Hi all,
I am new to SPM, and I am confused by the statistical analysis steps
of the sample (face_rep_SPM2.tar.gz) provided by SPM website. Would
you please help me to find some more details about using SPM2,
especially the statistical analysis of fMRI. My current problem is
that I do not understand the following words:
Press <Results>
Select 'SPM.mat' file
Press 'Define new contrast'
Enter - contrast = '1 0 0 1 0 0 1 0 0 1' [remaining 0's autofilled]
contrast name = 'Canonical HRF: Faces > Baseline'
press 'done'
I do not know the meaning of entered contrast, and the role it plays.
Thank you for your time!
Regards,
Kingaza
<div class="moz-text-flowed" style="font-family: -moz-fixed"> Advanced Event-related fMRI Example Data Set
============================================
Repetition priming for famous and nonfamous faces
Created R. Henson 8/7/00
WDCN & ICN, UCL
[log in to unmask]
Modified for SPM2 by R. Henson 13/5/03
One subject's data from Henson et al. (2002) Cerebral Cortex -
FOR TEACHING PURPOSES ONLY (to illustrate facilities of SPM) -
PLEASE DO NOT CITE WITHOUT PERMISSION
Experiment
----------
2x2 factorial event-related fMRI
One session (one subject)
(Famous vs. Nonfamous) x (1st vs 2nd presentation) of faces against
baseline of chequerboard
2 presentations of 26 Famous and 26 Nonfamous Greyscale photographs,
for 0.5s, randomly intermixed, for fame judgment task (one of two
right finger key presses).
Parameteric factor "lag" = number of faces intervening between
repetition of a specific face + 1
Minimal SOA=4.5s, with probability 2/3 (ie 1/3 null events)
Continuous EPI (TE=40ms,TR=2s) 24 descending slices (64x64 3x3mm2),
3mm thick, 1.5mm gap
Following images are in the Data directory...
sM03953_0005_*.{img,hdr,mat}
------------------------
351 Analyze format functional images (*img contain data, *hdr is Analyze header)
64x64x24 3mmx3mmx4.5mm voxels
< View images with <DISPLAY> button on main SPM window - Note orbitofrontal
and inferior temporal drop-out and ghosting (Effects:brighten if necessary) >
usM03953_0005_*.{img,hdr,mat}
-----------------------------
Coregister all images to first, and then unwarp and reslice
using sinc interpolation (*mat files contain orientation information)
Type:
Select <REALIGN & UNWARP> in the upper left SPM window
'Number of subjects' - 1
'Num Sessions for subject 1' - 1
Select all 351 sM03953_0005_*.{img} and press 'Done'
Pull down menu - 'Model field changes w.r.t'- select 'Pitch & Roll'
Pull down menu - select 'Create what?' - select 'All images + mean'
Output:
u* resliced, unwarped files
mean* mean of resliced files (for normalisation)
rp_sM03953_0005_0006.txt
movement parameters (6x351 values)
spm2.ps
graphical output of translations and rotations
<In spm2.ps, note movement and rotation in three directions>
ausM03953_0005_* .{img,hdr,mat}
-------------------------------
351 functional images temporally realigned to middle slice 12 (of 24)
using sinc interpolation (top slice = 24 = acquired first)
Type:
Press <SLICE TIMING> in the upper middle SPM window
'Number of subjects/sessions' - 1
Select all usM03953_0005_*.img and press 'Done'
Pull down menu - select 'Descending (first slice = top)'
Prompt - Reference Slice - enter '12' (default)
Prompt - Interscan interval (TR) - enter '2.0'
Prompt - Acquisition time (TA) - enter '1.9...' (default)
Output:
au* files interpolated realigned images
< Can use <ImCalc> to create difference image between ausM03953_0005_0100.img
and usM03953_0005_0100.img to see effect of interpolation about middle slice >
wausM03953_0005_*.{img,hdr,mat}
-------------------------------
