Presentation on theme: "MfD Voxel-Based Morphometry (VBM)"— Presentation transcript:
1 MfD 2013-2014 Voxel-Based Morphometry (VBM) Andrea Gajardo-VidalClarisse Aichelburg
2 OVERVIEW VBM: Basic Approach, Basic Idea & Application, Advantage The original VBM: Spatial normalisation, segmentation, smoothing and statistical analysis.VBM: new approachWhat does SPM show in VBM?: DARTEL toolboxSome important applications (literature hints)Further Considerations: Potential Confounds, Statistical AnalysisTake home messagesReferences
3 1.VBM: Basic ApproachVoxel-wise comparison of the local volume of GM between two groups of subjectsinvolves spatially normalizing high-resolution images from all participants into the same stereotactic space, segmenting and smoothing GM segmentsVoxel-wise parametric statistical tests comparing the smoothed GM images from the two groups is performed
4 1.VBM: Basic Idea & Application Not biased to one particular structure, instead assesses anatomical differences throughout the brainInvestigates patterns of brain changein developmentin diseasedue to learning/ brain plasticityAs well as neuroanatomical correlates of subject characteristicsScoresTraitsGenetic influences
5 1.VBM: Advantagelink between an individual’s performance or trait as measured in an ecologically valid environment (outside the scanner) to brain structures measurements obtained in the scannercan administer multiple tasks to the same set of participants, and correlations between tasks can be analyzedone-to-one mapping between a cognitive function and the structure of a brain regionwhen a single region is identified it needs to be interpreted in the context of known functions of the region
6 2. VBM: The original proposal (Ashburner & Friston, 2000) Spatial normalisation: involves transforming all the subjects’ data to the same stereotactic space.Tissue segmentation:the spatially normalized images are next partitioned into gray matter (GM), white matter (WM), cerebrospinal fluid (CSF) and three other background classes, using a modified mixture model cluster analysis technique which identifies voxel intensity distributions of particular tissue types.Smoothing: the gray matter images are now smoothed by convolving with an isotropic Gaussian kernel. This makes the subsequent voxel-by-voxel analysis comparable to a region of interest approach.Logit transform: Often, prior to performing statistical tests on measures of concentration, the data are transformed using the logit transformation in order to render them more normally distributed.Statistical analysis: the final step of a VBM analysis involves a voxel-wise statistical analysis. This employs the general linear model (GLM), a flexible framework that allows a variety of different statistical tests such as group comparisons and correlations with covariates of interest.
9 VBM: New approachMore recently, there was a change to the VBM framework, such that the pre-processed data were scaled such that the total volume of tissue in each structure is preserved after warping the data to a standard reference space . This correction is by scaling by the Jacobian determinant of the deformation and is colloquially known as MODULATION.” The result is that the pre-processed data represent a quantitative measure (tissue volume per unit volume of spatially normalized image).Unified segmentation and spatial normalisationMore flexible group-wise normalisation using DARTELVolume-preserving transformation/warpingGaussian smoothingOptional computation of tissue totals/globalsVoxel-wise statistical analysis
10 Segmentation T1 images are partioned into: grey matterwhite matterCSFExtra tissue maps can be generated (2 additional ones)Segmentation is achieved by combining :Tissue Probability maps/ Bayesion Priors (TPMs) based on general knowledge about normal tissue distribution withMixture model cluster analysis (which identifies voxel intensity distributions of particular tissue types in the original image).***In order to separate grey matter from other tissues, there must be a high contrast between grey matter and other surrounding tissues. If grey matter is not clearly visible, then it cannot be precisely segmented
11 Segmentation: TPMs – Tissue prior probability maps Each TPM indicates the prior probability for a particular tissue at each point in MNI spaceTPMs are warped to match the subject.ICBM Tissue Probabilistic Atlases. These tissue probability maps are kindly provided by the International Consortium for Brain Mapping, John C. Mazziotta and Arthur W. Toga.
12 Segmentation Intensities in the image fall into roughly 3 classes. Intensity information in the image itselfIntensities in the image fall into roughly 3 classes.SPM assigns a voxel to a tissue class based on its intensity relative to the others in the image.Each voxel has a value between 0 and 1, representing the probability of it being in that particular tissue class.
