VBM Voxel-based morphometry

Slides:



Advertisements
Similar presentations
DARTEL John Ashburner 2008.
Advertisements

VBM Susie Henley and Stefan Klöppel Based on slides by John Ashburner
A Growing Trend Larger and more complex models are being produced to explain brain imaging data. Bigger and better computers allow more powerful models.
SPM5 Segmentation. A Growing Trend Larger and more complex models are being produced to explain brain imaging data. Bigger and better computers allow.
Overview of SPM p <0.05 Statistical parametric map (SPM)
Concepts of SPM data analysis Marieke Schölvinck.
Detecting Subtle Changes in Structure
MRI preprocessing and segmentation.
Gordon Wright & Marie de Guzman 15 December 2010 Co-registration & Spatial Normalisation.
Voxel-Based Analysis of Quantitative Multi-Parameter Mapping (MPM) Brain Data for Studying Tissue Microstructure, Macroscopic Morphology and Morphometry.
Introduction to Functional and Anatomical Brain MRI Research Dr. Henk Cremers Dr. Sarah Keedy 1.
Basics of fMRI Preprocessing Douglas N. Greve
OverviewOverview Motion correction Smoothing kernel Spatial normalisation Standard template fMRI time-series Statistical Parametric Map General Linear.
MfD Voxel-Based Morphometry (VBM)
Coregistration and Normalisation By Lieke de Boer & Julie Guerin.
Preprocessing: Coregistration and Spatial Normalisation Cassy Fiford and Demis Kia Methods for Dummies 2014 With thanks to Gabriel Ziegler.
Realigning and Unwarping MfD
Multiple testing Justin Chumbley Laboratory for Social and Neural Systems Research Institute for Empirical Research in Economics University of Zurich With.
Multiple testing Justin Chumbley Laboratory for Social and Neural Systems Research Institute for Empirical Research in Economics University of Zurich With.
Bayesian models for fMRI data Methods & models for fMRI data analysis 06 May 2009 Klaas Enno Stephan Laboratory for Social and Neural Systems Research.
Spatial preprocessing of fMRI data Methods & models for fMRI data analysis 25 February 2009 Klaas Enno Stephan Laboratory for Social and Neural Systrems.
Spatial preprocessing of fMRI data
The structural organization of the Brain Gray matter: nerve cell bodies (neurons), glial cells, capillaries, and short nerve cell extensions (axons and.
Multiple comparison correction Methods & models for fMRI data analysis 29 October 2008 Klaas Enno Stephan Branco Weiss Laboratory (BWL) Institute for Empirical.
SPM+fMRI. K space K Space Mechanism of BOLD Functional MRI Brain activity Oxygen consumptionCerebral blood flow Oxyhemoglobin Deoxyhemoglobin Magnetic.
Preprocessing II: Between Subjects John Ashburner Wellcome Trust Centre for Neuroimaging, 12 Queen Square, London, UK.
Surface-based Analysis: Intersubject Registration and Smoothing
Voxel-Based Morphometry John Ashburner Wellcome Trust Centre for Neuroimaging, 12 Queen Square, London, UK.
FMRI Preprocessing John Ashburner. Contents *Preliminaries *Rigid-Body and Affine Transformations *Optimisation and Objective Functions *Transformations.