Normalised interpolated realigned functional images
The normalisation parameters are generated from the mean realigned
image, normalised to the EPI template using affine and nonlinear warps
Optional: To save filespace, can change default voxel size to 3x3x3,
rather than interpolating to 2x2x2:
Press <DEFAULTS> in SPM main window
Pull down menu - 'Spatial Normalization'
Pull down menu - 'Defaults for - writing normalized'
Pull down menu - 'Preserve concentrations'
Pull down menu: Bounding box - 'Default'
Pull down menu: Voxel sizes - '3 3 3'
Pull down menu: Interpolation - 'Trilinear'
Pull down menu: To wrap images - 'no wrap'
Type:
Press <NORMALISE> near upper, middle SPM window
Pull down menu - 'Determine Parameters and Write Normalised'
Select 'Template image' - EPI.img
Select 'Source image, subj1' - meanusM03953_0005_0006.img
Select 'Images to write' - meanusM03953_0005_0006.img
AND
all 351 ausM03953_0005_*.img
Select 'Source image,subj2' - press Done (ie select nothing; tells SPM no more subjects)
Output:
wau* files normalised, interpolated, realigned functionals
wmeanusM03953_0005_0006.img
normalised mean realigned functional
meanusM03953_0005_0006_sn3d.mat
normalisation parameters
spm2.ps
graphical output of normalisation process (appended)
< View normalised images with <DISPLAY> button - Note interpolation
to new voxel sizes and image dimensions >
< View normalisation transformations in spm2.ps (second page) >
< Check registration with <CHECK_REG> button - select
wmeanusM03953_0005_0006.img, EPI.img in SPM Templates dir and
single_sub_T1.img in SPM Canonical dir >
swausM03953_0005_*.{img,hdr,mat}
--------------------------------
Smoothed normalised realigned images
Smoothed with an isotropic Gaussian kernel with FWHM = 8mm (for statistical inference using theory of Gaussian fields)
Type:
Press <SMOOTH> in upper right SPM window
Prompt - smoothing FWHM - enter '8'
Select all wausM03953_0005_*.img and press 'Done'
Output:
swaus* Smoothed normalised images
OPTIONAL step: normalising structural image
-------------------------------------------
Coregister mean EPI and structural T1 images
Type:
Press <COREGISTER> in upper right SPM window
Prompt - No. subjects enter - '1'
Pull down menu - 'Coregister only'
Select target image - 'meanusM03953_0005_0006.img'
Select object image - 'sM03953_0007.img'
Select other images - <select nothing - press done>
< Can check coregistration by pressing <CHECK REG> in bottom of SPM
window and selecting sM*img and meanus*img - Note warping of
structural in z-direction near top of brain due to EPI distortion >
Output: (changes *mat orientation file for structural image)
Normalising structural T1 image with deformation parameters from EPIs.
Press <NORMALIZE> in upper right SPM window
Pull down menu - 'Write normalised only'
Select parameters file for subj1 - 'meanusM03953_0005_0006_sn3d.mat'
Select images to write normalized - 'sM03953_0007.img'
Select parameters file for subj2 - 'Done' (no more)
Output: wsM03953_0007.img
CatStats - Categorical Factorial Analysis
-----------------------------------------
IMPORTANT:
Because interpolated data to middle slice, need to synchronise
the model with the middle slice (rather than the default top slice).
Press <DEFAULTS> in SPM main window
Pull down menu - 'Statistics - FMRI'
Upper tail F prob. threshold - enter default (or enter '0.01')
Number of Bins/TR - enter 24
Sampled bin - change default to '12' (corresponds to middle slice)
NB: If change default F-threshold for storing data to 0.01 in
Defaults menu, more likely to save raw data for some voxels activated
(but bigger data file produced).
