voxelsor500voxels,whichiscompletely
arbitrary),simplyrunthisfirstwithoutspecifyinganextentthreshold.Thiswillgiveyouanoutput(i.e.,thestandardSPM
glassbrainwithsignificanteffects).Whenyouclick“Table”(SPMmainmenu)youwillgetatablewithallrelevantvalues
(MNI coordinates, p-values, cluster size etc). Below the table you will find additional information, such as “Expected
NumberofVoxelsperCluster”.Rememberthisnumber(thisisyourempiricallydeterminedextentthreshold).Re-runSPM
!Resultsetc.andspecifythisnumberwhenaskedforthe“ExtentThreshold”.Thereisalsoahiddenoptionin“CAT12!
Datapresentation!ThresholdandtransformspmT-maps”todefinetheextentthresholdintermsofap-valueortouse
the“ExpectedNumberofVoxelsperCluster”.
27
SpecialCases
CAT12forlongitudinaldata
BACKGROUND
ThemajorityofVBMstudiesarebasedoncross-sectionaldata,whereoneimageisacquiredforeach
subject.However,inordertotracke.g.learningeffectsovertimelongitudinaldesignsarenecessary,
where additional time-points are acquired for each subject. The analysis of these longitudinal data
requiresacustomizedprocessing,thatconsidersthecharacteristicsofintra-subjectanalysis.Whilefor
cross-sectionaldataimagescanbeprocessedindependentlyforeachsubjectlongitudinaldatahasto
be registered to the baseline image (or mean image) for each subject. Furthermore, spatial
normalizationisestimatedforthebaselineimageonlyandappliedtoallimages(Figure5).Additional
attention is then needed for the setup of the statistical model. The following section will therefore
describedatapreprocessingandmodelsetupforlongitudinaldata.
Textandfigureinpreparation
Fig4.:FlowdiagramforprocessinglongitudinaldatawithCAT12.Thisfiguredemonstrates the steps for processing longitudinal data. After an initial realignment, the mean of the realigned images is calculated (mean) and used as reference image in a subsequent realignment.Therealignedimages(rix)arethencorrectedforsignalinhomogeneitieswith
regard to the reference mean image. Spatial normalization parameters are estimated in
thenextstepusingthesegmentationsofthemeanimage.Thesenormalizationparameters are applied to the segmentations of the bias-corrected images (p1mrix). The resulting normalizedsegmentations(wp1mrix)arefinallyagainrealigned.
Preprocessingoflongitudinaldata-overview
The CAT12 Toolbox supplies a batch for longitudinal study design. Here, for each subject the
respectiveimagesneedtobeselected.Intra-subjectrealignment,biascorrection,segmentation,and
normalizationarecalculatedautomatically.Preprocessedimagesarewrittenaswp1mr*andwp2mr*
for grey and white matter respectively. To define the segmentation and normalization parameters,
thedefaultsincat_defaults.mareused.
28
CHANGESETTINGSFORPREPROCESSING
Youcanchangethetissueprobabilitymap(TPM)usingtheGUIorbychangingtheentryinthefile
cat_defaults.m. Any parameters that cannot be changed using the GUI have to be set in the file
cat_defaults.m:
Changeyourworkingdirectoryto“/toolbox/CAT12”inyourSPMdirectory:
!select“Utilities!cd”intheSPMmenuandchangetotherespectivefolder.
Thentype“opencat_defaults.m”inyourmatlabcommandwindow.Thefilewillopenintheeditor.If
youareunsurehowtochangethevalues,openthemodule“SegmentData”inthebatcheditorfor
reference.
PREPROCESSINGOFLONGITUDINALDATA
CAT12!Segmentlongitudinaldata
Parameters:
o Data<-X!New:Subject!Subject!Longitudinaldataforonesubject!SelectFiles
![selectrawdata]!Done
- Selectallvolumesforeachsubject.AstheToolboxdoesnotsupportmultispectral
data yet (i.e., different imaging methods for the same brain, such as T1-, T2-,
diffusion-weighted or CT images), it is recommended to choose a T1-weighted
image.
- Select“New:Subject”toadddataforanewsubject
Thedataforeachsubjectshouldbelistedasone“subject”intheBatchEditor,i.e.
thereareasmanysubjectslistedasincludedintheanalysis.
!Forallotheroptionsyoucanfollowtheinstructionsforpreprocessingofcross-sectional
data as described before. Please note that not all writing options are available for
longitudinaldata.
Forthenamingconventionsofallwrittenfilessee“Namingconventionofoutputfiles”.ThefinalGM
segmentsaremwp1r*,thefinalWMsegmentsarenamedmwp2r*ifyouhaveselectedtooptionto
modulatethedata.Withoutmodulationtheleading“m”isomitted.
29
Statisticalanalysisoflongitudinaldata-overview
Themaininterestinlongitudinalstudiesisthecommonchangeoftissuevolumeovertimeinagroup
of subjects or the difference in these changes between two or more groups. The setup of the
statisticalmodelneededtoassessthesequestionswillbedescribedontwoexamples.First,thecase
of only one group of 4 subjects with 2 time points each (e.g. normal aging) is presented.
Subsequently, the case of two groups of subjects with 4 time points per subject will be described.
These examples should cover most analyses – the number of time points / groups just have to be
adapted. In contrast to the analysis of cross-sectional data as described before we have to use the
flexiblefactorialmodelthatconsidersthatthetimepointsforeachsubjectaredependentdata.
STATISTICALANALYSISOFLONGITUDINALDATAINONEGROUP
CAT12→StatisticalAnalysis→BasicModels
Parameters:
o Directory<-X!SelectFiles![selecttheworkingdirectoryforyouranalysis]!Done
o Design!“FlexibleFactorial”
*
SPM is internally handling some
" Factors!“New:Factor;New:Factor”
keyword factors such as “subject” or
Factor
“repl”. If you use “subject” as
• Name![specifytext(e.g.,”subject”)*]
keyword for the first factor the
conditions can be easier defined by
• Independence!Yes
onlylabelingthetimepointsasinput
• Variance!EqualorUnequal
(seebelow).
