FEAT 1 Practical

Practical Overview

This tutorial leads you through some standard single-subject analyses with FEAT. There may be moments when you are waiting for programs to run; during those times take a look at the FEAT manual (in particular look at the Detail section). We also suggest that you do read it carefully after the course, before using FEAT.

Find the practical data

cd ~/fsl_course_data/fmri

Example real audio-visual dataset

cd av

The dataset fmri.nii.gz is from an audio-visual experiment. Auditory stimulation was applied as an alternating "boxcar" with 45s-on-45s-off and visual stimulation was applied as an alternating "boxcar" with 30s-on-30s-off. The TR is 3 seconds. We have extracted just 45 timepoints and 5 slices from the original 4D data, to speed up the analysis.

Type Feat &
(Feat_gui & on Mac)
Press Select 4D data and select fmri.nii.gz (don't just type "fmri.nii.gz" in the file select popup or you probably won't end up setting the full pathname; use the file-select icon on the right to select the input data). FEAT warns you that, because the dataset only has a few slices in it, some default settings, such as whether brain-extraction is run on the FMRI data, have been changed - don't worry about this for now.
FEAT now knows how many time points (volumes) you have (45 in this cut-down dataset). If your TR (time between 3D volumes) is not 3s, change it (it is right for this dataset).
The High pass filter cutoff is preset to 100secs. This is chosen to remove the worst of the low frequency trends, and is also long enough to avoid removing the signal of interest. In general you need to ensure that this is not set lower than your maximum stimulation period (in this case 45*2=90secs).
Press the Pre-stats tab to look at the preprocessing steps. Make sure BET brain extraction is deselected, as we only have a few slices of data. All the other default pre-processing steps are fine for this dataset.
Select the Stats tab and press Full model setup to setup the GLM details.
Change the Number of EVs to 2 (we have two conditions to model separately - audio and visual).
Setup EV1 (the visual stimulation timing): set Off to 30, On to 30 and Phase to 30. This describes a square wave of total period 60s. It starts with an ON period, hence the phase setting, which shifts the waveform forward in time.
Setup EV2 (the auditory stimulation timing): set Off to 45, On to 45 and Phase to 45. This describes a square wave of total period 90s.
Now setup the Contrasts. Set the Number of contrasts to 2, set the first (OC1) to [1 0] and the second (OC2) to [0 1]. Thus the first output colour overlay image produced will show visual activation as only EV1 is used, and the second will show only auditory activation. Note that you can give each contrast a title. So, entitle the contrasts "Visual" and "Auditory" accordingly.
Now setup the F-tests. Set the Number of F-tests to 1. Select both contrasts. Thus the third output colour overlay image produced will show where either visual or auditory activation (or both) occur; i.e. it will show both on a single image.
Press View design. Make sure you understand the resulting design matrix. Time goes down the page, with every 10 TRs ticked off on the left. The red bar shows the width of the highpass filter (the amplitude of any signal much longer than it will get reduced). There are 4 columns in the design (1 and 3 are the ones you just set up and 2 and 4 are temporal derivatives of 1 and 3 - this will be explained in the lectures). The contrasts appear at the bottom of the image, with the F-test to the right of the contrasts. Note, that you can make the design matrix display disappear just by clicking on it once. For now, leave the design matrix display up (Press View design again if necessary).
Now, return to the EVs tab and FOR BOTH EVs change Convolution from Gamma to None and deselect Apply temporal filtering and deselect Add temporal derivative.
Press View design and note how you are now viewing the underlying conditions that were entered.
Restore the original settings one step at a time FOR BOTH EVs, viewing the design after each change (change Convolution to Gamma, then select Apply temporal filtering, then select Add temporal derivative). Make sure you understand the changes in the design matrix and their importance.
When you are happy that the design matrix is restored, press Done.
Look at the Post-stats section - the defaults are fine; cluster-based thresholding will be carried out.
Select the Registration tab. By default FEAT will register the middle-timepoint FMRI image (saved as example_func in the .feat output directory) directly to the standard space template. We recommend in general turning on the Main structural image option so that the lowres FMRI image is first registered to a brain-extracted highres structural image from the same subject; this highres is then registered to the standard space template, and then the two registrations are combined to give an example_func2standard.mat transform which can be used later to resample the FMRI stats into standard space. If, as here, the 4D data only contains a few slices, then even before registration to the highres image, it is a good idea to register example_func to a whole-head EPI image which contains basically the same slices as the 4D data.
Select Initial structural image, set the file to epiwholehead.nii.gz and set the DOF to 3; this is translation only, as we only have a few slices in the timeseries data which correspond to certain slices in the epiwholehead image, and we do not want to allow the image to be rotated. Set the Main structural image file to structural_brain.nii.gz with 7 DOF (note that we have already run BET on this, and have reduced the resolution of the structural image to save time on the highres to standard registration). Leave Standard image turned on with MNI152_T1_2mm_brain.nii.gz selected and set the DOF to 12. Instead of linear (12 DOF), the more accurate nonlinear registration is normally recommended to MNI space by selecting the "nonlinear" option, but we use linear here as it is faster for this practical.
You are now ready to run FEAT. Press Go. A web browser should appear, and as FEAT completes the different stages of processing, you will see messages appear in the Log section. Whether the web browser (and indeed the FEAT GUI) is left displayed or is closed, FEAT will continue to run in the background. For now, leave the web browser open so that you can monitor FEAT's progress. FEAT will take 2-5 minutes to complete.
Whilst FEAT is running, run fslview to have a quick look at the different images mentioned above: start with
fslview epiwholehead.nii.gz
then exit this and view structural_brain.nii.gz and finally exit this and view fmri.nii.gz. Note that when viewing the 4D image you can see the image time series as a movie by pressing on the movie icon, and you can also see time series plots by pressing Tools->Timeseries.
Again; whilst FEAT is still running, we will now use FSLView to create a hand-drawn mask in standard space that will be used later to find out about activation statistics from within the mask.
- Load the standard space template image $FSLDIR/data/standard/MNI152_T1_2mm into FSLView ($FSLDIR is an environment variable indicating the directory in which FSL is installed, you can type echo $FSLDIR to see what this is set to). Inside FSLView you can use the "File->Open Standard" menu option to find these standard space images quickly.
- Create an editable mask image with the same dimensions as the standard space template, by clicking on MNI152_T1_2mm in the image list (at the bottom) and then pressing File -> Create Mask.
- Move the cursor around until you are in the middle of the visual cortex around about slice number 33 (see hint if you are unsure where this should be). Turn on "drawing mode" by pressing the button with the pencil icon. Increase the pencil drawing width by increasing the value in the top-right scrollbox, for example, to 10. Click and drag over a region of the brain roughly corresponding to the visual cortex, in the axial view. Whenever you let go of the mouse button, the region you have been drawing will get written into the mask overlay image. You can move up to the next slice using the up arrow next to the Z co-ordinate box, and do that one as well.
- When you have drawn the mask in a few slices, save the mask image to file, by making sure the mask is selected in the image list, and then pressing File -> Save As. Choose a filename, for example, vismask.nii.gz in your ~/fsl_course_data/fmri/av directory. Press OK to save.
When FEAT has finished, look carefully at the various sections of the web page report, including motion correction plots in the Pre-stats section, the colour-rendered activation images and timeseries plots in the Post-stats section, and the Registration results. Note that if you click on the activation images you get a table of cluster co-ordinates.

