One additional comment, you don't have to use 1 and -1. You could, and
I'd actually recommend the following approach:
For each condition, create an onset file with 1s and 0s. Then you will
have N EVs for the psychological variable. For the interaction, in
FSL, you can should multiple each EV by the physiological regressor.
This will create N interaction regressors. Contrasts between them can
be computed with contrasts.
NOTE: In SPM, the physiological regressor is deconvolved prior to
forming the interaction.
In SPM, as I've tested this, splitting the interaction term into N
components for N conditions: (1) improves the model fit, (2) increases
the sensitivity, and (3) decreases false positives. We've submitted a
paper on this to Neuroimage.
Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Postdoctoral Research Fellow, GRECC, Bedford VA
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.
On Thu, Jul 14, 2011 at 6:27 AM, David Soto <[log in to unmask]> wrote:
> Hello, we have created a step by step manual to perform the PPI in FSL.....we have tried it in one data set and we did not get
> significant results, only uncorrected stuff. Therefore we wonder whether we are doing something wrong at some stage.
> Below I am posting our PPI 'manual', which may be also useful for other people - it this turns out to be fine.
> Would anyone who has performed PPI successfully in FSL comment on the following? I appreciate this is a bit of an onerous task
> but one who may help other people to do PPI in FSL.
> Cheers
>
>
> PPI analysis
> 1. Set up onset files (PSY regressor).
> Create a 3 column .txt file containing onsets, durations and weight for each condition of interest. Weight should be 1 or -1, relating to the contrast of interest
>
> 1. Create file for each subject containing the time series for the seed region (PHYS regressor).
> Create spherical seed region (3mm radius should be enough) using fslmaths command described above and mm coordinates.
> Open Misc => featquery.
> o Number of feat directories = no. of subjects
> o Select feat directories: navigate to EPI.feat directory for each subject
> o Stats images of interest: do not select any
> o Input ROI selection: Use mask image defined above
> o Output options
> Do not use atlas
> Select ‘Do not binarise mask’ and ‘Create timeseries plots’
> Set output directory name to something sensible (otherwise it will automatically create a directory named ‘featquery’; this can get confusing after you’ve run a few analyses).
> Click Go. Results will pop up in browser.
> Check the images to ensure that the mask is in the right location in every subject’s space.
>
>
> 1. First level analysis: click on FEAT fMRI analysis.
> Data
> Number of inputs: 1
> Select 4D data: navigate to EPI.feat/filtered_func_data for subject 1
> Output directory: define PPI directory within subject directory
> Total volumes: total number – number deleted previously. (e.g. 420-6=414).
> Number of delete volumes: 0, since the first 6 were deleted during the first FEAT analysis.
> Set TR = 2.2s and high pass filter cutoff = 100s
>
> Pre-stats
> Don’t select anything.
> Set smoothing to ~5mm
>
> Stats
> Use FILM prewhitening
> Full model setup
> o Number of EVs = 3(PSY, PHYS, PPI) plus any other regressors of no interest (e.g. wm_match, wm_nomatch, dummytask etc.)
> o EV1: PSY (onsets of variable of interest; e.g. onsets of priming blocks)
> Basic shape: Custom (3 column format)
> Filename: navigate to location of 3 column .txt file containing onset, duration and weight for each condition.
> Convolution: Double-Gamma HRF
> Apply temporal derivative and temporal filtering
> o EV2: PHYS (timeseries of seed region)
> Basic shape: Custom (1 entry per volume)
> Filename: navigate to the directory containing the output of the featquery analysis performed in step 2 (timeseries data). Select the mean_mask_ts.txt file
> Convolution: none (this is BOLD data and has already been convolved by the brain)
> Switch off temporal derivative and temporal filtering
> o EV3: PPI
> Basic shape: interaction
> Between EVs: select EV1 and EV2
> Make zero: set to Centre for PSY EV (EV1) and Mean for PHYS EV (EV2).
> Do not orthogonalise
> Switch off temporal derivative and temporal filtering
> o Remaining EVs
> Set up as for first level in original FEAT analysis
> Basic shape = custom (3 column); filename = condition onset text file; convolution = Double-Gamma HRF
> Do not orthogonalise
> Apply temporal derivative and temporal filtering
> Click Contrasts and F-tests
> o Set up 3 contrasts, for EV1, EV2 and EV3
> o Define the contrasts using the number keys (e.g. [1 0 0 0 0 0 ] for EV1, [0 1 0 0 0 0 0 ] for EV2 etc.)
>
> Post-stats
> Leave all defaults
>
> Registration
> Click Main structural image
> o Select T1 from individual subject
> o Select Full Search (ensures that images are in same orientation)
> o Select 7 DOF (degrees of freedom; 7 is sufficient as registration is within same brain)
> Standard space: MNI152_T1_2mm
> o Select Full Search
> o Select 12 DOF (coregistering to different brain)
>
>
> Save first level analysis as, e.g. firstlevel_PPI_precuneus.fsf in first subject’s directory
> Replicate to all other subjects using code listed above.
> Run FEAT from command line.
>
>
> 1. Higher level analysis
> Click on FEAT fMRI analysis
> Data
> 1. In top left hand box, select Higher-level analysis; top right-hand box = stats + post-stats
> Select ‘Inputs are lower level FEAT directories’
> Selects the directory for each subject containing all copes
> 1. Number of inputs = number of subjects
> 2. Select feat directories
> Navigate to each subject directory and select the .feat directory for the PPi analysis.
> Select all individually, then click OK.
> 1. Leave ‘use lower level copes’ boxes selected for all copes.
> 2. Specify output directory for the 2nd level, otherwise FSL will create one within the directory of the first subject.
> If the directory name specified was ‘secondlevel’, FSL will save all stats file in another directory named ‘secondlevel.gfeat’.
>
> Stats
> 1. Select Mixed effects: FLAME 1 if there are more than 10 subjects in the analysis. With fewer than 10 subjects, select Mixed effects: FLAME 1+2, which is considerably slower but more accurate with small samples.
> 2. Click Full Model Setup, EVs tab
> Number of main EVs = 1 (EV1 = group)
> Change weight of each EV (subject) to 1
> o Calculates the mean of each contrast across subjects
> View design: should show a vertical line
> 1. In Contrasts and F-tests tab, the only contrast to run is C1, group mean (selected by default)
> 2. Click Done
>
> Post-stats
> 1. Leave all defaults.
> Press Go to run analysis.
>
|