Run PPI in FSL

Functional Connectivity & PPI in FSL

PLEASE NOTE: This work is in progress. Follow these steps at your own risk.

Psychophysiological Interaction (PPI) is a type of fMRI functional connectivity analysis that is specifically useful for looking at the interaction of two regions during a block of time (i.e. a task). This following describes a process for executing a PPI with FSL, based on a block designed experiment.

Essential web references for running a PPI analysis with FSL

http://www.fmrib.ox.ac.uk/Members/joreilly/what-is-ppi/ 
http://www.fmrib.ox.ac.uk/Members/joreilly/how-to-run-a-ppi-analysis-in-feat

References of interest

Friston, KJ, Buechel, C, et al. (1997).
Psychophysiological and Modulatory Interactions in Neuroimaging. Neuroimage, 6, 218-229.
http://dx.doi.org/10.1006/nimg.1997.0291
Gitelman DR, Penny WD, Ashburner J, et al. (2003). 
Modeling regional and psychophysiologic interactions in fMRI: the importance of hemodynamic deconvolution. NeuroImage, 19(1), 200-7.
http://dx.doi.org/10.1016/S1053-8119(03)00058-2
Fair DA, Schlaggar BL, Cohen AL, et al. (2007). 
A method for using blocked and event-related fMRI data to study 'resting state' functional connectivity. NeuroImage, 35(1), 396-405. 
http://dx.doi.org/10.1016/j.neuroimage.2006.11.051
Murphy K, Birn RM, Handwerker DA, et al. (2009). 
The impact of global signal regression on resting state correlations: are anti-correlated networks introduced?. NeuroImage, 44(3), 893-905. 
http://dx.doi.org/10.1016/j.neuroimage.2008.09.036

Overview

The basic PPI model is: H(s)=H(t)+H(s*t) where s is the seed timeseries, t is the task, s*t is the interaction between seed region and task, and H is HRF convolution. Our sequence of an analysis is:

1. fMRI pre-processing of raw data (skull stripping, motion correction, high-pass filtering) and (optionally) removal of nuisance variables (motion, CSF & white matter timeseries)
2. Determining a seed region of interest, based on a functional activation, an anatomical region, or a meta-analytic ROI.
3. Obtaining an ROI mask in each subject’s native functional space and extracting the mean time series for the region.
4. Running the PPI analysis using the FSL General Linear Model with regressors for the task, the seed region timeseries, and the interaction of these two terms.

Steps

Preprocessing Model

Goal: obtain a cleaned up BOLD functional file after brain extraction, filtering, and motion correction.

• Enter raw 4D data as input
• BET (skull stripping)
• MCFlirt (motion correction)
• Highpass filter

ROI selection

Goal: Obtain a mask file for your Region Of Interest. Your ROI might be a voxel (based on peak activation from your fMRI analysis or meta-analysis), a sphere around a voxel, or an anatomical region (from a cortical/subcortical parcellation or an atlas).

Voxel processing

Work in progress. Will take voxel from standard space, get a sphere around voxel, register to functional space for each subject, and get mean timeseries.

MPRAGE processing (optional)

Goal: obtain an anatomical region of interest mask, along with white matter and CSF masks.

• BET MPRAGE
• Run FAST tissue-type segmentation on MPRAGE
• Register MPRAGE to functional space
• Transform CSF, white matter, gray matter masks to functional space
• Extract mean timeseries for CSF, WM voxels in filtered_func_data file output by preprocessing model; combine into one file

Nuisance Model (optional)

Goal: Obtain the residuals file that accounts for nuisance variables such as CSF and white matter signal.

• Enter WM and CSF mean timeseries
• Add motion parameters to model
• Set contrasts to 1; these are not important as we're only interested in the res4d file, not the COPE images

Get seed timeseries

The mean timeseries for your seed ROI will always come from the same functional file you use to do the PPI analysis step.

If you model out nuisance parameters:

• nuisance.feat/stats/res4D.nii.gz is the residual functional data that you will use in the actual analysis.

Important: Res4D has a mean ~0, so you must add (10000 * mask) to create input data. You also must put '–odt float' at the end of the fslmaths command to get proper values.

>> fslmaths mask –mul 10000 –add Res4D Res4D_input –odt float

• Extract seed timeseries from res4d_10000.nii.gz file

If not:

• ppi_preprocessing.feat/filtered_func_data.nii.gz

PPI Model

• Highpass filter at 10000
• Prewhiten
• Set up EVs for:
  1. Psychological Regressor: your task, custom entry with 1 column, convolution Double Gamma (or whatever model you prefer)
  2. Physiological Regressor: your seed timeseries, custom entry with 1 column
  3. Interaction: the intereaction between Psych and Phys, basic shape is interaction, make zero Center for Psych and Mean for Phys.
• Set up contrasts as:

1 0 0

0 1 0

0 0 1

0 0 -1

Which will give copes for psych mean, phys mean, interaction positive mean, and interaciton negative mean.

Group Analysis

• Enter single-subject COPEs into higher level mixed-effect simple OLS or FLAME model.
• Notes:
  • Contrast masking can be used to separate the effects of positive correlations and negative correlations in group differences.
  • A gray matter mask in MNI152 2mm standard space can be used as a pre-threshold mask in order to limit analysis to gray matter voxels.

Useful scripts

PPI scripts contains:

• run_preprocessing_model.sh
• mprage_processing.sh
• run_nuisance_model.sh
• get_seed_ts.sh
• run_ppi_model.sh

• design_preprocessing_template.fsf
• design_nuisance_template.fsf
• design_ppi_template.fsf