task-fMRI Analysis questions : parcellation, data type, smoothing, concatenation

[Austin Cooper]

Hello there neuroscientist maestros,

I wonder if you can clarify a few things for the task fMRI statistical analysis I want to run with my freshly cleaned data.

(1) Parcellation List

• I am not only interested in cortical regions, but also the thalamus, so I hope you can clarify which data file will be best for this. On this note, I see that in another post that HCP_MMPv1 should be referenced for both ParcellationFileList and ParcellationList variables. Could you please explain this? Will it still make use of, what  I think to be, the final clean file titled: "all_fMRI_data_Atlas_MSMAll_hp0_clean.dtseries.nii"?

(2) ideal data type to reference

• Now, in the vein of this *clean.dtseries.nii file, which I presume to be my highest quality data file, I see that in the example .fsf files that the data file mentioned is "tfMRI_{task_type}_LR.nii.gz". Why is this? Is this equivalent to the fake .nii file that is created since FSL does not yet have the capacity to work with CIFTI based data?

(3) smoothing

• Also, when it comes to smoothing, I have used a 2mm FWHM kernel during surface processing. Does this mean that whatever smoothing I choose within the .fsf file will be in addition to the smoothing that occurred in the surface processing stage?

(4) to concatenate or not to concatenate

• I have processed data for all individual runs (4 movie scans and rest) and also for all files concatenated
• do you recommend running all statistical analyses with the concatenated data, and does this provide more power, or am I mistaken?

Thank you for your continuous and timely help. Your wisdom is cherished.

Austin Cooper

[Glasser, Matt]

You can use this file to include subcortical structures:

 

https://balsa.wustl.edu/file/87B9N

 

ParcellationList is the string you add to the results name to mention that you parcellated it.  ParcellationFileList is the path to the dlabel file. 

 

Don’t worry about the paths in the .fsf files.  You want to use ${fMRIName}_Atlas_MSMAll_hp0_clean.dtseries.nii.

 

You can set original and final smoothing to 2 and no additional smoothing will be performed. 

 

Regarding concatenation, you would need to concatenate your designs if you ran them on concatenated data.  Typically what people do is to run first level analyses on single runs and to incorporate a second level to the analysis (a capability built into the task analysis script) to do the stats across runs.  The HCP-YA data have examples of this. 

 

Matt.

--
You received this message because you are subscribed to the Google Groups "HCP-Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hcp-users+...@humanconnectome.org.
To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/b55d99fa-f387-4f25-9e77-c7160f596b3dn%40humanconnectome.org.

 

---

The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail.

[Austin Cooper]

Thank you Matt!

I'm wondering if I'm completely confused when it comes to the first level analysis, so can you correct me if I'm wrong?

• Is it not the case that if I want to compute a statistical difference map between, let us say, movie and rest, that all of their data must be found in the same fMRI data file? eb
  • or is it still possible to compute this difference even if both their respective data is found in distinct files?
    • Would what I am proposing with the concatenated data be the same as running a first level analysis for both rest and movie data, separately, and then combinging and finding the difference between the two in the second level analysis?

Sorry for my naivety! I've been uncertain about this for some time now so I'd like to know if you can enlighten me with your abundant wisdom! 

Austin

[Glasser, Matt]

This is not typically done.  The reason is that you make a huge number of assumptions about the scanner by comparing from run to run.  When we typically concatenate data across runs (e.g., in a resting state analysis), we demean the data and also variance normalize it by unstructured noise to try to eliminate these effects as much as is possible.  Demeaning would actually remove a huge amount of the interesting differences in such a comparison.  Honestly, if you had a rest run and a movie run and wanted to compare them in this way, you would actually compare the mean fMRI volume of the runs (and hope the scanner didn’t change anything).  One time I played with run means.  I found that a huge effect was actually the brain settling in the skull and the CSF redistributing. 

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/dcbd94dc-6fc8-4d84-8e19-0d49018dbd7cn%40humanconnectome.org.

[Andreas Bartsch]

Hi Matt,

 

>I found that a huge effect was actually the brain settling in the skull and the CSF redistributing.

 

this is interesting.

Do you mean the relative position of the brain within the skull changed, with CSF redistribution, depending on, for example, the amount of ante-/retroflexion?

 

Cheers,

Andreas

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/449DEE3F-BD72-4826-8E0B-DC3EE02F34AB%40wustl.edu.

[Glasser, Matt]

Hi Andreas,

 

As I recall, the effect was most prominent in the ventricles.