4 trial types:
N1 - First presentation of Nonfamous face
N2 - Second presentation of Nonfamous face
F1 - First presentation of Famous face
F2 - Second presentation of Famous face
Change to Stats directory, and type in Matlab window:
'load sots'
=> 1 Matlab cell array variable 'sots'
sots - Stimulus Onset Times (in TRs) for each trial type,
ordered N1,N2,F1,F2
Type in Matlab window:
'sots{1}'
Stimulus Onset Times (in TRs) for N1
Type:
Press <fMRI>
Option - Specify design or data - select 'design'
Prompt - Interscan interval (TR) - enter '2'
Prompt - Scans per session - enter '351'
Option - Specify design in scans or secs - select 'scans'
Pull down menu - Select basis set - 'hrf (with time and dispersion derivatives)'
Option - interactions among trials (Volterra) - 'no'
Prompt - Number of conditions or trials - enter '4'
Prompt - Condition or trial name 1 - enter 'N1'
Prompt - Vector of onsets (N1) - enter 'sots{1}'
Prompt - Duration[s] (events=0) - enter '0'
Option - parametric modulation - press 'none'
Prompt - Condition or trial name 1 - enter 'N2'
Prompt - Vector of onsets (N1) - enter 'sots{2}'
Prompt - Duration[s] (events=0) - enter '0'
Option - parametric modulation - press 'none
Prompt - Condition or trial name 1 - enter 'F1'
Prompt - Vector of onsets (N1) - enter 'sots{3}'
Prompt - Duration[s] (events=0) - enter '0'
Option - parametric modulation - press 'none'
Prompt - Condition or trial name 1 - enter 'F2'
Prompt - Vector of onsets (N1) - enter 'sots{4}'
Prompt - Duration[s] (events=0) - enter '0'
Option - parametric modulation - press 'none'
Prompt - user specified regressors - enter '6'
Prompt - [351] regressor1 - type 'spm_load'
<SPM input is evaluated; spm_load is a matlab function that selects a file>
[select the movement parameters from the realignment, which
are in the RealignUnwarp directory as:
rp_SM03953_0005_0006.txt
a text file with 351 rows and 6 columns for the 6 affine params]
Prompt - name of regressor 1 - press 'return' for default (doesn't matter)
[repeat for next five regressors]
< NOTE THAT ADDING THE MOVEMENT PARAMETERS TO THE DESIGN MATRIX MAKES
THE UNWARPING STEP OF PREPROCESSING SOMEWHAT REDUNDANT, BUT THEY ARE
ADDED HERE MAINLY FOR TEACHING PURPOSES >
Type:
Press <Review design>
Select 'SPM.mat' and press 'Done'
Open Graphics window
Select 'explore fMRI design' in UI Window and select trial-type wish
to view
Design matrix has 351 rows (scans) and (4x3)+6+1 columns (covariates)
Columns are organised:
N1 - canonical HRF (basis function 1)
N1 - temporal derivative (basis function 2)
N1 - dispersion derivative (basis function 3)
N2 - canonical HRF (basis function 1)
...
movement confound1 (user-specified) regressor
...
mean (session) effect
< Note 2nd presentations necessarily later in time on average than 1st
Thus the repetition effect below is confounded with time. This was not
true of the analyses reported in Henson et al (2002) >
Type:
Press <fMRI>
Option - Specify design or data - select 'data'
Select 'SPM.mat' and press 'Done'
Select all 351 'swausM03952_0005_*.img' files and press 'Done'
Option - remove Global effects - press 'none'
Option - High-pass filter - press 'Specify'
Prompt - session cutoff period (secs) - enter '128' (default)
Option - Model intrinsic correlations - 'AR(1)'
<Design matrix, correlations, and selections appear in Graphics window>
Type:
Press <Estimate>
Select 'SPM.mat' and press 'Done'
<pause while parameters estimated - watch progress in Matlab window>
Main Effect of Faces minus Baseline (t-contrast on canonical activations)
-------------------------------------------------------------------------
Type:
Press <Results>
Select 'SPM.mat' file
Press 'Define new contrast'
Enter - contrast = '1 0 0 1 0 0 1 0 0 1' [remaining 0's autofilled]
contrast name = 'Canonical HRF: Faces > Baseline'
press 'done'
Select 'Canonical HRF: Faces > Baseline' (default)