• Grandmeanscaling!No
• ANCOVA!No
Factor
• Name![specifytext(e.g.,“time”)]
• Independence!No
• Variance!EqualorUnequal
• Grandmeanscaling!No
• ANCOVA!No
" Specify Subjects or all Scans & Factors ! “Subjects” ! “New: Subject; New:
Subject;New:Subject;New:Subject;”
Subject
• Scans![selectfiles(thesmoothedGMvolumesofthefirstSubject)]
• Conditions!“12”[fortwotimepoints]
Subject
- Scans![selectfiles(thesmoothedGMvolumesofthesecondSubject)]
30
"
- Conditions!“12”[fortwotimepoints]
Subject
• Scans![selectfiles(thesmoothedGMvolumesofthethirdSubject)]
• Conditions!“12”[fortwotimepoints]
Subject
- Scans![selectfiles(thesmoothedGMvolumesofthefourthSubject)]
- Conditions!“12”[fortwotimepoints]
Maineffects&Interaction!“New:Maineffect”
Maineffect
• Factornumber!2
Maineffect
• Factornumber!1
o Covariates*(seetextboxinexamplefortwo-sampleT-test)
o Masking
" ThresholdMasking!Absolute![specifyvalue(e.g.,“0.1”)]
" ImplicitMask!Yes
" ExplicitMask!<None>
o GlobalCalculation!Omit
o GlobalNormalization
" Overallgrandmeanscaling!No
o Normalization!None
STATISTICALANALYSISOFLONGITUDINALDATAINTWOGROUPS
CAT12→StatisticalAnalysis→BasicModels
Parameters:
o Directory<-X!SelectFiles![selecttheworkingdirectoryforyouranalysis]!Done
o Design!“FlexibleFactorial”
*
SPM is internally handling some
" Factors ! “New: Factor; New: Factor; New: keyword factors such as “subject” or
Factor”
“repl”. If you use “subject” as
keyword for the first factor the
Factor
conditions can be easier defined by
• Name![specifytext(e.g.,”subject”)*]
onlylabelingthetimepointsasinput
(seebelow).
31
"
• Independence!Yes
• Variance!Equal
• Grandmeanscaling!No
• ANCOVA!No
Factor
• Name![specifytext(e.g.,”group”)]
• Independence!Yes
• Variance!Unequal
• Grandmeanscaling!No
• ANCOVA!No
Factor
• Name![specifytext(e.g.,“time”)]
• Independence!No
• Variance!Equal
• Grandmeanscaling!No
• ANCOVA!No
Specify Subjects or all Scans & Factors ! “Subjects” ! “New: Subject; New:
Subject;New:Subject;New:Subject;”
Subject
• Scans ! [select files (the smoothed GM volumes of the 1st Subject of first
group)]
• Conditions!“[1111;1234]’“[firstgroupwithfourtimepoints]
Donotforgettheadditionalsinglequote!Otherwiseyou
havetodefinetheconditionsas“[11;12;13;14]“
Subject
• Scans![selectfiles(thesmoothedGMvolumesofthe2ndSubjectoffirst
group)]
• Conditions!“[1111;1234]’“[firstgroupwithfourtimepoints]
Subject
• Scans![selectfiles(thesmoothedGMvolumesofthe3rdSubjectoffirst
group)]
• Conditions!“[1111;1234]’“[firstgroupwithfourtimepoints]
Subject
• Scans ! [select files (the smoothed GM volumes of the 4th Subject of
secondgroup)]
• Conditions!“[2222;1234]’“[secondgroupwithfourtimepoints]
Subject
• Scans![selectfiles(thesmoothedGMvolumesofthe1stSubjectofsecond
group)]
32
"
• Conditions!“[2222;1234]’“[secondgroupwithfourtimepoints]
Subject
• Scans ! [select files (the smoothed GM volumes of the 2nd Subject of
secondgroup)]
• Conditions!“[2222;1234]’“[secondgroupwithfourtimepoints]
Subject
• Scans ! [select files (the smoothed GM volumes of the 3rd Subject of
secondgroup)
• Conditions!“[2222;1234]’”[secondgroupwithfourtimepoints]
Subject
• Scans ! [select files (the smoothed GM volumes of the 4th Subject of
secondgroup)]
• Conditions!“[2222;1234]’”[secondgroupwithfourtimepoints]
Maineffects&Interaction!“New:Interaction;New:Maineffect”
Interaction
• Factornumbers!23[Interactionbetweengroupandtime]
Maineffect
• Factornumber!1
o Covariates*(seetextboxinexamplefortwo-sampleT-test)
o Masking
" ThresholdMasking!Absolute![specifyvalue(e.g.,“0.1”)]
" ImplicitMask!Yes
" ExplicitMask!<None>
o GlobalCalculation!Omit
o GlobalNormalization
" Overallgrandmeanscaling!No
o Normalization!None
33
Contrasts
Longitudinal data in one group (example for
twotimepoints)
T-test
•
•
1-1
ForTime1>Time2:
-11
ForTime1<Time2: F-test
•
1-1
ForanydifferencesinTime:
Longitudinaldataintwogroups(examplefor
twotimepoints)
T-test
•
•
•
•
1-100
ForTime1>Time2inGroupA:
00-11
ForTime1>Time2inbothGroups:
1-11-1
ForTime1>Time2&GroupA>GroupB:
1-1-11
ForTime1>Time2inGroupB:
F-test
•
•
•
•
•
ForanydifferencesinTimeinGroupA:
1-100
ForanydifferencesinTimeinGroupB:
001-1
ForMaineffectTime:
1-11-1
ForMaineffectGroup:
ones(1,n)/n-ones(1,n)/n0ones(1,n1)/n1-ones(1,n2)/n2
ForInteractionTimebyGroup:
1-1-11
Heren1andn2arethenumberofsubjectsinGroupAandBrespectivelyandnisthenumberoftime
points.Pleasenotethatthezerointhe5thcolumnforthemaineffectGroupindicatesthecovariateof
nointerest(e.g.TIV).
34
AlteredWorkflowsforVBM-analyses
Background
For most analyses the VBM Toolbox will supply all tools needed. That is, as the new segmentation
algorithmisnotdependentonTissueProbabilityMaps(TPMs)anymoreandaspredefinedDARTELtemplates for healthy adults exist, most questions can be assessed using the toolbox settings.
However, for some special cases such as analyses in children or special patient populations, the
toolboxsettingsmightnotbeoptimal.ForthesecasestheCAT12Toolboxprovidesanintegrationinto
the SPM12 environment that can be used to optimize the preprocessing. In the following, we will
presentstrategieshowtodealwiththesespecialcases.
StandardVBMpreprocessing:Input,Outputandwheretomodify
ThefirstmoduleoftheCAT12Toolbox(“SegmentData”)processesallpreprocessingstepsexceptfor
thesmoothing.Basically,ittakesstructuralvolumesandTPMsasinput.Itwillthensegmentthedata,
applyaregistrationtoMNISpace(eitherrigidoraffine)andsubsequentlyanon-lineardeformation.
