Skip to content

glm-practice.md

Latest commit

 

History

History
executable file
·
411 lines (328 loc) · 21 KB

glm-practice.md

File metadata and controls

executable file
·
411 lines (328 loc) · 21 KB
Raw

Introduction to Functional Imaging of the Human Brain

Hands-on session with SPM - statistical data analysis

Małgorzata Wierzba, PhD (m.wierzba@nencki.edu.pl)
Laboratory of Emotions Neurobiology
Center of Excellence for Neural Plasticity and Brain Disorders BRAINCITY
Nencki Institute of Experimental Biology, Polish Academy of Sciences

Setup

  1. Copy the demo-ds005460-with-derivatives directory from the shared materials to your local machine.
    • It is a subset of the original dataset, containing data from three subjects: sub-0312f, sub-1512e, and sub-2811d.
    • However, besides (some of) raw data, it now also contains preprocessed data in the derivatives/fmriprep subdirectory.
.
├── dataset_description.json
├── derivatives
│   └── fmriprep
│       ├── sub-0312f
│       │   └── func
│       │       ├── sub-0312f_task-cet_desc-confounds_timeseries.json
│       │       ├── sub-0312f_task-cet_desc-confounds_timeseries.tsv
│       │       ├── sub-0312f_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.json
│       │       └── sub-0312f_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.nii.gz
│       ├── sub-1512e
│       │   └── func
│       │       ├── sub-1512e_task-cet_desc-confounds_timeseries.json
│       │       ├── sub-1512e_task-cet_desc-confounds_timeseries.tsv
│       │       ├── sub-1512e_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.json
│       │       └── sub-1512e_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.nii.gz
│       └── sub-2811d
│           └── func
│               ├── sub-2811d_task-cet_desc-confounds_timeseries.json
│               ├── sub-2811d_task-cet_desc-confounds_timeseries.tsv
│               ├── sub-2811d_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.json
│               └── sub-2811d_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.nii.gz
├── participants.json
├── participants.tsv
├── sub-0312f
│   └── func
│       └── sub-0312f_task-cet_events.tsv
├── sub-1512e
│   └── func
│       └── sub-1512e_task-cet_events.tsv
├── sub-2811d
│   └── func
│       └── sub-2811d_task-cet_events.tsv
└── task-cet_events.json

  1. Open MATLAB and navigate to the demo-ds005460-with-derivatives directory (the one you just created on your machine).

  2. Start SPM by typing:

spm fmri

Wait, what just happened?!

In the CLIMATE BRAIN dataset, data preprocessing was performed using fMRIPrep rather than SPM. While the general logic is similar, there are some differences between these software packages.

  • One important difference is that they use different naming conventions for the output files. In particular, fMRIPrep follows the so called Brain Imaging Data Structure (BIDS) format.
  • BIDS is a common standard for naming files and directories. For instance:
    • Raw data are located in the root directory of the dataset, while derived data are stored in the derivatives subdirectory. In particular, derivatives/fmriprep contains fMRIPrep output derived from the raw data.
    • Data for each subject are placed in subdirectories named sub-<label>, where string <label> is substituted with the unique identifier of each subject, e.g. sub-0312f, sub-1512e, sub-2811d.
    • anat contains anatomical data, while func contains functional data.
    • ... and so on! (for the full BIDS specification, go here)
  • Such consistent way to organize neuroimaging datasets makes it easy to understand and share data across research groups and institutions.
  • This is why BIDS is now widely adopted by the scientific community, and making your data BIDS-compliant is sometimes even a requirement. For instance, OpenNeuro allows users to host neuroimaging data for free, as long as it is BIDS-formatted.
  1. Navigate to the demo-ds005460-with-derivatives directory. The files that you see in the derivatives/fmriprep are selected fMRIPrep outputs.
    • These files are both sufficient and necessary to proceed with the statistical analysis of fMRI data.
    • All intermediate and auxiliary files have been removed.
  2. All files are organized according to BIDS.
    • Look at the name of this file: sub-1512e_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.nii.gz.
    • Try to guess what it is just by looking at its name and its location in the directory structure.

Let's start!

We will start by performing the subject-level analysis (i.e., first-level analysis) separately for each subject. We will start with sub-1512e and then follow by processing the remaining subjects' data. If we have time, we will proceed to the group-level analysis (i.e., second-level analysis).

But before we start...

Some important things to keep in mind about fMRIPrep:

  • All NIfTI images produced by fMRIPrep are compressed (i.e., .nii.gz format).
  • By default, fMRIPrep does not perform smoothing because some analyses require unsmoothed data.
  • fMRIPrep stores head-motion parameters (translations and rotations) in a large TSV file together with other confounds.

Therefore, there is still some preprocessing left to do before we can move on to the statistical analysis. This will be covered in the following sections, so be patient!