Calculate the high-pass filter cutoff from the design

Using the command line utility cutoffcalc you can obtain estimates of a 'safe' high-pass filter cutoff value (in seconds) to be used in FEAT.

Within the .feat directory your design matrix is saved as a simple text file 'design.mat'. You can create these and corresponding contrast matrices using either the Feat GUI or the Glm GUI. Either one will save the design matrices and statistical model options. In this case we will try the Glm GUI. Open this GUI (type Glm). In the General Linear Model window create a simple design matrix, e.g. a 30s on/off blocked design. In the GLM Setup window set the number of timepoints to 45. Now view the design (press View design in the General Linear Model window).
Set High pass filter cutoff (s) to 0 in the GLM Setup window (to turn off the filter - but note that if the cutoff resets to 1.0 then just ignore this as it will still do no filtering). Then save this GLM design to disk, calling it "newdesign". Run the cutoffcalc command on newdesign.mat with the following:
cutoffcalc -i newdesign.mat -t 0.9 --tr=3.0
The output is the High pass filter cutoff value (in seconds) that can be entered in the FEAT Data tab. Here we have set the TR explicitly to 3.0 seconds and the acceptable retained variance (of the filtered EV) to 90%. The retained variance value of 90% is rather lenient here due to the small number of timepoints, with values between 95% and 99% being used on more typical datasets.
Note that if you are interested in the effect of the HPF (highpass filter) then you can go back and forth, changing the HPF cutoff value and inspecting the difference this makes on the design matrix using the View design button.