The non-linear deformation parameters can be calculated via the low dimensional SPM default
approachorthehighdimensionalDARTELalgorithmandthepredefinedtemplates.Figure5depicts
thispreprocessingworkflowandhighlightspossibilitieswheretomodify.
Figureinpreparation
Fig. 5: Flow-chart of the preprocessing steps within the module “Segment Data”.
Marked in red are those steps, where the preprocessing can be customized. Per
default,thebuilt-inDARTELnormalizationworkswiththeCAT12DARTELtemplates
of550healthyadultcontrolsubjects.Affineregisteredtissuesegmentscanbeused
to create customized DARTEL-templates, which can then be used to replace the
defaultDARTELtemplate.
35
Adaptingtheworkflows
Customizedtissueprobabilitymaps-overview
FordataonchildrenitwillbeagoodideatocreatecustomizedTPMs,whichreflectageandgenderof
the population. The TOM8 Toolbox (available via: https://irc.cchmc.org/software/tom.php) provides
the means to customize these TPMs. To learn more about the TOM toolbox, see also
http://dbm.neuro.uni-jena.de/software/tom/.
CUSTOMIZEDTISSUEPROBABILITYMAPS
AbouttheTOM8Toolbox:
!selectModule“createnewtemplate”
!select“TOM.mat”(youwillhavetodownloadthisfiletogetherwiththetoolbox)
!writepriors/templateassinglefile
!forallothersusedefaultsettingsormodify.For“Age”eitheravectororameanage
(whenusingtheaverageapproach)mustbespecified.
ImplementationintoCAT12:
CAT12→SegmentData
Parameters:
o OptionsforinitialSPM12affineregistration
" TissueProbabilityMap(!SelectyourcustomizedTPMshere)
CustomizedDARTEL-template-overview
Forallcasesthatincludeatleast50-100subjectsacustomizedDARTELtemplatecanbecreated.That
is,greymatterandwhitemattertissuesegmentsofallsubjectsareusedtocreateameantemplateof
the study sample. As the CAT12 toolbox writes all files needed to create these templates (“DARTEL
export”), this requires only two additional steps. In order to use these newly created DARTELTemplateswiththeCAT12Toolbox,anaffineregistrationoftheDARTELexportshouldbeused.From
these affine registered segments customized DARTEL templates can then be created and used with
theCAT12module“SegmentData”.
36
CUSTOMIZEDDARTEL-TEMPLATE
PleasenotethattheuseofanownDARTELtemplatewillresultindeviationsandunreliableresultsfor
anyROI-basedestimationsbecausetheatlaseswilldiffer.
SeveralstepsareneededtocreatenormalizedtissuesegmentswithcustomizedDARTELTemplates.
ThesestepscanbebatchedusingdependenciesintheBatchEditor.Thelaststepcanbererunusing
the customized templates, if additional output files are needed. In the first step the T1 images are
segmented, and the tissue segments normalized to the Tissue Probability Maps using an affine
transformation.Startwithselectingthemodule“SegmentData”.
CAT12!SegmentData
Parameters:
!foralloptionsexcept“writingoptions”usesettingslikefora“standard”VBManalysis.
o WritingOptions
•
•
•
•
“GreyMatter”!“Modulatednormalized”!“No”
“GreyMatter”!“DARTELexport”!“affine”
“WhiteMatter”!“Modulatednormalized”!“No”
“WhiteMatter”!“DARTELexport”!“affine”
These settings will produce the volumes “rp1*-affine.nii” and “rp2*-affine.nii”, which are the grey
(rp1)andwhite(rp2)mattersegmentsafteraffineregistration.Thefollowingmodulescanbechosen
directlyinthebatcheditor(SPM!Tools!DARTELTools!RunDARTEL(createTemplates),and
SPM!Tools!CAT12!CAT12:SegmentData).Itmakessensetoaddandspecifythesemodules
togetherwiththe“SegmentData”modulewithintheBatchEditorandtosetdependencies.
SPM!Tools!DartelTools!RunDARTEL(createTemplates)
Parameters:
Images!selecttwotimes“new:Images”
"
"
Images:!selectthe“rp1*-affine.nii”filesorcreateadependency.
Images:!selectthe“rp2*-affine.nii”filesorcreateadependency.
!allotheroptions:usedefaultsormodify
SPM!Tools!DartelTools!NormalisetoMNIspace
37
Parameters:
DartelTemplate!selectthefinalcreatedtemplatewiththeending“_6”.
o Selectaccordingto!“ManySubjects”
Flowfields:!selectthe“u_*.nii”filesorcreateadependency.
• Images: ! New: Images ! select the “rp1*-affine.nii” files or create a
dependency.
• Images: ! New: Images ! select the “rp2*-affine.nii” files or create a
dependency.
o Preserve:!“PreserveAmount”
"
o GaussianFWHM:!usedefaultsormodify
!allotheroptions:usedefaultsormodify
Please note, that a subsequent smoothing is not necessary before statistics if you have used the
option“NormalisetoMNIspace”withadefinedGaussianFWHM.
Re-useofcustomizedDARTEL-templates
OptionallyyoucanalsousethenewcreatedcustomizedDARTELtemplatesintheCAT12Toolbox.This
mightbehelpfulifnewdataareclose(intermsofage)tothedatathathavebeenusedforcreating
theDARTELtemplate.Then,itisnotmandatorytoalwayscreateanewtemplateandyoucansimply
use the previously created DARTEL template. In that case an additional registration to MNI (ICBM)
spacehastobeapplied:
SPM!Tools!DartelTools!RunDARTEL(PopulationtoICBMRegistration)
Parameters:
DartelTemplate!selectthefinalcreatedtemplatewiththeending“_6”.
SPM!Util!Deformations
Parameters:
Composition!“New:DeformationField”
" DeformationField:!selectthe“y_*2mni.nii”filefromstepabove.
o Output!“New:Pushforward”
"
Applyto!selectall“Template”fileswiththeending“_0”to“_6”.
38
"
"
"
Outputdestination!Outputdirectory!selectdirectoryforsavingfiles.
Field of View ! Image Defined ! select final “Template” file with the ending
“_6”.
Preserve!PreserveConcentrations(no“modulation”).