Unpacking the data

IMPORTANT: All NIfTI images produced by fMRIPrep are compressed (i.e., .nii.gz format).

Since SPM can only work with uncompressed NIFTI, we need to unzip them first.

  1. Use the Batch button, then select BasicIO > File Operations > Gunzip Files
  2. Click on the File Set, then Specify.
  3. Navigate to derivatives/fmriprep/sub-1512e. From func select the preprocessed functional image sub-1512e_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.nii.gz. When you're ready, click Done.
  4. Click the green "play" button to run this step!
  5. Navigate to your files to check whether the .nii.gz file was successfully uncompressed into a .nii file.
  6. Save your batch (if needed) and close it.

Smoothing

IMPORTANT: By default, fMRIPrep does not perform smoothing because some analyses require unsmoothed data.

The type of analysis we are planning to perform requires that the data be smoothed.

  1. Use the Smooth button.
  2. Click on the Images to smooth, then navigate to the derivatives/fmriprep/sub-1512e/func subdirectory.
  3. Important: Make sure to select all the preprocessed funtional images! Click on the Frames field, replace 1 with Inf, the press the Enter key.
  4. Now, right-click on the right pane and from the context menu select Select All. You should see 385 files added. If you're ready, click Done.
  5. Change the FWHM field to 6 6 6.
  6. Click the green "play" button to run this step!
  7. Navigate to your files to check if the smoothing was sucessful. You will recognize the resulting output file by the s prefix.
  8. Save your batch (if needed) and close it.

Extracting head motion parameters

IMPORTANT: fMRIPrep stores head-motion parameters (translations and rotations) in a large TSV file together with other confounds.

For our analysis, we will only need six motion parameters, formatted in a way that SPM can use (i.e., a .txt file without a header).

  1. fMRIPrep stores head-motion parameters in the sub-1512e_task-cet_desc-confounds_timeseries.tsv file. Open this file in any text editor and try to locate the six motion parameters. Can you recognize them by name?
  2. Close SPM for now.
  3. In MATLAB, navigate to the root directory of your dataset, e.g. demo-ds005460-with-derivatives.
  4. In MATLAB, click New Script, then copy and paste the following code snippet:
subjects = {'sub-0312f', 'sub-1512e', 'sub-2811d'};

for s = 1:3
	sub = subjects{s};

	input = fullfile('./derivatives/fmriprep', sub, 'func', [sub, '_task-cet_desc-confounds_timeseries.tsv']);
	output = fullfile('./derivatives/fmriprep', sub, "func", [sub, '_motion_parameters.txt']);

	T = readtable(input, "FileType", "text", "Delimiter", "\t");
	cols = ["trans_x", "trans_y", "trans_z", "rot_x", "rot_y", "rot_z"];
	writematrix(T{:,cols}, output, "Delimiter", "tab");
end
  1. Give it a name (e.g., extract_motion_parameters.m) and save it in your user-specific MATLAB directory. You should be able to find this directory in your home directory, it will look like this: ~/Documents/MATLAB.
  2. Run the script!
  3. Navigate to your files to see if the sub-1512e_motion_parameters.txt file was indeed created. It will be in the derivatives/fmriprep/sub-1512e/func subdirectory.
  4. In fact, the script extracted motion parameters for all three subjects. You can verify that the motion parameter files were created in the other subjects’ directories as well.
  5. Clear MATLAB workspace, before going back to SPM. To do that, type this in the MATLAB console:
clear all
  1. Go back to SPM:
spm fmri

First-level analysis

We are almost ready to start the first-level analysis.

But before we start...

Before we can start fMRI data analysis, we need to know the experimental task that was performed in the scanner. In particular:

  • What conditions (experimental, control) was the subject exposed to while performing the task?
  • For each condition, we need to know exactly:
    • When it occurred (onsets)?
    • How long it lasted (durations)?

Here, we will analyse data from the Carbon Emission Task (CET), which involves participants making a series of decisions between receiving a monetary bonus (a self-interest choice) and retiring carbon emission certificates (a climate-friendly choice).

  1. Go to the CLIMATE BRAIN publication and read the description of the CET task. Make sure to take a look at Figure 3. This will help you understand the difference between "CET" (experimental condition) and "dummy" (control condition).
  2. Navigate to your files and in the raw data look for the sub-1512e_task-cet_events.tsv file.
    • The file will be located in the demo-ds005460-with-derivatives/sub-1512e/func subdirectory.
    • Open this file in any text editor.
  3. As you can see from the contents of this file, all trials can generally be divided into CET and dummy.
  4. Note that the onset and duration columns provide information about the onset and duration of each trial, respectively (in seconds).
    • Onsets for the CET trials are:
    5.7918
20.8334
51.1668
66.9674
83.2502
99.5425
130.6259
147.4176
162.9592
193.0510
208.3343
224.8760
256.7088
273.0094
306.5838
322.8761
338.1678
354.7095
387.7924
403.3346
418.8763
451.9591
467.2514
498.3342
514.8764
530.6765
546.4682
578.3099
593.3432
609.8848
640.7183
657.5100
689.8428
705.1434
736.9851
753.7768
    • Onsets for the dummy trials are:
    36.1257
114.5836
178.0009
240.9177
290.0422
371.7596
435.1674
483.2931
561.5098
625.6761
674.3011
721.4434
    • Both CET and dummy trials have a duration of 10 seconds.