press 'done'
Option - mask with other contrast(s) - press 'no'
Prompt - title for contrast - enter default
Option - p-value adjustment to control - select 'FWE' (family-wise error)
Prompt - threshold {p value} - enter '0.05' (default)
Prompt - & extent threshold {voxels} - enter '0' (default)
< In Graphics window, notice bilateral temporoccipital, left motor and right frontal activations >
Type:
press 'volume'
<describe list of p-values and coordinates>
left mouse on cursor on coordinates at top of table
<red cursor on MIP moves to left fusiform region>
press 'cluster'
<note coordinates should be 39 -72 -18>
press overlays - sections
select in 'wsM03953_0007.img' in Structural directory
(or 'wmeanusM03953_0005_0006.img' in Data directory)
<latter is more valid, since contains same EPI artifacts as data,
but less easier to visualise anatomically>
press overlays - render
select 'render_no_cereb.mat' in Structural directory
< a canonical rendered image with cerebellum artificially removed >
option - style - press 'old'
press 'plot'
Pull down menu - 'Event-related responses'
Prompt - which trials - enter 'N11'
Pull down menu - 'fitted response and adjusted data'
press 'plot'
Pull down menu - 'Event-related responses'
Prompt - which trials - enter 'N1'
Pull down menu - 'fitted response and PSTH'
All Effects of Interest (reduced model removing all confounds)
--------------------------------------------------------------
Type:
Press <Results>
Select 'SPM.mat' file
Click on F-contrasts
Press 'Define new contrast'
Enter - contrast '13:19' in columns for reduced design
[treats columns 13-19 as confounds, leaving 1-12]
contrast name = 'Effects of Real Interest'
press 'done'
Select 'Effects of Real Interest'
press 'done'
Option - mask with other contrast(s) - press 'no'
Prompt - title for contrast - enter default
Option - corrected height threshold - press 'yes'
Prompt - threshold {p value} - enter '0.05' (default)
Click on red MIP cursor and drag near to anterior right fusiform blob
Right click and select 'goto nearest local maxima'
(or press 'Cluster' and select 2nd to top value in table)
<note coordinates should be 45 -48 -27 - right fusiform face area>
press 'plot'
Pull down menu - 'Contrast of parameter estimates'
Prompt - which contrast - enter 'Effects of real interest'
<columns 1,4,7,10 are canonical HRF for N1,N2,F1,F2 - note
suppression effect from F1 to F2 greater than from N1 to N2>
<columns 2,5,8,11 are temporal derivative - note fact that
derivatives close to zero suggests model timing is okay>
press 'plot'
Pull down menu - 'Fitted responses'
Pull down menu - which contrast - 'Effects of Real Interest'
Prompt - fitted or adjusted - 'adjusted'
Pull down menu - plot against - 'Scan or time'
< model in red and data in blue across whole timeseries >
< can change attrib:XLim to [0 100] to see more clearly >
Effects of Fame (F-contrast for two-tailed tests on canonical)
--------------------------------------------------------------
Type:
Press <Results>
Select 'SPM.mat' file
Click on F-contrasts
Press 'Define new contrast'
Enter - contrast type '-1 0 0 -1 0 0 1 0 0 1'
contrast name = 'Canonical HRF: F vs N'
press 'done'
Select 'Canonical HRF: F vs N'
press 'done'
Option - mask with other contrast(s) - press 'NO'
Prompt - title for contrast - enter default
Option - corrected height threshold - press 'no'
Prompt - threshold {p value} - enter '0.001' (default)
Prompt - & extent threshold {voxels} - enter '10'
<Note left mid-temporal, temporal pole and inferior frontal regions>
press 'volume'
left mouse on coordinates <-39 -6 -18> in Table
<red cursor on MIP moves to left temporal pole cluster>
press 'plot'
Pull down menu - 'Contrast of parameter estimates'
Prompt - which contrast - enter 'Effects of REAL interest'
< Note columns 7 and 10 (canonical HRF for F1 and F2) are positive (face naming?)