Finally,thenewtemplatecanbeusedasdefaultDARTELtemplateforanynewdatathatarecloseto
thedatathathavebeenusedfortemplatecreation:
CAT12!SegmentData
Parameters:
o Volumes<-X!SelecttheoriginalT1imageslikeinthefirstmodule“SegmentData”.
o Extended Options for CAT12 segmentation ! “Spatial normalization Template” !
SelectnormalizedDARTELTemplate“wTemplate*_1.nii”
- For all other options use the same settings as in the first module, or
modify.
o WritingOptions!selecttheoutputfilesjustlikeinanystandardVBManalysis:
Howtoproceed
All steps described above are just an adaption of the CAT12 Toolbox module “Segment Data”. As
always it is a good idea to save the applied modules and to perform quality control. Here, the
modules“Displayonesliceforallimages”and“Checksamplehomogeneity”fromtheCAT12Toolbox
willbehelpful.Pleasealsonotethattheuseof
Othervariantsofcomputationalmorphometry
Deformation-basedmorphometry(DBM)
Background
DBMisbasedontheapplicationofnon-linearregistrationprocedurestospatiallynormaliseonebrain
toanotherone.Thesimplestcaseofspatialnormalisationistocorrecttheorientationandsizeofthe
brains. In addition to these global changes, a non-linear normalisation is necessary to minimise the
remainingregionaldifferencesbymeansoflocaldeformations.Ifthislocaladaptationispossible,the
deformations now reveal information about the type and localization of the structural differences
betweenthebrainsandcanundergosubsequentanalysis.
Differences between both images are minimized and are now coded in the deformations. Finally, a
mapoflocalvolumechangescanbequantifiedbyamathematicalpropertyofthesedeformations–
the Jacobian determinant. This parameter is well known from continuum mechanics and is usually
39
usedfortheanalysisofvolumechangesinflowingliquidsorgases.TheJacobiandeterminantallowsa
directestimationofthepercentagechangeinvolumeineachvoxelandcanbestatisticallyanalyzed
(Gaseretal.2001).Thisapproachisalsoknownastensor-basedmorphometrybecausetheJacobian
determinantrepresentssuchatensor.
Adeformation-basedanalysiscanbecarriedoutnotonlyonthelocalchangesinvolumebutalsoon
theentireinformationofthedeformations,whichalsoincludesthedirectionandstrengthofthelocal
deformations (Gaser et al. 1999; 2016). Since each voxel contains three-dimensional information, a
multivariatestatisticaltestisnecessaryforanalysis.AmultivariategenerallinearmodelorHotelling’s
T2testiscommonlyusedforthistypeofanalysis(Gaseretal.1999;Thompsonetal.1997).
AdditionalStepsinCAT12
CAT12!SegmentData
Parameters:
o Writingoptions!Jacobiandeterminant!Normalized!Yes
-
Inordertosavetheestimatedvolumechangeschangethewritingoptionforthe
normalizedJacobiandeterminantto“yes”.
Changesinstatisticalanalysis
Follow the steps for the statistical analysis as described for VBM, select the smoothed Jacobian
determinants(e.g.swj_*.nii)andchangethefollowingparameters:
CAT12→StatisticalAnalysis→BasicModels
Parameters:
o Covariates:Don’tuseTIVascovariate
o Masking
" ThresholdMasking!None
" ImplicitMask!Yes
" ExplicitMask!../spm12/tpm/mask_ICV.nii
Surface-basedmorphometry(SBM)
Background
Surface-basedmorphometryhasseveraladvantagesoverusingvolumetricdataalone.Forinstance,
brainsurfacemesheshavebeenshowntoincreasetheaccuracyofbrainregistrationcomparedwith
Talairach registration (Desai et al. 2005). Brain surface meshes also permit new forms of analyses,
such as gyrification indices that measure surface complexity in 3D (Yotter et al. 2011b) or cortical
thickness(Gaseretal.2016).Furthermore,inflationorsphericalmappingofthecorticalsurfacemesh
40
raisestheburiedsulcitothesurfacesothatmappedfunctionalactivityintheseregionscanbeeasily
visualized.
Localadaptivesegmentation
Gray matter regions with high iron concentration, like the motor cortex and the occipital regions,
often have increased intensities that lead to misclassications. In addition to our adaptive MAP
approach for partial volume segmentation we use an approach that allows adaptation of local
intensitychangesinordertodealwithvaryingtissuecontrast(Dahnkeetal.2012a).
Corticalthicknessandcentralsurfaceestimation
We use a fully automated method that allows for measurement of cortical thickness and
reconstructionsofthecentralsurfaceinonestep.Itusesatissuesegmentationtoestimatethewhite
matter (WM) distance, then projects the local maxima (which is equal to the cortical thickness) to
other gray matter voxels by using a neighbor relationship described by the WM distance. This
projection-based thickness (PBT) allows the handling of partial volume information, sulcal blurring,
andsulcalasymmetrieswithoutexplicitsulcusreconstruction(Dahnkeetal.2012b).
Topologicalcorrection
In order to repair topological defects we use a novel method that relies on spherical harmonics
(Yotteretal.2011a).First,theoriginalMRIintensityvaluesareusedasabasistoselecteithera“fill”
or“cut”operationforeachtopologicaldefect.Wemodifythesphericalmapoftheuncorrectedbrain
surfacemesh,suchthatcertaintrianglesarefavoredwhilesearchingfortheboundingtriangleduring
reparameterization.Then,alow-passfilteredalternativereconstructionbasedonsphericalharmonics
ispatchedintothereconstructedsurfaceinareasthatpreviouslycontaineddefects.
Sphericalmapping
Asphericalmapofacorticalsurfaceisusuallynecessarytoreparameterizethesurfacemeshintoa
common coordinate system to allow inter-subject analysis. We use a fast algorithm to reduce area
distortion resulting in an improved reparameterization of the cortical surface mesh (Yotter et al.
2011c).
Sphericalregistration
Wehaveadaptedthevolume-baseddiffeomorphicDartelalgorithmtothesurface(Ashburner,2007)
to work with spherical maps (Yotter et al. 2011d). We apply a multi-grid approach that uses
reparameterizedvaluesofsulcaldepthandshapeindexdefinedonthespheretoestimateaflowfield
thatallowsdeformingasphericalgrid.
AdditionalStepsinCAT12
CAT12!SegmentData
Parameters:
o Writingoptions!Surfaceandthicknessestimation!Yes
41
Use projection-based thickness to estimate cortical thickness and to create the
centralcorticalsurfacefortheleftandrighthemisphere.