13 4. DARTEL toolboxDARTEL=Diffeomorphic Anatomical Registration using Exponentiated Lie algebra (DARTEL).More sophisticated registration model which improves image registrationMore precise inter-subject alignment (multiple iterations: more than registrations rather than registration parameters from the previous model).More sensitive to identify more accurate localization
14 Spatial normalisation with DARTEL Limited flexibility registration has been criticisedMNI/ICBM templates/priors are not universally representativeVBM is crucially dependent on registration performanceInverse transformations are useful, but not always well-definedMore flexible registration requires careful modelling and regularisation (prior belief about reasonable warping)The DARTEL toolbox combines several methodological advances to address these limitationsEvaluations show DARTEL performs at state-of-the art (see Klein et al., (2009) NeuroImage 46(3): )
15 Modulation (preserved amount) Voxel intensities can be further modulated by determinant of Jacobean of warps: i.e, whether want to compare gray matter density (unmodulated) or volume (modulated).During modulation voxel intensities are multiplied by the local value in the deformation field from normalisation, so that total GM/WM signal remains the sameChange of intensity now represents volume relative to templateMultiplication of the warped (normalised) tissue intensities so that their regional or global volume is preserve
16 Modulated vs. Unmodulated VolumeComparison between absolute volumes of GM or WM structuresUseful for looking at the effects of degenerative diseases or atrophyUnmodulatedConcentration/ densityproportion of GM (or WM) relative to other tissue types within a regionHard to interpretIt may be useful for highlighting areas of poor registration (perfectly registered unmodulated data should show no differences between groups)
17 SmoothingThe Jacobian-corrected warped tissue class images would then be blurred by low-pass filtering the imagesWith isotropic Gaussian kernelusually between 7 & 14 mmChoice of kernel changes statsEffect: data becomes more normally distributedEach voxel contains average GM and WM concentration from an area around the voxel (as defined by the kernel)Brilliant for statistical tests (central limit theorem)Compensates for inexact nature of spatial normalisation, “smoothes out” incorrect registration
18 SmoothingThe data are typically convolved with a Gaussian, so the result is a weighted count of voxels containing the tissue (see figure).***The degree of blurring should relate to the accuracy with which the data can be registered, more blurring if the intersubject registration is less accurate.
20 Statistical analysis :GLM Statistical analysis of the preprocessed data are performed by fitting a GLM at each voxel.Y = Xβ + εVoxel wise (independent statistical tests for every single voxel).The general principle: a design matrix is specified, which model the source of variance among the data. For example, if were N subjects in the study , then the design matrix would contain N arrows.Outcome: statistical parametric maps, showing areas of significant difference/ correlationsLook like blobsUses same software as fMRI
21 Using the GLM for VBMTypically the matrix may contain blocks that represent the group from which each subject belongs (Ashburner, 2008).e.g., compare the GM/ WM differences between 2 groups (healthy controls vs. some population of patients)H0: there is no difference betweenthese groupsβ: other covariates, not just the mean
22 Validity of the statistical tests in SPM Errors (residuals) need to be normally distributed throughout brain for stats to be validInvalidates experiments that compare one subject with a groupCorrection for multiple comparisonsValid for corrections based on peak heights (voxel-wise)This requires smoothness of residuals to be uniformly distributed but it’s not in VBM because of the non-stationary nature of underlying neuroanatomyBigger blobs expected in smoother regions, purely by chanceAlternativesImprove normalisation
27 When we normalise to MNI space we also are smoothing your images. With this final step, the pre-processing is done!Now we can perform statistical analysis of the preprocessed data.
28 VBM: most important applications to healthy subjects A Voxel-Based Morphometric Study of Ageing in 465 Normal Adult Human Brains (Good et al., 2001): introduction of an optimised version of VBMRegional effect of age: Our data support the theory of a heterogeneic response of various compartments of the brain to ageing. We observed accelerated loss of grey matter volume symmetrically in both parietal lobes (angula gyri), preand postcentral gyri, insula, and anterior cingulate cortex.
29 VBM: most important application to diseased subjects Voxel-Based Morphometry of the Human Brain: Methods and Applications (Mechelli et al., 2005)For example in: developmental and congenital disorders , autism , bipolar disorders, temporal lobe epilepsy, Down’s syndrome Parkinson’s disease , Huntington's disease , Alzheimer’s disease and primary progressive aphasia . The use of VBM with highly distorted brains presents special challenges however, due to the difficulties that arise during spatial normalisation. Of particular interest, from a methodological point of view, is a investigation of temporal lobe epilepsy by Keller et al. (2004). The authors used both standard and optimised VBM and furthermore compared “modulated” and “non-modulated” procedures.