Introduction to SPM SPM fMRI Course London, May 2012 Guillaume Flandin Wellcome Trust Centre for Neuroimaging University College London.
VBM Voxel-Based Morphometry
Voxel-Based Morphometry John Ashburner Wellcome Trust Centre for Neuroimaging, 12 Queen Square, London, UK.
Voxel Based Morphometry
Lecture 24: Cross-correlation and spectral analysis MP574.
Co-registration and Spatial Normalisation
MNTP Trainee: Georgina Vinyes Junque, Chi Hun Kim Prof. James T. Becker Cyrus Raji, Leonid Teverovskiy, and Robert Tamburo.
1 Detecting Subtle Changes in Structure Chris Rorden –Voxel Based Morphometry Segmentation – identifying gray and white matter Modulation- adjusting for.
Preprocessing of FMRI Data fMRI Graduate Course October 23, 2002.
Anatomical Measures John Ashburner zSegmentation zMorphometry zSegmentation zMorphometry.
FIL SPM Course Oct 2012 Voxel-Based Morphometry Ged Ridgway, FIL/WTCN With thanks to John Ashburner.
DTU Medical Visionday May 27, 2009 Generative models for automated brain MRI segmentation Koen Van Leemput Athinoula A. Martinos Center for Biomedical.
Coregistration and Spatial Normalisation
Voxel Based Morphometry
Voxel-based morphometry The methods and the interpretation (SPM based) Harma Meffert Methodology meeting 14 april 2009.
Statistical Parametric Mapping Lecture 11 - Chapter 13 Head motion and correction Textbook: Functional MRI an introduction to methods, Peter Jezzard, Paul.
Bayesian models for fMRI data Methods & models for fMRI data analysis November 2011 With many thanks for slides & images to: FIL Methods group, particularly.
MINC meeting 2003 Pipelines: analyzing structural MR data Jason Lerch.
Spatial Smoothing and Multiple Comparisons Correction for Dummies Alexa Morcom, Matthew Brett Acknowledgements.
SPM Pre-Processing Oli Gearing + Jack Kelly Methods for Dummies
The general linear model and Statistical Parametric Mapping I: Introduction to the GLM Alexa Morcom and Stefan Kiebel, Rik Henson, Andrew Holmes & J-B.
MfD Co-registration and Normalisation in SPM
Methods for Dummies Voxel-Based Morphometry (VBM)
Baseline and longitudinal patterns of brain atrophy in MCI patients, and their use in prediction of Short- term conversion to AD: Results from ADNI Xuejiao.
Surface-based Analysis: Inter-subject Registration and Smoothing
The general linear model and Statistical Parametric Mapping
Voxel-based Morphometric Analysis
Surface-based Analysis: Intersubject Registration and Smoothing
Keith Worsley Keith Worsley
Computational Neuroanatomy for Dummies
Contrasts & Statistical Inference
Voxel-based Morphometric Analysis
From buttons to code Eamonn Walsh & Domenica Bueti
Voxel-based Morphometric Analysis
SPM2: Modelling and Inference
Preprocessing: Coregistration and Spatial Normalisation
Voxel-based Morphometric Analysis
Anatomical Measures John Ashburner
Contrasts & Statistical Inference
Bayesian Inference in SPM2
Contrasts & Statistical Inference
Presentation transcript:

VBM Voxel-based morphometry Nicola Hobbs & Marianne Novak Thanks to Susie Henley

Overview Background Pre-processing steps Analysis Multiple comparisons Pros and cons of VBM Optional extras

Background VBM is a voxel-wise comparison of local tissue volumes within a group or across groups Whole-brain analysis, does not require a priori assumptions about ROIs; unbiased way of localising structural changes Can be automated, requires little user intervention  compare to manual ROI tracing

Basic Premise Spatial normalisation (alignment) into standard space Segmentation of tissue classes Modulation - adjust for volume changes during normalisation Smoothing - each voxel is a weighted average of surrounding voxels Statistics - localise & make inferences about differences

VBM Processing

Step 1: normalisation Aligns images by warping to standard stereotactic space Affine step – translation, rotation, scaling, shearing Non-linear step Adjust for differences in head position/orientation in scanner global brain shape Any remaining differences (detectable by VBM) are due to smaller-scale differences in volume

SPATIALLY NORMALISED IMAGE NORMALISATION ORIGINAL IMAGE SPATIALLY NORMALISED IMAGE TEMPLATE IMAGE

SPATIALLY NORMALISED IMAGE 2. Tissue segmentation Aims to classify image as GM, WM or CSF Two sources of information a) Spatial prior probability maps b) Intensity information in the image itself GREY MATTER WHITE MATTER CSF SPATIALLY NORMALISED IMAGE Assigns each voxel a (probable) tissue type based on a combination of spatial prior probability maps and the voxel intensity Also includes image intensity non-uniformity correction;

a) Spatial prior probability maps Smoothed average of GM from MNI Intensity at each voxel represents probability of being GM SPM compares the original image to this to help work out the probability of each voxel in the image being GM (or WM, CSF)

b) Image intensities Intensities in the image fall into roughly 3 classes SPM can also assign a voxel to a tissue class by seeing what its intensity is 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 Includes correction for image intensity non-uniformity

Generative model Segmentation into tissue types Bias Correction Normalisation These steps cycled through until normalisation and segmentation criteria are met

Step 3: modulation Corrects for changes in volume induced by normalisation Voxel intensities are multiplied by the local value in the deformation field from normalisation, so that total GM/WM signal remains the same Allows us to make inferences about volume, instead of concentration

Modulation i modulation i / δV normalisation X δV E.g. During normalisation TL in AD subject expands to double the size Modulation multiplies voxel intensities by Jacobian from normalisation process (halve intensities in this case). Intensity now represents relative volume at that point

Is modulation optional? Unmodulated data: compares “the proportion of grey or white matter to all tissue types within a region” Hard to interpret Not useful for looking at e.g. the effects of degenerative disease Modulated data: compares volumes Unmodulated data may be useful for highlighting areas of poor registration (perfectly registered unmodulated data should show no differences between groups)

Step 4: Smoothing Convolve with an isotropic Gaussian kernel Each voxel becomes weighted average of surrounding voxels Smoothing renders the data more normally distributed (Central Limit theorem) Required if using parametric statistics Smoothing compensates for inaccuracies in normalisation Makes mass univariate analysis more like multivariate analysis Filter size should match the expected effect size Usually between 8 – 14mm

Smoothing SMOOTH WITH 8MM KERNEL 8 mm Images are smoothed using a Gaussian kernel (we usually use 8mm FWHM) The intensity of each voxel is therefore a weighted average of the surrounding voxels Makes data more normally distrubuted 8 mm

VBM: Analysis What does the SPM show in VBM? Cross-sectional VBM Multiple comparison corrections Pros and cons of VBM Optional extras

VBM: Cross-sectional analysis overview T1-weighted MRI from one or more groups at a single time point Analysis compares (whole or part of) brain volume between groups, or correlates volume with another measurement at that time point Generates map of voxel intensities: represent volume of, or probability of being in, a particular tissue class Get map of voxel intensities Eg memory test and hippo vol

What is the question in VBM analysis? Control AD Take a single voxel, and ask: “are the intensities in the AD images significantly different to those in the control images for this particular voxel?” eg is the GM intensity (volume) lower in the AD group cf controls? ie do a simple t-test on the voxel intensities

Statistical Parametric Maps (SPM) Repeat this for all voxels Highlights all voxels where intensities (volume) are significantly different between groups: the SPM SPM showing regions where Huntington’s patients have lower GM intensity than controls Colour bar shows the t-value

VBM: group comparison Intensity for each voxel (V) is a function that models the different things that account for differences between scans: V = β1(AD) + β2(control) + β3(covariates) + β4(global volume) + μ + ε + β3(age) + β4(gender) + β5(global volume) + μ + ε V = β1(AD) + β2(control) In practice, the contrast of interest is usually t-test between β1 and β2 Voxel intensity is a function that models all the different things that account for differences between scans. Use AD vs controls as example. Beta value is slope of association of scans or values at that voxel Covariates are explanatory or confounding variables eg “is there significantly more GM in the control than in the AD scans?”

VBM: correlation Correlate images and test scores (eg Alzheimer’s patients with memory score) SPM shows regions of GM or WM where there are significant associations between intensity (volume) and test score V = β1(test score) + β2(age) + β3(gender) + β4(global volume) + μ + ε Contrast of interest is whether β1 (slope of association between intensity & test score) is significantly different to zero Combine group comparison and correlation analyses.

