fMRI Analysis with FSL
Note
Under construction
FEAT
Launch FSL FEAT GUI
In the terminal, type fsl & to launch FSL, and click on the “FEAT FMRI Analysis” icon to launch the FEAT GUI.
Tip
Navigating to the data folder first
It is recommended that the command is called after navigating to the directory where the data is located, so that loading the image data will be easier.
Tip
Avoiding spaces in file and folder names
Avoiding having spaces in the file and folder names. This is considered a good practice in general when operating in Unix/Linux environment. Consider replacing spaces with underscores or dashes.
(Optional) Load FEAT Configuration
If you have a FEAT configuration file (.fsf) from a previous analysis, you can load it by clicking on the Load button. This will import all the settings from the configuration file, and you’ll be able to modify from there.
Slice timing correction
Click on the “Pre-stats” tab, and select “Slice timing correction” from the drop-down menu. Choose “Use slice timings file”, and select the file that corresponds to the input data. The file should be a text file with one number per line, corresponding to the slice acquisition order.
Important
Check whether your DICOM data contains slice timing information
It is important to check with radiographer whether the DICOM data contains slice timing information. Data collected UHS MRI scanners undergo anonymisation process, which can remove slice timing information before the data is transferred to the user. Typically, slice timing information can be preserved when the DICOM images are burnt onto a DVD from the scanners.
Head motion correction
Under “Pre-stats” tab, select “MCFLIRT” from the drop-down menu of motion correction. This will enable motion correction.
Note
Motion correction is not always necessary
While head motion correction is critical, it is not always a mandatory preprocessing step. For instance, motion correction is not accurate on edge slices, where the cerebrospinal fluid (CSF) flows in and out of the imaging volume. Performing motion correction in this case can actually lead to inaccurate voxel’s slice position. See Fultz et al. 2019 for details.
Coregistration
Click on the “Registration” tab, and under “Main structural image”, select the individual’s T1-weighted brain image.
Tip
Perform brain extraction on the T1-weighted image beforehand
Registration may be affected by large field of view (FOV) of the T1-weighted image and the skull. Therefore, performing brain extraction on the T1-weighted image beforehand can improve the registration accuracy. Even though BET brain extraction can be enabled in FEAT, it’s good idea to provide a brain-extracted T1-weighted image directly if possible.
Normalisation
You’ll need to choose brain atlas for this step. In FSL there are a range of standard brain atlases available, in which we will be using the MNI152 T1 2mm atlas (MNI152_T1_2mm_brain.nii.gz). This is a commonly used standard brain in neuroimaging research.
Tip
How to find the path of FSL installation
In a terminal, type whereis fsl. This will return the path of the FSL installation.
Run FEAT
You are now ready to run FEAT. Press Go to start the analysis. FEAT should normally take a few minutes to complete.
Viewing FEAT Analyses Results
In a file browser (e.g. Windows File Explorer), navigate to the output folder of the FEAT analysis. This folder should end with the suffix .feat. Inside this folder, there should be a file called report.html. Double-clicking on this file, a web browser should open and display the status and results of the FEAT analysis, which is a convenient way to view the results.
While FEAT analysis is still running, it will display “STILL RUNNING” in the main page. When the analysis is complete, open FSLeyes and load the preprocessed fMRI image (filtered_func_data) from the .feat folder.
FSLeyes: a brief intro
FSLeyes is the image viewer released with FSL version ≥ 5.0.10. FSLeyes does not perform any processing or analysis on images.
A complete user guide of FSLeyes can be found here. FSLEyes has the following features:
Orthographic (3 orthogonal slicings) and lightbox (multiple slices) views
Multiple simultaneous views (orthographic and/or lightbox)
Timeseries display (via graphs or movie loops)
Multiple semi-transparent colour-overlays
Simple freehand image editing (mask drawing)
3D rendering
Starting FSLeyes
Basic image viewing
Unlinking Cursors
Viewing multiple images
Viewing Timeseries (4D images)
Viewing Atlases
FSLeyes - 3D mode
Brain Extraction Tool (BET)
Accurate brain extraction is crucial for carrying out structural analysis that involves segmentation. In FSL, it is straightforward to perform brain extraction by running BET, but obtaining accurate results will involve some skill and diligence.
For command-line version, you can type bet to learn about the usage description:
bet <input> <output> [options]
where input and output stand for filenames, and options can be many, or none, of the available extra options.
Varying the fractional intensity threshold parameter (-f)
The -f option in bet controls the fractional intensity threshold that distinguishes brain from non-brain. By default the value is set on 0.5, and when it is smaller, the brain estimate gets larger. In command line, try setting the -f option from 0.2 to 0.8, in turn, to see the effect it has. Save these outputs with different names and load them into FSLeyes.
Troubleshooting Brain Extraction
The section describes some of the more problematic brain extraction cases, which are common with images that have large FOV and/or substantial bias field.
Using the gradient threshold option (-g)
Dealing with large FOV
crop image first to remove the neck
provide an estimate of the centre of the brain
use other BET options that might be more robust
(Optional) FSLUTILS
fslinfo and fslhd
fslstats
fslmaths
fslmaths is a very general image calculator and can be used to perform a variety of manipulations of images.
As an example, here we have extracted two images from a functional dataset, image0 and image1. We’d like to calculate the difference between two consecutive timepoint images, which may be used as part of a quality assessment. We will call the output imdiff. To do so, run the following command:
fslmaths image0 -sub image1 imdiff
and view the output (imdiff).
Now, to calculate this as a percent difference image, run the following command:
fslmaths imdiff -div image0 -mul 100 imdiffpercent
which will first take the difference image (imdiff), divide by the first of the original images (image0), multiply by 100, and output imdiffpercent. View the output, and run fslstats.