Zurich SPM Course 2014 Voxel-Based Morphometry & DARTEL

Course

Zurich SPM Course 2014

Voxel-Based Morphometry& DARTELGed Ridgway, LondonWith thanks to John Ashburnerand the FIL Methods Group1Examples applications of VBMMany scientifically or clinically interesting questions might relate to the local volume of regions of the brainFor example, whether (and where) local patterns of brain morphometry help to:Distinguish groups (e.g. schizophrenics and healthy controls)Explain the changes seen in development and aging Understand plasticity, e.g. when learning new skillsFind structural correlates (scores, traits, genetics, etc.)Some more unusual examplesVBM and altruismMorishima et al. (2012) DOI:10.1016/j.neuron.2012.05.021individual differences in GM volume in TPJ not only translate into individual differences in the general propensity to behave altruistically, but they also create a link between brain structure and brain function

VBM for fMRI in the presence of atrophyGoll et al. (2012) PMID:22405732

fMRI adjusted for VBM

VBM and political orientationRyota Kanai, Tom Feilden, Colin Firth, Geraint ReesPolitical Orientations Are Correlated with Brain Structure in Young Adults. DOI:10.1016/j.cub.2011.03.017

Tissue segmentation for VBMHigh-resolution MRI reveals fine structural detail in the brain, but not all of it reliable or interestingNoise, intensity-inhomogeneity, vasculature, MR Intensity is usually not quantitatively meaningful (in the same way that e.g. CT is)fMRI time-series allow signal changes to be analysedQuantitative MRI is possible though, and promising, see e.g. Draganski et al. (2011) PMID:21277375Regional volumes of the three main tissue types: gray matter, white matter and CSF, are well-defined and potentially very interestingVoxel-Based MorphometryIn essence VBM is Statistical Parametric Mapping of regional segmented tissue density or volume

The exact interpretation of gray matter density or volume is complicated, and depends on the preprocessing steps usedIt is not interpretable as neuronal packing density or other cytoarchitectonic tissue propertiesThe hope is that changes in these microscopic properties may lead to macro- or mesoscopic VBM-detectable differences7VBM overviewUnified segmentation and spatial normalisationMore flexible groupwise normalisation using DARTELModulation to preserve tissue volumeOtherwise, tissue densityOptional computation of tissue totals/globalsGaussian smoothingVoxel-wise statistical analysisVBM in picturesSegment

Normalise

VBM in pictures

Segment

Normalise

Modulate

Smooth

VBM in pictures

Segment

Normalise

Modulate

Smooth

Voxel-wise statistics

11VBM in picturesSegment

Normalise

Modulate

Smooth

Voxel-wise statistics

beta_0001con_0001ResMSspmT_0001FWE < 0.0512VBM SubtletiesModulationHow much to smoothInterpreting resultsAdjusting for total GM or Intracranial VolumeStatistical validityMultiplication of warped (normalised) tissue intensities so that their regional total is preservedCan detect differences in completely registered areasOtherwise, we preserve concentrations, and are detecting mesoscopic effects that remain after approximate registration has removed the macroscopic effectsFlexible (not necessarily perfect) warping leaves less112/31/31/32/31111Native

intensity =tissue probabilityModulatedUnmodulatedSee also http://tinyurl.com/ModulationTutorial Modulation(preserve amounts)Clarify, modulation not a step (as spm2) but an option in the segment and the normalise GUIs14

Modulation(preserve amounts)Top shows unmodulateddata (wc1), with intensity or concentration preservedIntensities are constant

Below is modulated data (mwc1) with amounts or totals preservedThe voxel at the cross-hairs brightens as more tissue is compressed at this point15SmoothingThe analysis will be most sensitive to effects that match the shape and size of the kernelThe data will be more Gaussian and closer to a continuous random field for larger kernelsUsually recommend >= 6mmResults will be rough and noise-like if too little smoothing is usedToo much will lead to distributed, indistinct blobsUsually recommend ROI: no subjective (or arbitrary) boundariesVBM < ROI: harder to interpret blobs & characterise errorInterpreting findingsThickeningThinningMis-classifyMis-registerMis-registerContrastFoldingInterpreting findingsVBM is sometimes described asunbiased whole brain volumetryRegional variation in registration accuracySegmentation problems, issues with analysis maskIntensity, folding, etc.But significant blobs probably still indicate meaningful systematic effects!20Adjustment for nuisance variablesAnything which might explain some variability in regional volumes of interest should be consideredAge and gender are obvious and commonly usedConsider age+age2 to allow quadratic effectsSite or scanner if more than one(Note: model as factor, not covariate; multiple columns of dummies)Interval in longitudinal studiesSome 12-month intervals end up months longerTotal grey matter volume often used for VBMChanges interpretation when correlated with local volumes (shape is a multivariate concept See next slide)Total intracranial volume (TIV/ICV) sometimes more powerful and/or more easily interpretable, see alsoBarnes et al., (2010), NeuroImage 53(4):1244-55Adjustment for total/global volumeShape is really a multivariate conceptDependencies among volumes in different regionsSPM is mass univariateCombining voxel-wise information with global integrated tissue volume provides a compromiseUsing either ANCOVA or proportional scaling