but columns 1 and 4 (canonical HRF for N1 and N2) are close to zero >
Effects of Repetition (F-contrast on canonical and derivative)
--------------------------------------------------------------
Type:
Press <Results>
Select 'SPM.mat' file
Click on F-contrasts
Press 'Define new contrast'
Enter - contrast type '1 0 0 -1 0 0 1 0 0 -1 0 0
0 1 0 0 -1 0 0 1 0 0 -1 0
0 0 1 0 0 -1 0 0 1 0 0 -1'
contrast name = 'Canonical + Derivatives: 1 vs 2'
press 'done'
Select 'Canonical + Derivative: 1 vs 2'
press 'done'
Option - mask with other contrast(s) - press 'yes'
Select - 'Canonical HRF: Faces > Baseline'
Prompt - uncorrected mask p-value - enter '0.05'
Option - nature of mask - 'inclusive'
Prompt - title for contrast - enter default
Option - corrected height threshold - press 'no'
Prompt - threshold {p value} - enter '0.001' (default)
Prompt - & extent threshold {voxels} - enter '10'
<note combined probability of ~.0005 given orthogonal contrasts>
press <Results-fig> in Graphics window to get new window for table
press 'volume'
left mouse on cursor on coordinates third from top of table
<red cursor on MIP moves to right occipitotemporal region 45 -60 -15>
<most likely in inferior occipital sulcus on this subject, but
in fusiform in group data>
press 'plot'
Pull down menu - 'Contrast of parameter estimates'
Prompt - which contrast - enter 'Canonical + Derivatives: 1 vs 2'
<note positive canonical (and temporal derivative) indicates, given direction
of contrast, that 1st presentations are greater than 2nd presentations
(see Henson et al (2002) for further details)>
press 'plot'
Pull down menu - 'Event-related responses'
Pull down menu - 'which effect?' - 'F1'
Pull down menu - 'plot in terms of...' - 'fitted response and PSTH'
Toggle - hold - turn on in Input window
Pull down menu - 'Event-related responses'
Pull down menu - 'which effect?' - 'F2'
Pull down menu - 'plot in terms of...' - 'fitted response and PSTH'
<note smaller event-related response for F2 than F1>
<effect can also be seen by plotting the event-relate response for
each condition, one after the other, and using the "hold on" option>
< Other interesting contrasts:
a) F-contrast on movement parameters - big edge effects (but not in fusiform)!
b) Trial-specific F-contrasts for N1 and for F1 - note more left
temporofrontal for F1
c) Derivatives only - explain additional variability?>
ParStats - Parametric Differential Analysis
-------------------------------------------
This is an alternative statistical model to the CatStats one above,
that also caters for effects of repetition lag. Repetition is expanded
to a continuous parametric modulation by lag (see below). Fame is also
treated as a parametric modulation of +1 or -1. A third modulation
is the fame x lag interaction (simply the multiplication of the
fame and lag parametric modulations).
Because the models are correlated, the main results are similar.
(Note that it is not exactly the same as the lag analysis in the
Henson et al. 2000, for which lag effects were modelled for
second presentations only).
1 trial types:
Face - Presentation of any type of face (N1, N2, F1 and F2 collapsed)
3 parametric modulations
Lag - Exponentially decaying function of number of intervening items
Fame - Famous or Nonfamous (1/-1)
FamexLag - Interaction of above two
In Stats directory, type in Matlab window:
'load lags'
=> 4 Matlab vectors:
sots - Stimulus Onset Times (in TRs) for all faces
lag - Exponentially decaying function of number of faces intervening
between first and second presentation of a face:
= exp(-l/50)
where l = number of intervening faces (for first presentations,
ie N1 and F1, number = Inf, so lags = 0).
< This particular exponential form was based on the options offered
in SPM99; see SPM99 version of this demo, and Henson et al (2000).
