So far, all of the analyses you've done have used the GLM (General Linear Model). In a sense, all the GLM is doing is looking for voxels that resemble your task. While this approach is typically the standard type of analysis in neuroimaging research, there are many other types of analysis that greatly extend this approach. We will briefly touch on two of those approaches today: model-free independent components analysis (ICA) and psychophysical interaction (PPI).

FSL has tool that does ICA on your data (MELODIC). Using this tool is very easy since you do not have to specify a model or really do anything beyond giving it your data. The objective is simple: reduce the dimensionality of your data by decomposing it into a set of maximally independent components. One key benefit of running MELODIC is the ability to remove components that you believe to be “noise”. Another benefit stems from the fact that it is model-free. Thus, you make no assumptions regarding the timing or shape of hemodynamic responses, which may prove very useful when the underlying shape/timing of a hemodynamic response is unknown. However, as you may have guessed, the major drawback to this approach is interpretability.

For more information on MELODIC, please see this page.

We have a few data sets that we can work with today. We will use some of the data from FUNC_2, which you may remember K-Space lab. We will also take a look at one of the block runs of FUNC_4. I've created two different versions of the FUNC_4 data: one with an artifact and one without an artifact (we'll look at these later).

1. First you'll want to open MATLAB (and add the appropriate paths). You'll also need to open the Terminal so that you can open FSL, but before you open FSL, you will want to navigate to the classwork directory (this is just to make your life easier).
2. Within the Examples directory under –/Datasets/Class.01/ you will find a directory called MELODIC+PPI_sample. Copy this directory to your FSL directory under the Classwork folder. (remember to use the recursive copy flag cp -r input destination)
   • input: /afs/acpub/project/neurobio381/datasets/Class.01/Examples/MELODIC+PPI_sample
   • destination: /afs/acpub/project/neurobio381/classwork/Fall_2008/Students/YOUR_DIRECTORY/FSL/. (as always, don't forget the “.” at the end!)
3. Now you need to open MELODIC, which will be one of the lower tabs on the FSL GUI.
4. The MELODIC GUI is set up very similarly to the one in FEAT, which you should be very familiar with by now.
5. Data tab: you should load up run1.nii.gz (it should have 50 time points) in the FUNC2_K-Space. The TR for this and the rest of the data you'll be working with today is 2 seconds.
6. Prestats tab: use the standard options that we've used in the past (i.e., motion correction, interleaved slice-timing correction, BET, and smooth to 8mm)
7. Registration tab: don't worry about inputting a main structural for this and the other analyses. You will, however, want to change the DOF for putting the Func into standard space to 12.
8. That's all you have to do. Now click Go.

So, it's really easy to run MELODIC. Now let's take a look at some results that I generated earlier. Look in the MELODIC_Examples folder that was in the directory your copied over earlier.

PPI (psychophysiological interactions) is a method for finding out whether the correlation in activity between two distant brain areas is different in different psychological contexts – in other words whether there is an interaction between the psychological state and the functional coupling between two brain areas.

For more information about PPI and detailed instructions on how to implement it in FSL, please see this page

• Within the MELODIC+PPI_sample directory you copied over to your FSL directory, you should be able to find a directory called PPI
• The PPI directory contains a sample PPI analysis that was run on the first run of FUNC_4. For the “seed region”, I extracted a small sphere from supplementary motor area (SMA).
• This analysis will show how the psychological task (e.g., the flash-squeezing) interacts with activity in SMA across different brain regions. Look at the results in the *.feat directory. The 3rd regressor is our PPI regressor. Notice how a very small region of posterior cingulate interacts with the task and activity in SMA.
• Of course this isn't the best example of the use of PPI. Can you think of better examples? Or how you might apply this technique to your own experiment?