30 5.Further Considerations: Potential Confounds Behavioural and MRI measurements must be reliablee.g. images must be acquired from the same scanner and same MR sequencesParticipantssample sizethe larger the sample size, the greater the power to detect differencesthough if the effect size is large, differences can be observed with smaller cohortse.g. age, gender ratios, education, disease severity across groups
31 5.Further Considerations: Potential Confounds ProcessingVBM analyses results should reflect systematic volumetric differences, such as folding or thickness, rather than artifacts, such as misclassification or misregistrationsteps often vary across studies (Whitwell and Jack, 2005)
32 5.Further considerations Different brains have different sizesRegional volumes are likely to vary as a function of the whole brain volumeNeed to correct for “global brain volume”Total GM volumeWhole brain volumeTotal intra-cranial volumeBrains of similar size with GM differences globally and locally (Mechelli et al., 2005)
33 5.Further considerations: Statistical Analysis The multiple-comparison problemIn VBM, depending on your resolutionvoxelsstatistical testsSo at p < .0550000 false positivesdifferent options for correcting for multiple comparisonse.g. Bonferroni, FEW, FDRchanging the p value and using different corrections will change the number of voxels that exceed the significance threshold
34 5. Further Considerations: Correlations Correlations do not imply causal relationshipsVBM Independent ROI Analysesinvestigating possible anatomical convergence between functional and morphological methodsBrain stimulation techniques can provide independent support for a causal link between structure and functiondisrupting the function of a region via brain stimulation (TMS, tDCS) can confirm the functional involvement of the area in a task
35 5. Further Considerations: Correlations Positive & Negative CorrelationsEvidence for the notion of “the more brain the better”versus evidence associating better performance with less brain volumedepends on specific brain region as well as particular process assessedWhat is the physiological link between the interindividual variability in a given structure and a cognitive ability?significance of “macroscopic” variations, whatever their direction, is not yet understood in terms of “microscopic” variations (Eriksson et al., 2009)
36 6. Take home messagesVBM performs voxel-wise statistical analysis on smoothed (modulated) normalised tissue segmentsSPM8 performs segmentation and spatial normalisation in a unified generative modelBased on Gaussian mixture modelling, with DCT-warped spatial priors, and multiplicative bias fieldThe new segment toolbox includes non-brain priors and more flexible/precise warping of themSubsequent (currently non-unified) use of DARTEL improves normalisation for VBM.
37 6. Take home messagesVBM uses the machinery of SPM to localise patterns in regional volumetric variationUse of “globals” as covariates is a step towards multivariate modelling of volume and shapeMore advanced approaches typically benefit from the same preprocessing methodsNew segmentation and DARTEL close to state of the artThough possibly little or no smoothing
38 6. Take home messages Consider potential confounds and correlations 10 Rules for reporting VBM studies (Ridgway et al., 2008)set out rationale of study and describe data fullyexplain how brain segmentations are produceddescribe method of inter-subject spatial normalizationmake your statistical design transparentbe clear about the significance of your findingspresent results unambiguouslyclarify and justify any non-standard statistical analysisguard against common pitfallsrecognize the limitations of the techniqueinterpret your results cautiously and in context
39 7. Key papers & References Ashburner & Friston (2000, NeuroImage, Voxel-Based Morphometry – The Methods)Ridgway et al. (2008, NeuroImage, Ten simple rules for reporting voxel-based morphometry studies)Kanai & Rees (2011, Nature Reviews Neuroscience, The structural basis of inter-individual differences in human behaviour and cognition)Logothetis (2008, Nature, What we can do and what we cannot do with fMRI)Mechelli, Price, Friston & Ashburner (2005). Voxel-based morphometry of the human brain: methods and applications. Current Medical Imaging Reviews, 1:Ashburner (2009). Computational anatomy with the SPM software. Magnetic Resonance Imaging, 27: 1163 – 1174Good, C., Johnsrude, I. S., Ashburner, J., Henson, R. N. A., Friston, K. J., Frackowiak, R. S. J. (2001). A voxel-based morphometric study of ageing in 465 normal adult human brains. NeuroImage 14(1), 21-36
40 Thank you for listening… and special thanks to our expert John Ashburner