Now, we can finally start the first-level analysis!

Model specification

  1. Use the Specify 1st-level button.
  2. Click on Directory.
    • Here, we need to provide the directory where we want SPM to save the first-level results.
    • Navigate to your files. Inside the derivatives directory, create a new results subdirectory, and within this directory, create a subject-specific folder named sub-1512e.
    • The same can be achived by running this command in the MATLAB console:
    mkdir derivatives/results/sub-1512e
    • At the end, your directory structure should look like this: demo-ds005460-with-derivatives/derivatives/results/sub-1512e.
    • Go back to SPM and select this newly created directory.
  3. Specify Units for design as Seconds.
  4. Specify Interscan interval as 2.
    • This refers to the repetition time (TR), which is the time required to acquire a single functional volume.
    • In our case, the TR is 2 seconds. You can verify this information in the sub-1512e_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.json file, by opening it in any text editor.
  5. Click on Data & Design, then New: Subject/Session.
  6. Click on Scans, then Specify, then navigate to the derivatives/fmriprep/sub-1512e/func subdirectory.
  7. Important: Make sure to select all the smoothed funtional images! Click on the Filter field and change it to ss.*. Then click on the Frames field, replace 1 with Inf, then press the Enter key.
  8. Now, right-click on the right pane and from the context menu select Select All. You should see 385 files added. If you're ready, click Done.
  9. Click on Conditions:
    • Here we need to specify any conditions (experimental, control) that the subject was exposed to while performing the task.
    • As discussed above, we have two conditions (CET and dummy). Therefore we click New: Condition twice.
    • For each condition, specify Name, Onsets, and Durations. If in doubt, refer to the "But before we start..." section above.
  10. Click on Multiple regressors, then Specify. Navigate to derivatives/fmriprep/sub-1512e/func subdirectory and select the sub-1512e_motion_parameters.txt file, containing motion parameters.
  11. Click the green "play" button to run this step!
  12. Take a look at the output and review your experimental design! Do you know what different columns correspond to? Is this a block design or an event-related design?
  13. Save your batch, but don't close it this time!

Model estimation

  1. While in the batch editor, select SPM > Stats > Model estimation.
  2. Click on Model estimation, then Select SPM.mat and this time Dependency instead of Specify.
  3. Choose the only available option on the list, fMRI model specification.
  4. Click the green "play" button to run both steps!
    • SPM will issue a warning and ask if you want to overwrite the existing model. Confirm by clicking Continue (we do this only for educational purposes).
  5. Save changes to your batch, but don't close it this time!

Contrast definition

  1. While in the batch editor, select SPM > Stats > Contrast Manager.
  2. Click on Contrast Manager, then Select SPM.mat and this time Dependency instead of Specify.
  3. From the list of available dependencies, choose Model estimation.
  4. Click on Contrast Sessions and then New: T-contrast.
    • Specify Name as CET vs dummy.
    • Specify Weights vector as [1 -1].
  5. Click on Contrast Sessions again and then New: T-contrast
    • Specify Name as dummy vs CET.
    • Specify Weights vector as [-1 1].
  6. Change Delete existing contrasts to Yes.
  7. Click the green "play" button to run all steps!
    • SPM will issue a warning again, proceed by clicking Continue (we do this only for educational purposes).
  8. Save changes to your batch and close it. :)

Viewing and thresholding results

  1. In the main SPM menu, use the Results button and navigate to the SPM model you have just created, i.e. derivatives/results/sub-1512e/SPM.mat.
  2. Select the contrast you'd like to view, e.g. CET vs dummy, then click Done.
  3. Don't apply any masking.
  4. For your p value adjustment choose FWE, then confirm that you want to use the threshold of 0.05.
  5. Repeat steps 1-3, but this time choose none instead of FWE. Also, use the threshold of 0.001 and extent threshold of 0.
  6. What do you see?
    • Are the results the same or different from what you see in the CLIMATE BRAIN publication?
    • Try to navigate to coordinates x = -6, y = 26 and z = 37, just to make sure.
    • Discuss your ideas with the tutor.

Results table

The results table provides cluster-level and peak-level statistics.