CAT12!SurfaceTools!Resample&SmoothSurfaces
Parameters:
o SurfaceData<-X!Selectthesurfacedata(e.g.[lr]h.thickness.*)
o SmoothingFilterSizeinFWHM[usedefaultsormodify]
- 12-18mmkernelsarewidelyusedforSBMandIrecommendtostartwitha
valueof15mm.
o Splitjobintoseparateprocesses
-
-
Inordertousemulti-threadingtheCAT12segmentationjobwithmultiple
subjects can be split into separate processes that run in the background.
YoucanevencloseMatlab,whichwillnotaffecttheprocessesthatwillrun
inthebackgroundwithoutGUI.Ifyoudonnotwanttorunprocessesinthe
backgroundthensetthisvalueto0.
Keep in mind that each process needs about 1.5..2GB of RAM, which
shouldbeconsideredtochoosetherightnumberofprocesses.
Please further note that no additional modules in the batch can be run
except CAT12 segmentation. Any dependencies will be broken for
subsequentmodules.
Extractoptionalsurfaceparameters
Youcanalsoextractadditionalsurfaceparametersthathavetoberesampledandsmoothedwiththe
beforementionedtool.
CAT12!SurfaceTools!Extract&MapSurfaceData!ExtractAdditionalSurfaceParameters
Parameters:
o CentralSurfaces<-X!Selectthecentralsurfacedata(e.g.[lr]h.central.*)
o Gyrificationindex
- Extract gyrification index (GI) based on absolute mean curvature. The
methodisdescribedinLudersetal.NeuroImage,29:1224-1230,2006.
o Corticalcomplexity(fractaldimension)
ExtractCorticalcomplexity(fractaldimension)whichisdescribedinYotteretal.Neuroimage,56(3):
961-973,2011.
Warning:Estimationofcorticalcomplexityisveryslow!
o Sulcusdepth
42
Extractsqrt-transformedsulcusdepthbasedontheeuclideandistancebetweenthecentralsurface
anditsconvexhull.
Transformationwithsqrt-functionisusedtorenderthedatamorenormallydistributed.
Changesinstatisticalanalysis
For statistical analysis of surface measures you can use the common 2nd level models that are also
usedforVBM.However,youhavetoanalyzeeachhemisphere(indicatedby“lh”and“rh”)separately.
Follow the steps for the statistical analysis as described for VBM and change the following
parameters:
CAT12→StatisticalAnalysis→BasicModels
Parameters:
o Covariates:Don’tuseTIVascovariate
o Masking
" ThresholdMasking!None
" ImplicitMask!Yes
" ExplicitMask!<none>
Asinputyouhavetoselecttheresampledandsmoothedfiles(seeabove).Athicknessfileoftheleft
hemispherethatwasresampledandsmoothedwiththedefaultFWHMof15mmisnamedas:
S15mm.lh.thickness.resampled.name.gii
where“name”iftheoriginalfilenamewithoutextension.Pleasenotethatthesefilesaresavedinthe
surf-folderasdefault.
Do not use the “Estimate” function in the SPM window, but rather use the respective function in
CAT12.ThisallowstooverlayyourresultsautomaticallyontotheFreesurferaveragesurface:
CAT12!StatisticalAnalysis!EstimateSurfaceModels
Regionofinterest(ROI)analysis
CAT12 allows estimation of tissue volumes (and additional surface parameters such as cortical
thickness)fordifferentvolumeandsurface-basedatlasmaps(Gaseretal.2016).Alloftheseresults
areestimatedinnativespacebeforeanyspatialnormalization.Theresultsforeachdatasetarestored
asXMLfilesinthelabeldirectory.TheXMLfilecatROI[s]_*.xmlcontainsinformationofallatlasesas
data structure for one dataset and the optional “s” indicates surface atlases. You can use the CAT
function“cat_io_xml”toreadtheXMLdataasstructure.
PleasenotethatinthebetaversionofCAT12theseXMLfileswerenotautomaticallysavedbecause
theoptionforROIanalysiswasdisabled.Iftheoption"cat12.output.ROI"incat_defaults.missetto
"0" no ROI XML files are saved and you have to set this parameter to "1" and preprocess the data
43
againifyouareinterestedinROIanalysisofvolumedata(forsurfacedatatheseXMLfilesarealways
extractedafterpreprocessing,thusitisnotnecessarytodothisforsurfacedata).
Additionalstepsforsurfacedata
WhileROI-basedvaluesforVBM(volume)dataareautomaticallysavedinthelabelfolderasXMLfile
it is necessary to additionally extract these values for surface data. This has to be done after
preprocessing the data and creating cortical surfaces. You can extract ROI-based values for cortical
thicknessbutalsoforanyothersurfaceparameterthatwasextractedusingthe“ExtractAdditional
SurfaceParameters”function:
CAT12!ExtractROIData!ExtractROI-basedsurfacevalues
Parameters:
o (Left) Surface Data Files <-X ! Select surface data files such as lh.thickness.* for all
subjects
StatisticalanalysisofROIdata
Finally, the XML files of several subjects can be analyzed using an already existing SPM design with
CAT12!AnalyzeROIs.Here,theSPM.matfileisusedinordertogetinformationaboutallrespective
label files, but also about your design (including all covariates/confounds you have modeled). Thus,
thesamestatisticalanalysisthatissavedintheSPM.matfileisappliedtoyourROIdata.Youcanthen
selectacontrast,athresholdandameasurestoanalyze(e.g.Vgm,Vwm,thickness,gyrification...)and
canchoosebetweendifferentatlasmaps.Theresultswillbeprintedandsavedasthresholdedlog-p
volumeorsurfacemap:
logPThreshold_NameOfContrast_NameOfAtlas_Measure.nii
[lr]h.logPThreshold_NameOfContrast_NameOfAtlas_Measure.gii
ThesemapscanbeoptionallyvisualizedusingCAT12 !DisplayResults !SliceOverlayforvolume
mapsorCAT12!DisplayResults!Displaysurfaceresultsforsurfacemaps.
In order to analyze different measures (e.g. Vgm/Vwm for volumes or thickness/gyrification for
surfaces)youcanuseanyexistingvolume-basedanalysistoextractdifferentvolumemeasuresorany
existingsurface-basedanalysistoextractdifferentsurfacemeasures.Togiveanexample:Anexisting
SPM.matfilewithaVBManalysisofGMallowsyoutoanalyzeROImeasuresforboth,GMaswellas
WM.Thus,itisnotnecessarytohaveaSPM.matfileofaVBManalysisofWM.Thesameholdsfor
surface-basedanalysis.IfyouhaveanexistingSPM.matfortheanalysisofcorticalthicknessyoucan
also estimate ROI analysis for gyrification or fractal dimension. However, it is necessary to extract
before ROI-based measures for each subject using CAT12 ! Extract ROI Data ! Extract ROI-based
surfacevalues.