Correcting for Multiple Comparisons 200,000 voxels per scan ie 200,000 t-tests If you do 200,000 t-tests at p<0.05, by chance 10,000 will be false positives Bad practice… A strict Bonferroni correction would reduce the p value for each test to 0.00000025 However, voxel intensities are not independent, but correlated with their neighbours Bonferroni is therefore too harsh a correction and will lose true results

Familywise Error SPM uses Gaussian Random Field theory (GRF)1 Using FWE, p<0.05: 5% of ALL our SPMs will contain a false positive voxel This effectively controls the number of false positive regions rather than voxels Can be thought of as a Bonferroni-type correction, allowing for multiple non-independent tests Good: a “safe” way to correct Bad: but we are probably missing a lot of true positives 1 http://www.mrc-cbu.cam.ac.uk/Imaging/Common/randomfields.shtml

False Discovery Rate FDR more recent q value FDR q<0.05 False Discovery Rate FDR more recent It controls the expected proportion of false positives among suprathreshold voxels only Using FDR, q<0.05: we expect 5% of the voxels for each SPM to be false positives (1,000 voxels) Bad: less stringent than FWE so more false positives Good: fewer false negatives (ie more true positives) But: assumes independence of voxels: avoid….? Voxel

VBM Pros VBM Cons 1. Objective analysis 2. Do not need priors – more exploratory 3. Automated VBM Cons 1. False positives: misregistration, FDR False negatives: FWE More difficult to pick up differences in areas with high inter-subject variance: low signal to noise ratio Comparing VBM to less automated or more manual tools measuring regional tissue volumes

Other VBM Issues Optimised VBM: GM to GM warping, then applied to whole brain image (better GM alignment); Good et al, Neuroimage 2001 (SPM 2) Diffeomorphic warping: DARTEL Multivariate techniques: including classification/SVM Longitudinal scan analysis: two time points especially

18 iterations to form template Ashburner Neuroimage 2007

Standard preprocessing: areas of decreased volume in depressed subjects DARTEL preprocessing: areas of decreased volume in depressed subjects

Longitudinal VBM Baseline and follow-up image are registered together non-linearly (fluid registration), NOT using spm software Voxels at follow-up are warped to voxels at baseline Represented visually as a voxel compression map showing regions of contraction and expansion

Fluid Registered Image FTD (semantic dementia) Voxel compression map 1 year expanding contracting

Optimised VBM Native space images Standard space images 1. Affine registration to SPM2 T1 template 4. Normalisation using parameters from step 3; GM is well-aligned Standard space images Standard space images 2. Segmentation 5. Segmentation GM segments GM segments 3. Estimate normalisation parameters for GM segments to SPM2 GM template 6. Modulation: correcting for spatial changes introduced in normalisation Normalisation parameters GM to GM Mod GM 7. Masking: segments are multiplied by binary region to exclude any non-brain Masked GM 8. Smoothed at 8mm FWHM Smoothed, Masked, mod GM

Resources and references http://www.fil.ion.ucl.ac.uk/spm (the SPM homepage) http://imaging.mrc-cbu.cam.ac.uk/imaging/CbuImaging (neurimaging wiki homepage) http://www.mrc-cbu.cam.ac.uk/Imaging/Common/randomfields.shtml (for multiple comparisons info) Ashburner J, Friston KJ. Voxel-based morphometry--the methods. Neuroimage 2000; 11: 805-821 (the original VBM paper) Good CD, Johnsrude IS, Ashburner J, Henson RN, Friston KJ, Frackowiak RS. A voxel-based morphometric study of ageing in 465 normal adult human brains. Neuroimage 2001; 14: 21-36 (the optimised VBM paper) Ridgway GR, Henley SM, Rohrer JD, Scahill RI, Warren JD, Fox NC. Ten simple rules for reporting voxel-based morphometry studies. Neuroimage 2008.