ICA
The following guide describes the Independent Component Analysis (ICA) and Dual Regression, both using FSL tools.
- 1) MELODIC
In terminal, run Melodic command (with the first capital letter). In the MELODIC GUI setup the group ICA:
a) Data
Number of inputs - corresponds to the number of individual NIFTI files to be analyzed.
Select 4D data - here insert the names of the NIFTI files including the full path. WARNING: If you plan to perform averaging of repeated measures, the files with identical task and subject have to directly follow each other.
Delete volumes - leave 0 unless you know what you are doing.
TR (s) - keep default value (which is derived from the first input file)
High pass filter cutoff (s) - set it to 120, unless you have reason for a different cutoff.
b) Pre-Stats
Alternative reference image - leave turned off
Motion correction - leave MCFLIRT
B0 unwarping - leave turned off, unless you want use fieldmaps for registration
Slice timing correction - change to Interleaved (0,2...)
Bet brain extraction - leave turned on
Spatial smoothing FWHM (mm) - change to 8
Intensity normalization - leave turned off
Temporal filtering Perfusion subtraction - leave turned off
Highpass - leave turned on
c) Registration
Expanded structural image - leave turned off
Main structural image - turn on
Select main structural images - here insert the names of the NIFTI files containing brain-stripped (e.g. using BET, the file names should end with '_brain') high resolution T1 images of the corresponding subject (see point 1.a) including the full path.
Linear - leave Normal search, change BBR to 6 DOF
Standard space - leave turned on
Standard space image - keep the default image, when recycling the design, make sure the path corresponds to the installed version of FSL (e.g. 5.0)
Linear - leave Normal search and 12 DOF
Nonlinear - turn on
Warp resolution (mm) - keep the default value (10)
Resampling resolution (mm) - keep the default value (4)
d) Stats
Variance-normalise timecourses - leave turned on
Automatic dimensionality estimation - leave turned on, unless you want to change it to something else
Single-session ICA - change to Multi-session temporal concatenation
e) Post-Stats
Threshold IC maps - leave the default value (0.5)
Background image - leave Mean highres
Output full stats folder - turn on
Timeseries model - leave empty
Timeseries contrasts - leave empty
Session/subjects model - group analysis model can be set here (useless)
Session/subjects contrasts - group analysis contrasts can be set here (useless)
Afterwards you can save your design or click Go to start the analysis. - 2) Create group desing
In terminal, run Glm command.
a) GLM Setup window
Timeseries design - change to Higher-level / non-timeseries design (due to bug, you need to do this also before loading a previously saved design)
# inputs - set to the number of individual NIFTI files, which compose input for the ICA
Wizard - for simple design a wizard can be useful
b) General Linear Model window: EVs tab
Number of main EVs - for an unpaired design, the number equals to the number of groups or factor levels. For a paired design, a number of individual subjects has to be added
Number of additional, voxel-dependent EVs - leave 0
Paste - copy/paste the design matrix to/from a text file
Group - label each independent group with a unique number
EV1, EV2... - here instert the design matrix for each corresponding explanatory variable and input file
View design - check visually your design
Efficiency - check visually your design
c) General Linear Model window: Contrasts & F-tests tab
Contrasts - set the number of contrasts
F-test - set the number of F-tests
Paste - copy/paste the contrast matrix to/from a text file
Title - set the title of a contrast
EV1, EV2... - set the values for corresponding EVs (see point 2.b)
C1, C2... - leave turned on
Save your design and close GLM. - 3) Dual Regression
In terminal, go to the directory, where the MELODIC output *.gica directory is saved. Run the following command:
dual_regression <ica directory>.gica/groupmelodic.ica/melodic_IC 1 <GLM design>.mat <GLM design>.con 5000 <ica directory>.gica.dr `cat <ica directory>.gica/.filelist` - 4) Checking the results
In the terminal, go to the Dual Regression output directory (<ica directory>.gica.dr). Run lsic *tfce_corrp* to list all the contrasts with values above 0.95. Write down the numbers of ICs (WARNING: the files are numbered starting with 0 as opposed to the numbering in the MELODIC report page). Check the components in the MELODIC report page (report.html within <ica directory>.gica and discard those containing artifacts. If no repeated measures have been performed, this is the ICA and DR outcome, otherwise proceed to the step 5. - 5) Averaging repeated measures (Clewett
D., 2011, University of Southern California)
In the terminal, go to the Dual Regression output directory (<ica directory>.gica.dr). Run:
drrepmeas dr_stage2_ic<4-digit number of the chosen IC>.nii.gz dr_stage2_ic<4-digit number of the chosen IC>.nii.gz ...
WARNING: This step assumes, that each two subsequent files are from the same subject! - 6) Randomise
Prepare the averaged GLM design as in step 2. This design will have fewer inputs than the original one. Run the following command for each averaged component (ic<number>_all.nii.gz) from the previous step:
randomise -i ic<number of the IC>_all -o <your output suffix> -d <averaged GLM design>.mat -t <averaged GLM design>.con -T - 7) Checking the results
Follow the instructions from the step 4 to evaluate the outcome.
Back to Process.
upol.cz