ForROIanalysisofsurfacesyoucanselecttheSPM.matfileoftheanalysiseitherfortheleftorright
hemisphere. The design should be the same and the ROI results will be always estimated for both
hemispheres.
Pleasenote,thatifyouhavemovedyourdataafterestimatingyouroriginalvoxel-orsurfaced-based
statisticstheneededROIfilescannotbefound.
44
OptionalextractionofROIdata
Finally,theXMLfilesofseveralsubjectscanbecombinedandsavedforfurtheranalysisasCSVfile
usingthe“EstimateMeanValuesinsideROI”function:
CAT12!ExtractROIData!EstimatemeanvaluesinsideROI
Parameters:
o XMLfiles<-X!Selectxmlfilesforeachsubjectthataresavedinthelabelfolder.
o Outputfile!Defineoutputnameforcsvfile.Thisnameisextendedbytheatlasname
andthenameofthemeasure(e.g.“Vgm”forgraymattervolume)
Foreachmeasure(e.g.“Vgm”forgraymattervolume)andeachatlasaseparateCSVfileiswritten.
Thisworksforbothvolumeaswellassurfacedata,butvolumeandsurfacedatahastobeprocessed
separately using this function (surface-based ROI values are indicated by an additional “s”, e.g.
''catROIs_''). You can use external software such as Excel or SPSS to read the resulting CSV files for
furtheranalysis.Takealsocareofthedifferentinterpretationbetween“.”and“,”dependingonyour
regionandlanguagesettingsonyoursystem.
UseofatlasfunctionsinSPM12
Moreover,youcanusethevolume-basedatlasesthatareprovidedwithCATalsoasatlasmapswith
SPM atlas functions. This is especially helpful if you have used the default VBM processing pipeline
because the CAT12 atlas maps are then in the same DARTEL space as your data. Thus, if you have
used the default VBM processing pipeline it is recommended to use the CAT12 atlases in DARTEL
spaceratherthantheSPMNeuromorphometricsatlas.InordertouseCAT12atlasesyouhavetocall
the cat_install_atlases function once that will copy the atlases to SPM. By default only
Neuromorphometics,LPBA40,andHammersatlasmapsareusedandareindicatedinthenamebya
leading''dartel_''.
Atlas maps for surfaces can be used with the function CAT12 ! Display Results ! Display Surface
Results. Here, the data cursor function allows you to display atlas regions under the cursor.
Furthermore,youcanusethe''Atlaslabeling''functiontoprintalistofatlasregionsoftheresulting
clusters.
Additionalinformationonnative,normalizedandmodulatedvolumes
Whenpreprocessingtheimages(see“FirstModule:SegmentData”,onpages5-6),thedecisionabout
thenormalizationparameterswilldeterminetheinterpretationoftheanalysisoutcomes.Pleasenote
thatsomeoftheoutputoptionsareonlyavailableintheexpertmode.
“Nativespace”producestissueclassimagesinspatialcorrespondencetotheoriginaldata.Although
thiscouldbeusefulforestimatingglobaltissuevolumes(e.g.,GM+WM+CSF=TIV)itisnotsuitableto
conductVBManalysesduetothemissingvoxel-wisecorrespondenceacrossbrains).Ofnote,ifoneis
interested in these global tissue volumes in native space (“raw values”), it won’t be necessary to
45
actually output the tissue class images in native space. The “Segment Data”-function automatically
generatesanxmlfileforeachsubject(cat_*.xml),whichcontainstherawvaluesforGM,WM,and
CSF.Thesubject-specificvaluescanbecombined(i.e.,integratedintoasingletextfile)byusingthe
function:CAT12!StatisticalAnalysis!EstimateTIV.
“Normalized”producestissueclassimagesinspatialcorrespondencetothetemplate.Thisisuseful
forVBManalyses(e.g.,“concentration”ofgraymatter;Goodetal.2001;Neuroimage).
“Modulatednormalized”producestissueclassimagesinalignmentwiththetemplate,butmultiplies
(“modulates”)thevoxelvaluesbytheJacobiandeterminant(i.e.,linearandnon-linearcomponents)
derived from the spatial normalization. This is useful for VBM analyses and allows comparing the
absoluteamountoftissue(e.g.,“volume”ofgraymatter;Goodetal.2001;Neuroimage).Pleasenote
thatbyusingthistypeofmodulationyouhavetouseTIVascovariateinordertocorrectfordifferent
brainsizes.AnalternativeapproachistouseglobalscalingwithTIVifTIViscorrelatingtoomuchwith
your covariates of interest. See the section about “Checking for Design Orthogonality” for more
information.
If you use the expert mode you can optionally choose to modulate your data for non-linear terms
only,whichwasthedefaultforpreviousVBMversions.Thisproducestissueclassimagesinalignment
withthetemplate,butmultipliesthevoxelvaluesbythenon-linearcomponentsonly.Thisisuseful
forVBManalysesandallowscomparingtheabsoluteamountoftissuecorrectedforindividualbrain
sizes. Of note, this option is similar to using “Affine+non-linear” (see above) in combination with
“global normalization” (when later building the statistical model using the traditional PET designs).
Although this type of modulation was used in previous VBM versions as default it is not
recommended anymore because the use of the standard modulation in combination with TIV as
covariategivesmorereliableresults(Maloneetal.2015;Neuroimage).
46
Namingconventionofoutputfiles
Please note that the resulting files of CAT12 are organized in separate subfolders (e.g. mri, report,
surf,label).Ifyoudon'twanttousesubfoldersyoucanchangetheoption"cat12.extopts.subfolders"
incat_default.mto"0".