  1. Take a closer look at the results table.
    • For each cluster, it provides the cluster size (kE) and p-values, adjusted using different methods.
    • For each cluster, it shows 3 local peaks, that are more than 8 mm apart.
    • For each peak, it provides the T statistics and p-values, adjusted using different methods.
    • Also, for each peak it gives you the [x, y, z] coordinates, i.e. its location in the MNI space.

Atlas labeling

You can also view the results in a graphical form, overlaid on a brain template.

  1. While in SPM's Results window, use the Overlays selector.
  2. Select Sections, then use the Prev path selector to navigate to spm25/canonical subdirectory.
  3. Choose single_subj_T1.nii as a template.
  4. Move the coursor around to view different clusters.
  5. Right-click on the image and from the context menu select: Display > Labels > Neuromorphometrics.
  6. What brain regions do the visible clusters represent?

Exporting the results

We will now learn how to export the results to a NIfTI file. This file can then be opened in any NIfTI viewer (e.g. MRIcron).

  1. While in SPM's Results window, use the Save selector.
  2. Choose Thresholded SPM and provide a meaningful name for your output file, e.g. CET-vs-dummy or dummy-vs-CET.
  3. Navigate to your files, find the file that you just created and try opening it in MRIcron.

Second-level analysis

But before we start...

Before we can proceed to the second level analysis, all the above steps need to be repeated for the remaining subjects.

Any survivors left...?

Wait, what about batch scripting?

In the shared materials, you will find two files: tutorial_batch_glm_job.m, which defines the batch, and tutorial_batch_glm.m, which runs it.

  1. Open the tutorial_batch_glm_job.m file in SPM using the Batch button.
  2. Notice that the batch contains all the steps we performed today!
  3. Notice that the subject-specific information has been removed and replaced with X characters. When we run the script, X characters will be replaced with the following:
    • a preprocessed functional image, i.e. sub-<label>_task-cet_space-MNI152NLin2009cAsym_desc-preproc_bold.nii.gz
    • a directory name for the subject-specific results, i.e. ./derivatives/results/sub-<label>
    • a file containing motion parameters, i.e. sub-<label>_motion_parameters.txt
  4. Now close SPM and open the tutorial_batch_glm.m directly in MATLAB. This script is has been prepared to run over all the three subjects. Since we have already processed the first subject, we only need to run the script for the remaining two subejcts, i.e. sub-0312f and sub-2811d.
  5. Edit the code by changing the list of subjects.
  6. Save the tutorial_batch_glm.m script and then run it, by pressing the green Run button in the MATLAB menu!
  7. Relax and enjoy the show! 🍿📺
  8. When it's done, clear MATLAB workspace, and then go back to SPM.
clear all
spm fmri
  1. Don't forget to look at the results for sub-0312f and sub-2811d! Are they similar or completely different?

Now, we are ready to start the second-level analysis!

Model specification

  1. Use the Specify 2nd-level button.
  2. Click on Directory.
    • Here, we need to provide the directory where we want SPM to save the second-level results.
    • Navigate to your files, in the derivatives/results directory create a new group subdirectory.
    • At the end, your directory structure should look like this demo-ds005460-with-derivatives/derivatives/results/group.
    • Go back to SPM and select this newly created directory.
  3. Click on Design and notice that at this point we can choose from many different types of analysis, such as one-sample t-test, paired t-test or ANOVA. In this case, we select the defaut option, i.e. one-sample t-test.
  4. Click on Scans, navigate to derivatives/results, then to each subject's results subdirectory and select con_0001.nii image for each subject. In total, you should have three files selected. These images represent the CET vs dummy contrasts we've seen earlier!
  5. Save your batch, but don’t close it!

Model estimation

  1. While in the batch editor, select SPM > Stats > Model estimation.
  2. Click on Model estimation, then Select SPM.mat and this time Dependency instead of Specify.
  3. Choose the only available option on the list, Factorial design specification.
  4. Save changes to your batch, but don’t close it this time!

Contrast definition

  1. While in the batch editor, select SPM > Stats > Contrast Manager.
  2. Click on Contrast Manager, then Select SPM.mat and this time Dependency instead of Specify.
  3. From the list of available dependencies, choose Model estimation.
  4. Click on Contrast Sessions and then New: T-contrast.
    • Specify Name as Group: CET vs dummy.
    • Specify Weights vector as 1.
  5. Click the green “play” button to run all steps!
  6. Save changes to your batch and close it.
  7. Use the Results button to view the group results, you already know all the steps! :)

Yes, but...

The group results are somewhat disappointing, as they look nothing like those presented in the CLIMATE BRAIN publication.

  1. Was this something you expected?
  2. Discuss your ideas with the tutor.

That's it!

Congratulations, you have just completed your first fMRI data analysis!