Featquery

Featquery allows you to calculate certain data statistics, either at a voxel of interest, or averaged over a region of interest using a mask. We will use the standard-space mask which we created earlier.

Start Featquery by typing Featquery &
(Featquery_gui & on Mac)
Select the fmri.feat directory created by your first analysis on the audio-visual dataset.
Featquery automatically reads the FEAT directory and gives you the appropriate options as to which statistics you can choose to investigate. Select the following statistics for both visual and auditory contrasts (i.e. numbers 1 and 2):
- stats/cope (unthresholded contrast of parameter estimate)
- stats/zstat (unthresholded z statistics)
- stats/tstat (unthresholded t statistics)
- thresh_zstat (thresholded z statistics)
Select the Convert PE/COPE values to % option
Select the mask that you created earlier as the Mask Image (note - Featquery can take either a standard-space mask OR a lowres one in the original dataspace OR a mask in the space of the structural image)
Select the Do not binarise mask (allow weighting) option
In the Output options section, select an atlas, for example the Harvard-Oxford Cortical Structural Atlas. The local maxima voxels reported by Featquery will be related to structures in the selected atlas.
Press Go and a web browser showing the estimated statistics should popup shortly (possibly a minute or two if things are busy). If your standard space visual mask is any good, the mean value for the visual zstat image should be a lot larger than for auditory!
Now select an atlas in the Input ROI selection panel of FEATQuery, for example, the Juelich Histological Atlas. Next, press the Select label button and select a relevant structure, for example GM Primary auditory cortex TE1.0 R. Featquery will automatically transform the relevant mask from the FSL atlases database into the space of your data and use that to create the output statistics. Press Go.

High-resolution Single-Session Overlays

You might want to see your low resolution data overlaid onto your high resolution image. Use Renderhighres (on mac Renderhighres_gui) and select your FEAT output directory. Select the Space to upsample to: standard option and also select the Background image: main structural option. When the processing has finished you can find the hr/rendered_*.nii.gz pictures in the FEAT directory and view them with fslview. Renderhighres takes a minute or two to run, as the images get resampled into high resolution using the "accurate", but slow, sinc interpolation method; whilst you are waiting you should start a new terminal and move on to the next section.

Custom Waveforms

We can load more complicated EVs into FEAT using Custom Waveforms. Here, a simulated dataset has been generated with some more complicated event-related conditions to model.

cd ~/fsl_course_data/fmri/art

Feat &

(Feat_gui & on Mac)

Select the data artdata.nii.gz. This dataset has already been preprocessed, so change Full analysis to Stats + Post-stats.
Go to Full model setup and set 3 EVs (there are three different things going on in the data with their own timings).
Condition 1 is a boxcar design so set EV1 to be Off=20 On=20 Phase=10.
Condition 2 is a jittered event-related design with inter-stimulus-interval (ISI) > 20s and mean(ISI)=25s; so set EV2's Basic shape to custom (3 column format); select the filename as jittered_isi_custom_file.dat. In the terminal have a quick look at the custom file with the cat command. The 3 columns are explained in the FEAT user guide.
Condition 3 is randomised event-related with ISI>3s and mean(ISI)=6s; set EV3's shape to custom (3 column format); select the filename as rand_isi_custom_file.dat.
Set 3 contrasts [1 0 0] [0 1 0] and [0 0 1] and an f-test [1 1 1]. Press Done. Turn off registration (select the Registration tab and deselect all registrations), press Go and wait for exciting results! On the FEAT report page look at the time series plots of data vs model.
If you have time, have a detailed look at the timeseries plots in the Post-stats section of the webpage report. For example, click on the zstat1 timeseries plot, to see more information about the fitting related to contrast 1. Find the Peristimulus plot for EV1; this shows the data points for all repeats of the condition described in EV1, collapsed down to one "average" version of that stimulus. In this case this means all repeats of the basic block-design on-off shape. Now go back and click on the zstat2 timeseries plot, to see more information about the fitting related to contrast 2. Find the Peristimulus plot for EV3; this shows the data points for all repeats of the events described in EV3, collapsed down to one "average" version of that stimulus. In this case this means all repeats of the brief event. Now start FSLView, and load in the example_func image in the FEAT output directory. FSLView goes into "FEAT mode" and shows you the data timeseries as well. If you now load in thresh_zstat1 and, from the menu in the timeseries widget, choose COPE1, as you click around, you will see the fitted model against the data (the data is shown in red and the fitted COPE1 model is shown in green).

This is the end of FEAT session 1. But don't forget to take a look at your Renderhighres results.