Images(savedinsubfolder"mri")
SegmentedImages:
#p[0123]*
[m[0]w]p[0123]*[_affine].nii
Bias,noiseandintensitycorrectedT1image: #m*
[w]m*.nii
Jacobiandeterminant
#j_
wj_*.nii
Deformationfield(inversefield)
y_(iy_)
y_*.nii(iy_*.nii)
*
filename
#
imagespaceprefix
Imagespaceprefix:
m
modulated
m0
modulatednon-linearonly(expertmodeonly)
w
warped(spatiallynormalizedusingDARTEL)
Imagespaceextension:
_affine
affineregisteredonly
Imagedataprefix:
p
partialvolume(PV)segmentation
0
PVlabel
1
GM
2
WM
3
CSF
47
Surfacesinnativespace(savedinsubfolder"surf")
SURF.TYPE.*.gii
SURF
left,righthemisphere[lh|rh]
TYPE
surfacedatafile[central|sphere|thickness|gyrification|...]
central-coordinatesandfacesofthecentralsurface
sphere-coordinatesandfacesofthesphericalprojectionofthecentralsurface
sphere.reg-coordinatesandfacesofthesphereaftersphericalregistration
thickness-thicknessvaluesofthesurface
sqrtsulc-sqrt-transformedvaluesofsulculdepthbasedontheeuclidiandistancebetween
thecentralsurfaceanditsconvexhull
gyrification-gyrificationvaluesbasedonabsolutemeancurvature
fractaldimension-fractaldimensionvalues(corticalcomplexity)
Surfaces in (normalized) template space (after resampling and smoothing; saved in subfolder
"surf")
FWHM.SURF.TYPE.resampled.*.gii
FWHM
filtersizeinFWHMaftersmoothing(e.g.s15mm)
SURF
left,righthemisphere[lh|rh]
TYPE
surfacedatafile[thickness|gyrification|fractaldimension|...]
thickness-thicknessvaluesofthesurface
sqrtsulc-sqrt-transformedvaluesofsulculdepthbasedontheeuclidiandistancebetween
thecentralsurfaceanditsconvexhull
gyrification-gyrificationvaluesbasedonabsolutemeancurvature
fractaldimension-fractaldimensionvalues(corticalcomplexity)
48
Imagesandsurfaceoflongitudinaldata
After processing longitudinal data the filenames additionally contain an "r" between the original
filenameandtheotherprefixestoindicatetheadditionalregistrationstep.Pleasealsonotethatonly
thebias,noiseandintensitycorrectedaverageimageofalltimepointsforeachsubjectissaved.
Reports(savedinsubfolder"report")
Global morphometric and image quality measures are stored in the cat_*.xml file. This file also
containsotherusefulinformationaboutsoftwareversionsandtheusedoptionsforpreprocessingthe
data.Youcanusethecat_io_xmlfunctiontoreaddatafromxml-files.Furthermore,areportforeach
datasetissavedaspdf-filecatreport_*.pdf.
Regionsofinterest(ROI)data(savedinsubfolder"label")
ROIdataisoptionallysavedasxml-filecatROI[s]_*.xml.Theoptional“s”indicatessurfaceatlases.
CallingCATfromtheUNIXcommandline
You can call CAT also from the Unix command line as shell script. This allows to run the process
completely in the background without any graphical output and distribute all given data to parallel
jobs.ThismightbehelpfulifyouwanttorunCATonacomputercluster.
Youcanusetheshellscriptscat_batch_cat.shorcat_batch_long.sh(forlongitudinaldata).Pleasecall
thescriptstoseemoreinformationhowtodefineforexampleoptionaldefaultsfiles.
Furthermore,thereexistsascriptdistribute_to_server.shtodistributejobstodifferentservers.
Technicalinformation
This toolbox is an extension of the segmentation in SPM12, but uses a completely different
segmentationapproach.3
AMAPSegmentation
ThesegmentationapproachisbasedonanAdaptiveMaximumAPosterior(AMAP)techniquewithout
theneedforaprioriinformationabouttissueprobabilities.Thatis,theTissueProbabilityMaps(TPM)
arenotusedconstantlyinthesenseoftheclassicalUnifiedSegmentationapproach(Ashburneret.al.
2005),butonlyforspatialnormalizationandtheinitialskull-stripping.ThefollowingAMAPestimation
isadaptiveinthesensethatlocalvariationsoftheparameters(i.e.,meansandvariance)aremodelled
as slowly varying spatial functions (Rajapakse et al. 1997). This not only accounts for intensity
inhomogeneitiesbutalsoforotherlocalvariationsofintensity.
3
TheclassicSPM12segmentationisstillusedinaddition,butonlytoinitiallyremovenon-braintissuefromtheimageand
togetastartingestimateforthesegmentation.
49
PartialVolumeSegmentation
Additionally, the segmentation approach uses a Partial Volume Estimation (PVE) with a simplified
mixedmodelofatmosttwotissuetypes(Tohkaetal.2004).Westartwithaninitialsegmentation
intothreepureclasses:graymatter(GM),whitematter(WM),andcerebrospinalfluid(CSF)basedon
theabovedescribedAMAPestimation.TheinitialsegmentationisfollowedbyaPVEoftwoadditional
mixedclasses:GM-WMandGM-CSF.Thisresultsinanestimationoftheamount(orfraction)ofeach
puretissuetypepresentineveryvoxel(assinglevoxels-givenbytheirsize-probablycontainmore
thanonetissuetype)andthusprovidesamoreaccuratesegmentation.
Denoising
Furthermore, we apply two denoising methods. The first method is a spatial-adaptive Non-Local
Means(SANLM)denoisingfilter(Manjónetal.2010)thatisappliedaftertheintensitynormalization.
This filter removes noise while preserving edges and is implemented as preprocessing step. The
secondmethodisaclassicalMarkovRandomField(MRF)approach,whichincorporatesspatialprior
informationofadjacentvoxelsintothesegmentationestimation(Rajapakseetal.1997)andispartof
the AMAP segmentation. The strength of the filters is automatically obtained by estimating the
remainingnoiseintheimage.
DartelNormalisation
Another important extension to the SPM12 segmentation is the integration of the Dartel
normalisation(Ashburner2007)intothetoolboxbyanalreadyexistingDarteltemplateinMNIspace.
Thistemplatewasderivedfrom555healthycontrolsubjectsoftheIXI-database(http://www.braindevelopment.org) and is provided in MNI space4 for six different iteration steps of Dartel
normalisation.Thus,forthemajorityofstudiesthecreationofsample-specificDarteltemplatesisnot
necessaryanymore5.