(ii) is globally thicker, but locally thinner than (i) either of these effects may be of interest to us.Fig. from: Voxel-based morphometry of the human brain Mechelli, Price, Friston and Ashburner. Current Medical Imaging Reviews 1(2), 2005.Note globals dont help distinguish the thickened or folded cortex...22VBMs statistical validityResiduals are not normally distributedLittle impact for comparing reasonably sized groupsPotentially problematic for comparing single subjects or tiny patient groups with a larger control group(Scarpazza et al, 2013; DOI: 10.1016/j.neuroimage.2012.12.045)Mitigate with large amounts of smoothingOr use nonparametric tests, e.g. permutation testing (SnPM)Though also not suitable for single case versus control group Smoothness is not spatially stationaryBigger blobs expected by chance in smoother regionsNS toolbox http://www.fil.ion.ucl.ac.uk/spm/ext/#NS Voxel-wise FDR is common, but not recommendedLongitudinal VBMThe simplest method for longitudinal VBM is to use cross-sectional preprocessing, but longitudinal statisticsStandard preprocessing not optimal, but unbiasedNon-longitudinal statistical analysis would be severely biased(Estimates of standard errors would be too small)Simplest longitudinal statistical analysis: two-stage summary statistic approach (like in fMRI)Contrast on the slope parameter for a linear regression against time within each subjectFor two time-points with interval approximately constant over subjects, equivalent to simple time2 time1 difference image24Longitudinal VBM variationsIntra-subject registration over time is much more accurate than inter-subject normalisationA simple approach is to apply one set of normalisation parameters (e.g. estimated from baseline images) to both baseline and repeat(s)Draganski et al (2004) Nature 427: 311-312More sophisticated approaches use nonlinear within-subject registration (ideally symmetric or unbiased)New Longitudinal Registration toolbox in SPM12(Ashburner & Ridgway, 2013; PMID: 23386806)Either pure TBM or combine with segmentation of within-subject averages (e.g. Rohrer et al, 2013; PMID: 23395096)25Spatial normalisation with DARTELVBM is crucially dependent on registration performanceThe limited flexibility of DCT normalisation has been criticisedInverse transformations are useful, but not always well-definedMore flexible registration requires careful modelling and regularisation (prior belief about reasonable warping)MNI/ICBM templates/priors are not universally representativeThe DARTEL toolbox combines several methodological advances to address these limitationsMotivation for using DARTELRecent papers comparing different approaches have favoured more flexible methodsDARTEL outperforms SPMs old DCT normalisationAlso comparable to the best algorithms from other software packages (though note that DARTEL and others have many tunable parameters...)Klein et al. (2009) is a particularly thorough comparison, using expert segmentationsResults summarised in the next slide

Part of Fig.1 in Klein et al.

Part of Fig.5 in Klein et al.

DARTEL TransformationsDisplacements come from integrating flow fieldsRegularise velocity not displacement(syrup instead of elastic)3 (x,y,z) DF per 1.5mm cubic voxel10^6 DF vs. 10^3 DCT bases

Scaling and squaring is used in DARTEL, more complicated again in latest work (Geodesic Shooting)Consistent inverse transformation is easily obtained, e.g. integrate -u

DARTEL objective functionLikelihood component (matching) Specific for matching tissue classesMultinomial assumption (cf. Gaussian)Prior component (regularisation)A measure of deformation (flow) roughness/energy (uTHu)Need to choose form and weighting(s) of regularisationDefaults usually work well (e.g. even for AD)But be aware that different regularisation is a different model, so can lead to differences in the resultsSimultaneous registration of GM to GM and WM to WM, for a group of subjectsGrey matter White matterGrey matter White matterGrey matter White matterGrey matter White matterGrey matter White matterTemplateSubject 1Subject 2Subject 3Subject 4DARTEL averagetemplate evolution

Rigid average(Template_0)Average ofmwc1 usingsegment/DCTTemplate6Template160 images from OASIS cross-sectional data (as in VBM practical)32

SummaryVBM performs voxel-wise statistical analysis on smoothed (modulated) normalised tissue segmentsSPM performs segmentation and spatial normalisation in a unified generative modelBased on Gaussian mixture modelling, with warped spatial prior probability maps, and multiplicative bias fieldSubsequent (non-unified) use of DARTEL improves spatial normalisation for VBM(and probably also fMRI...)Key references for VBMAshburner & Friston (2005) Unified Segmentation.NeuroImage 26:839-851Mechelli et al. (2005) Voxel-based morphometry of the human brain Current Medical Imaging Reviews 1(2)Ashburner (2007) A Fast Diffeomorphic Image Registration Algorithm. NeuroImage 38:95-113Ashburner & Friston (2009) Computing average shaped tissue probability templates. NeuroImage 45(2):333-341Ashburner & Friston (2011) Diffeomorphic registration using geodesic shooting and Gauss-Newton optimisation. NeuroImage 55(3):954-67 PMID:21216294