Matt.

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/3F95F4D5-949A-4F9C-A9EE-6FAB7BFE69B9%40radvisory.net.

[Andreas Bartsch]

Hi Matt,

 

did you come up with a potential explanation?

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/7C3A3895-3A53-406F-8F9E-711F02A733E7%40wustl.edu.

[Glasser, Matt]

Hi Andreas,

 

I thought it was redistribution of brain tissues from upright to supine positioning.

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/DFF852AD-A6A7-484F-911B-67E27C0C6E9E%40radvisory.net.

[Austin Cooper]

Hi Matt, thanks for this one.

According to your guidance and some of my knowledge on concatenation (primarily from this publication), I'm thinking it would be best to concatenate fMRI runs of the same nature (i.e. concatenate movie scans with movie scans, and checker board visual stimuli scans with themselves). 

I'm not sure, though, if this concatenation should  be done during the processing (i.e. set during the ICAFIXProcessingBatch.sh script, or after. My main concern is signal normalization, I'd like to ensure that the signal intensity between runs of the same type are normalized, and thus on the same "playing field". Is there any normalization that occurs during the HCP preprocessing? If not then I suppose it best that I use single-run FIX on all runs, normalize the data once preprocessing has finished, and then concatenate.

If you have any input I'd really appreciate knowing. 

[Tim Coalson]

MR FIX demeans and also normalizes using the unstructured noise amplitude before concatenating.  I believe it undoes both of these things on the split-out timeseries after the cleanup.  The concatenated runs result in a more stable ICA, and better identification and cleanup of the scanner/motion artifacts it is intended to deal with, so we generally do not recommend single-run ICAFIX.  As for the particular strategy, Day 1 as a single group is what I think we usually use.

Tim

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/6ce0cecc-4ecf-46d4-974f-97a42ca589fan%40humanconnectome.org.

[Glasser, Matt]

We recommend concatenation for cleaning of fMRI data with spatial and temporal ICA.  Resting state fMRI data are analyzed concatenated.  Task fMRI data typically are not.  You can always concatenate for one step of processing and then unconcatinate later (including reverting variance normalization and demeaning).  That is what we do for task fMRI. 

 

Each fMRI run is normalized to a grand mean 10000.  Runs are demeaned before concatenation.  Unstructured noise is equalized across runs using variance normalization. 

 

Actually, the problem with concatenating across runs for task fMRI analyses of the sort you mentioned comes from the demeaning step.

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/6ce0cecc-4ecf-46d4-974f-97a42ca589fan%40humanconnectome.org.

[Glasser, Matt]

We have done different things depending upon how much data we have.  We did all 3T tasks and all 7T retinotopy data for HCP-YA each in its own concatenation.  For HCP Lifespan, we just combined all fMRI data.

 

Matt.

[Austin Cooper]

Thanks a bunch.

Do you think the demeaning step is still problematic when concatenating tasks of the same type?

Also may you clarify the difference between the data found in these folders and verify if I've done things correctly.

I've run ICA-FIX:

    # set list of fMRI on which to run ICA+FIX, separate MR FIX groups with %, use spaces (or @ like dedrift...) to otherwise separate runs

    # the MR FIX groups determine what gets concatenated before doing ICA

    # the groups can be whatever you want, you can make a day 1 group and a day 2 group, or just concatenate everything, etc

    fMRINames="tfMRI_TASK1_AP@tfMRI_TASK2_AP@tfMRI_TASK3_AP@tfMRI_TASK4_AP@rfMRI_REST1_AP@tfMRI_MOVIE1_AP@tfMRI_MOVIE2_AP@tfMRI_MOVIE3_AP@tfMRI_MOVIE4_AP"

    # If you wish to run "multi-run" (concatenated) FIX, specify the names to give the concatenated output files

    # In this case, all the runs included in ${fMRINames} become the input to multi-run FIX

    ConcatNames="all_fMRI_data"  ## Use space (or @) to separate concatenation groups

followed by POST-FIX:

# List of fMRI runs

# If running on output from multi-run FIX, use ConcatName(s) as value for fMRINames (space delimited)

fMRINames="all_fMRI_data"

followed by MSMAII for REST:

# For MR FIX, set fMRINames to empty

fMRINames=""

# the original MR FIX parameter for what to concatenate. List all single runs from one concatenated group separated with @.