LocalAdaptiveSegmentation(LAS)
Beside WM-inhomogeneities, also the GM intensity can vary for different regions like the motor
cortex, the basal ganglia, or the occipital lobe. Although, these changes have an anatomical
background (e.g. iron content, myelenization), they depend on the MR-protocol and often lead to
GM-underestimationsforhigherintensitiesandCSF-overestimationsforlowerintensities.Therefore,
alocalintensitytransformationofalltissueclassesisusedinordertoreducethiseffectsintheimage
before the final AMAP segmentation. The strength of the changes is controlled by the LASstr
parameter,with0fornoLAS,smallvalues(0.01-0.5)forsmalladaptations,0.5foraverageadaptation
(default),andhighervalues(0.5-1)forstrongadaptations.
4
Thus,noadditionalMNInormalizationisnecessary.
5
For studies investigating data of children I still recommend creating a customized Dartel template. Of note, for this
optionarepresentativesamplewithasufficientnumberofsubjectsisrequired.
50
Skull-Stripping
CAT12containsarevisedgraph-cutbasedskull-strippingwithaarbitrarystrength,with0foramore
liberalandwiderbrainmasksand1foramoreaggressiveskull-stripping.Thedefaultis0.5andwas
successfullytestedonavarietyofdifferentimages.
Thestrengthparameteraffectsmultipleinternalparameters:
• Intensitythresholdstodealwithblood-vesselsandmeninges
• Distanceandgrowingparametersforthegraph-cut/region-growing
• Closingparametersthatfillthesulci
• Smoothingparametersthatallowsharperorwiderresults
Ifyoursegmentationsstillcontainskullandothernon-braintissue(e.g.dura)youcantrytoincrease
thestrength.Ifpartsofthebrainaremissinginthesegmentationsthestrengthcanbedecreased.
AffinePreprocessing(APP)
To improve the initial SPM segmentation an initial affine registration on a bias-corrected image is
appliedandtheintensityrangeislimitedtoavoidproblemsinspecialprotocols.Ifpreprocessingfails
a more aggressive version is available that applies a rough bias correction and removes non-brain
partsthebrainbeforetheinitialaffineregistration.
Cleanup
CAT12includesanewcleanuproutinethatusesmorphological,distanceandsmoothingoperationsto
removeremainingmeningesfromthefinalsegmentation.Thestrengthofthecleanupiscontrolledby
the cleanupstr parameter, with 0 for no cleanup, low values <0.5 for light cleanup, 0.5 for average
cleanup(default),and1forstrongcleanup.
Interpolation
CAT12 uses an internal interpolation in order to allow more reliable results also for low resolution
imagesandanisotropicspatialresolutions.Althoughaninterpolationcannotaddfurtherdetailstothe
images, some of the used functions benefit from the higher number of voxels and the common
stripedartefactsinmodulatedimagesarestronglydiminished.
51
References
J.Ashburner(2005):Unifiedsegmentation.Neuroimage26(3):839-51.
J.Ashburner(2007):Afastdiffeomorphicimageregistrationalgorithm.Neuroimage38(1):95-113.
E.Luders,P.M.Thompson,K.L.Narr,A.W.Toga,L.Jancke,C.Gaser(2006):Acurvature-basedapproachto
estimatelocalgyrificationonthecorticalsurface.Neuroimage,29(4):1224-30.
C.Gaser,H.-P.Volz,S.Kiebel,S.Riehemann,H.Sauer(1999):Detectingstructuralchangesinwholebrainbased
onnonlineardeformations–applicationtoschizophreniaresearch.Neuroimage,10:107-113.
C.Gaser,I.Nenadic,B.Buchsbaum,E.Hazlett,M.S.Buchsbaum(2001):Deformation-basedmorphometryand
itsrelationtoconventionalvolumetryofbrainlateralventriclesinMRI.Neuroimage,13:1140-1145.
C.Gaser,R.Dahnke(2016).CAT-AComputationalAnatomyToolboxfortheAnalysisofStructuralMRIData.
HBM2016.http://www.neuro.uni-jena.de/hbm2016/GaserHBM2016.pdf
R.Dahnke,R.Yotter,C.Gaser(2012a).Corticalthicknessandcentralsurfaceestimation.Neuroimage,65:336348.
R.Dahnke,G.Ziegler,C.Gaser(2012b).LocalAdaptiveSegmentation.HBM2012.http://www.neuro.unijena.de/hbm2012/HBM2012-Dahnke02.pdf
R.Desai,E.Liebenthal,E.T.Possing,E.Waldron,J.R.Binder(2005):Volumetricvs.surface-basedalignmentfor
localizationofauditorycortexactivation.Neuroimage26(4):1019-29.
C.D. Good, I.S. Johnsrude, J. Ashburner, R.N. Henson, K.J. Friston, R.S. Frackowiak (2001): A voxel-based
morphometricstudyofageingin465normaladulthumanbrains.Neuroimage.14(1Pt1):21-36.
I.B.Malone,K.K.Leung,S.Clegg,J.Barnes,J.L.Whitwell,J.Ashburner,N.C.Fox,G.R.Ridgway(2015):Accurate
automaticestimationoftotalintracranialvolume:anuisancevariablewithlessnuisance.Neuroimage104:36672.
J.Manjon,PCoupe,L.Marti-Bonmati,D.L.Collins,M.Robles(2010).AdaptiveNon-LocalMeansDenoisingof
MRImagesWithSpatiallyVaryingNoiseLevelsJournalofMagneticResonanceImaging,31:192-203.
J.C. Rajapakse, J.N. Giedd, J.L. Rapoport (1997): Statistical Approach to Segmentation of Single-Channel
CerebralMRImages.IEEETrans.Med.Imag.16(2):176-186.
J. Tohka, A. Zijdenbos, A. Evans (2004): Fast and robust parameter estimation for statistical partial volume
modelsinbrainMRI.Neuroimage23(1):84-97.
R.A.Yotter,R.Dahnke,P.M.Thompson,C.Gaser(2011a):TopologicalCorrectionofBrainSurfaceMeshesUsing
SphericalHarmonics.HumanBrainMapping,32(7):1109-24.
R.A. Yotter, G. Ziegler, I. Nenadic, P.M. Thompson,C. Gaser (2011b): Local cortical surface complexity maps
fromsphericalharmonicreconstructions.Neuroimage,56(3):961-973.
R.A. Yotter, P.M. Thompson,C. Gaser (2011c): Algorithms to Improve the Re-Parameterization of Spherical
MappingsofBrain.JournalofNeuroimaging,21(2):e134-47.
R.A.Yotter,G.Ziegler,P.M.Thompson,C.Gaser(2011d):DiffeometricAnatomicalRegistrationontheSurface.
HBM2011.
14-Jun-17
[email protected]
[email protected]
52