SPM2 no longer offers exponential modulation, so above done by hand >
fam - +1/-1
famlag - product of fam and lag (re-mean-corrected)
Type in Matlab window:
'lag{1}'
=> Lag modulations
Type:
Press <fMRI>
Option - Specify design or data - select 'design'
Prompt - Interscan interval (TR) - enter '2'
Prompt - Scans per session - enter '351'
Option - Specify design in scans or secs - select 'scans'
Pull down menu - Select basis set - 'hrf'
Option - interactions among trials (Volterra) - 'no'
Prompt - Number of conditions or trials - enter '1'
Prompt - Condition or trial name 1 - enter 'Faces'
Prompt - Vector of onsets - enter 'sots'
Prompt - Duration[s] (events=0) - enter '0'
Option - parametric modulation - press 'other'
Prompt - '# parameters' - enter '3'
Prompt - name of parameter 1 - 'Lag'
Option - parameters for lag - enter 'lag'
Option - polynomial order - '1'
Prompt - name of parameter 2 - 'Fame'
Option - parameters for lag - enter 'fam'
Option - polynomial order - '1'
Prompt - name of parameter 3 - 'FameXLag'
Option - parameters for lag - enter 'famlag'
Option - polynomial order - '1'
Prompt - user specified regressors - enter '6'
Prompt - [351] regressor1 - type 'spm_load'
<SPM input is evaluated; spm_load is a matlab function that selects a file>
[select the movement parameters from the realignment, which
are in the RealignUnwarp directory as:
rp_sM03953_0005_0006.txt
a text file with 351 rows and 6 columns for the 6 affine params]
Prompt - name of regressor 1 - press 'return' for default (doesn't matter)
[repeat for next five regressors]
Type:
Press <Review design>
Select 'SPM.mat' and press 'Done'
Open Graphics window
Select 'explore fMRI design' in UI Window and select trial-type wish
to view: note trials ordered:
Main effect of Faces - canonical HRF (basis function 1)
Main effect of Faces - temporal derivative (basis function 2)
Main effect of Faces - dispersion derivative (basis function 3)
Parametric lag modulation - canonical HRF (basis function 1)
...
movement confound1 (user-specified) regressor
...
mean (session) effect
Type:
Press <fMRI>
Option - Specify design or data - select 'data'
Select 'SPM.mat' and press 'Done'
Select all 351 'swausM03952_0005_*.img' files and press 'Done'
Option - remove Global effects - press 'none'
Option - High-pass filter - press 'Specify'
Prompt - session cutoff period (secs) - enter '128' (default)
Option - Model intrinsic correlations - 'AR(1)'
<Design matrix, correlations, and selections appear in Graphics window>
Type:
Press <Estimate>
Select 'SPM.mat' and press 'Done'
<pause while parameters estimated - watch progress in Matlab window>
Type:
Press <Results>
Select 'SPM.mat' file
Click on F-contrasts
Press 'Define new contrast'
Select 'F-contrast'
Enter - contrast = '0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 1'
contrast name = 'Effect of Lag'
press 'done'
Press 'Define new contrast'
Select 't-contrast'
Enter - contrast = '1 0 0
0 1 0
0 0 1'
contrast name = 'Main effect of Faces'
press 'done'
Select 'Effect of Lag'
press 'done'
Option - mask with other contrast(s) - press 'yes'
Select - 'Canonical: Faces > Baseline'
Prompt - uncorrected mask p-value - enter '0.05'
Option - nature of mask - 'inclusive'
Prompt - title for contrast - enter default
Option - corrected height threshold - press 'no'
Prompt - threshold {p value} - enter '0.001' (default)
Prompt - & extent threshold {voxels} - enter '10'
Type:
move cursor to largest right inferior temporal region
press 'cluster'
<select 48 -57 -15> - note coordinates close to CatStats repetition effect>
press 'plot'
Pull down menu - 'Plots of parametric responses'
<As parametric modulator increases (from 0 to 1, "into" the window), the
response decreases. Because the parametric modulator increases as the
number of intervening items decreases, this means that the response increases
as the lag (intervening items) increases - ie the repetition decrease is
transient - wearing off over intervening items. Note that to show this
properly, need to model lag for 2nd presentations only - present
demonstration is simply to show model similar to CatStats model>
</div>
|