mrfixNames="tfMRI_TASK1_AP@tfMRI_TASK2_AP@tfMRI_TASK3_AP@tfMRI_TASK4_AP@rfMRI_REST1_AP@tfMRI_MOVIE1_AP@tfMRI_MOVIE2_AP@tfMRI_MOVIE3_AP@tfMRI_MOVIE4_AP"

# the original MR FIX concatenated name (only one group)

mrfixConcatName="all_fMRI_data"

# @-separated list of runs to use for this new MSMAll run of MR FIX

mrfixNamesToUse="rfMRI_REST1_AP"

# FIX output concat name for this new MSMAll run of MR FIX

OutfMRIName="rest_fMRI_data_after_MSMAll"

followed by MSMAII for TASK (checker board stimuli presentation)::

# For MR FIX, set fMRINames to empty

fMRINames=""

# the original MR FIX parameter for what to concatenate. List all single runs from one concatenated group separated with @.

mrfixNames="tfMRI_TASK1_AP@tfMRI_TASK2_AP@tfMRI_TASK3_AP@tfMRI_TASK4_AP@rfMRI_REST1_AP@tfMRI_MOVIE1_AP@tfMRI_MOVIE2_AP@tfMRI_MOVIE3_AP@tfMRI_MOVIE4_AP"

# the original MR FIX concatenated name (only one group)

mrfixConcatName="all_fMRI_data"

# @-separated list of runs to use for this new MSMAll run of MR FIX

mrfixNamesToUse="tfMRI_TASK1_AP@tfMRI_TASK2_AP@tfMRI_TASK3_AP@tfMRI_TASK4_AP"

# FIX output concat name for this new MSMAll run of MR FIX

OutfMRIName="task_fMRI_data_after_MSMAll"

followed by MSMAII for MOVIE:

# For MR FIX, set fMRINames to empty

fMRINames=""

# the original MR FIX parameter for what to concatenate. List all single runs from one concatenated group separated with @.

mrfixNames="tfMRI_TASK1_AP@tfMRI_TASK2_AP@tfMRI_TASK3_AP@tfMRI_TASK4_AP@rfMRI_REST1_AP@tfMRI_MOVIE1_AP@tfMRI_MOVIE2_AP@tfMRI_MOVIE3_AP@tfMRI_MOVIE4_AP"

# the original MR FIX concatenated name (only one group)

mrfixConcatName="all_fMRI_data"

# @-separated list of runs to use for this new MSMAll run of MR FIX

mrfixNamesToUse="tfMRI_MOVIE1_AP@tfMRI_MOVIE2_AP@tfMRI_MOVIE3_AP@tfMRI_MOVIE4_AP"

# FIX output concat name for this new MSMAll run of MR FIX

OutfMRIName="movie_fMRI_data_after_MSMAll"

Lastly followed by DeDriftAndResample:

MRFixConcatNames="all_fMRI_data"

#SPECIAL: if your data used two (or more) MR FIX runs (which is generally not recommended), specify them like this, with no whitespace before or after the %:

#MRFixConcatNames=(concat12 concat34)

#MRFixNames=(run1 run2%run3 run4)

MRFixNames="tfMRI_TASK1_AP@tfMRI_TASK2_AP@tfMRI_TASK3_AP@tfMRI_TASK4_AP@rfMRI_REST1_AP@tfMRI_MOVIE1_AP@tfMRI_MOVIE2_AP@tfMRI_MOVIE3_AP@tfMRI_MOVIE4_AP"

#fixNames are for if single-run ICA FIX was used (not recommended)

fixNames="NONE"

#dontFixNames are for runs that didn't have any kind of ICA artifact removal run on them (very not recommended)

dontFixNames="NONE"

All of which results in these folders:

I'm questioning if I ran the MSMAII scripts properly; should I only run them for the RS scan? 

Also, is the DeDrift script properly set up/does it properly reference the data? Should I instead be referencing the output of the MSMAII script? 

And lastly, what is the difference between the data found in the folders directly under the concatenated folder? Is it just that they are not concatenated or is it also that they have NOT been demeaned?

Sorry for the array of questions! Any assistance you can provide is appreciated!!

Warm regards,

Austin

[Glasser, Matt]

What to do with the mean is tough if you are doing an activation analysis because it depends both on the brain and on the scanner.  For a correlation analysis it doesn’t matter.

 

What data is all this, HCP data or your data?

Matt.

 

From: Austin Cooper <austin....@gmail.com>
Date: Tuesday, October 10, 2023 at 12:03 PM
To: HCP-Users <hcp-...@humanconnectome.org>
Cc: "Glasser, Matt" <glas...@wustl.edu>
Subject: Re: ThRe: [hcp-users] task-fMRI Analysis questions : parcellation, data type, smoothing, concatenation

 

 

Thanks a bunch.

[Austin Cooper]

This is personal data from a Siemens 3T Prisma scanner. 

I think it's likely best that I normalize the data for each individual task (watching checker board stimuli in block based design) based on the mean activation during baseline blocks (this was recommended by my supervisor Dr. Amir Shmuel). I'm just uncertain whether I should take the task data from the concatenated folder or from the folders specific to each task (basically I'm still uncertain as to the actual difference between this data). 

What are your thoughts?

Warm regards from Montreal,

Austin

[Glasser, Matt]

How much resting state fMRI do you have?  I would not run MSMAll 3x, but rather pick a single best registration.  If you have a lot of resting state, you could do just resting state.  If not, perhaps you should just use all the data for that too.

 

The single runs that come out of MR+FIX and DeDriftAndResample, have their original means and variances back in.  The concatenated data are demeaned, and variance normalization has occurred between runs, but then the data are scaled back into intensity normalized mode (uniformly across all runs).   

[Austin Cooper]

I have 330 volumes of RS, do you reckon that this is sufficient? I figured that this likely isn't enough so I ran MSMAII on all the fMRI runs together. The output is a file named all_fMRI_data_after_MAMAll_Atlas_hp0_clean_vn.dtseries.nii which is found in a folder named all_fMRI_data_after_MAMAll. 

Does the subsequent processing stage, DeDriftAndResamplePipelineBatch.sh, utilize this data at all? You can see that the individual runs and the concatenated folder are all finished processing after the MSMAII stage (as seen in the image below), so I'm wondering how this MSMAII is found, since I haven't explicitly stated it's respective output name in the subsequent processing steps.

The dedrift...batch script variables are set in the following manner:

Overall, based on what I understand from what you've said, it seems most ideal to analyze data from the concatenated folder since everything is normalized and demeaned. Since I have 4 distinct fMRI TASK scans where the stimuli is all very similar (the checker board visual stimulation), I want to perform specific contrasts within this data, so I want it to be on the same "playing field" of intensity and mean signal. Now, this being said, can all these TASK specific contrasts be performed within the concatenated data (where the data also included movie and rest fMRI) or should it first be removed such that I have a TASK specific concatenated sequence? I'm uncertain if having all the other data in the timeseries that is analyzed within FSL will obstruct the statistical comparisons (i.e. will they impact the baseline measures and thus add unwanted signal to the analysis, or are they neglected as long as their respective EV is left at 0?

Thanks for your persistent help and clarification Matt!

Austin

[Glasser, Matt]

It’s probably better to use all the data.

 

No, everything is resampled and recleaned according to the final registration. 

 

Typically, we analyze the individual runs with FSL for a task GLM analysis.  For ICA analyses, we sometimes analyze all the data together.

 

Matt.

 

From: Austin Cooper <austin....@gmail.com>
Reply-To: "hcp-...@humanconnectome.org" <hcp-...@humanconnectome.org>
Date: Tuesday, October 24, 2023 at 2:12 AM
To: HCP-Users <hcp-...@humanconnectome.org>
Cc: "Glasser, Matt" <glas...@wustl.edu>, HCP-Users <hcp-...@humanconnectome.org>, Austin Cooper <austin....@gmail.com>
Subject: Re: ThRe: [hcp-users] task-fMRI Analysis questions : parcellation, data type, smoothing, concatenation

 

 

I have 330 volumes of RS, do you reckon that this is sufficient? I figured that this likely isn't enough so I ran MSMAII on all the fMRI runs together. The output is a file named all_fMRI_data_after_MAMAll_Atlas_hp0_clean_vn.dtseries.nii which is found in a folder named all_fMRI_data_after_MAMAll. 

 

Does the subsequent processing stage, DeDriftAndResamplePipelineBatch.sh, utilize this data at all? You can see that the individual runs and the concatenated folder are all finished processing after the MSMAII stage (as seen in the image below), so I'm wondering how this MSMAII is found, since I haven't explicitly stated it's respective output name in the subsequent processing steps.

 

The dedrift...batch script variables are set in the following manner:

Error! Filename not specified.

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/278ae365-b0df-4f58-a724-233bd01e48bdn%40humanconnectome.org.