Running PALM using Parcellated CIFTI data with TFCE

[Ashlea Segal]

Dear HCP Community,

I am trying to perform a seed-based group-level resting-state fMRI analysis using PALM. Following the instructions outlined in the Exploring the Human Connectome Course Materials (https://wustl.app.box.com/s/3b98a4mvgcwq0z0fhh4olvcqow1l5z95), I adapted the task-based fMRI code available on HCP Github to run the first and second analysis using parcellated data. Then to run the group level analysis, I used PALM. As with the example provided in the course materials, very few parcels were significant. However, when I ran the analysis at a vertex level incorporating TFCE, the significant regions broadly agree with the regions that have a higher t-statistic (as expected and aligning with the course materials).

Ideally, it would be great to run the entire analysis on dense maps, so I can run PALM with TFCE in the group-level analysis. However, this is incredibly computationally expensive as I have more than 1500 seeds, and 150 subjects - To run the first-level analysis on dense maps takes approx. 1 hour for a single seed per subject (i.e., 225,000 hours in total for all seeds/subjects).

I would like some advice on whether the options I have considered may be appropriate, or advice if someone has come across a similar hurdle and how they overcame it. 

1.     To build my own code to run ROI-based TFCE. It has been implemented in the CONN Toolbox; however, it doesn’t appear to be used widely and is not implemented in PALM, so I am unsure if it is appropriate. 

2.     After running the second-level analysis, convert the COPE.ptseries images into dense maps and run the third level analysis with PALM as normal. 

Thank you for your time,

Ashlea 

[Coalson, Timothy Scott (S&T-Student)]

1500 is a lot of seeds, do they have identified meanings, or is this basically a downsampled dense analysis?  I don't entirely follow your plan, but if you do a separate group-level PALM run for each seed, you will have a substantial multiple comparisons problem that PALM isn't aware of.  Resting state data generally doesn't fit very well into statistics intended for task data.

Tim

---

From: Ashlea Segal <ashlea...@gmail.com>
Sent: Monday, March 29, 2021 6:00 PM
To: HCP-Users <hcp-...@humanconnectome.org>
Subject: [hcp-users] Running PALM using Parcellated CIFTI data with TFCE

 

--
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/c2ddc0c3-07ab-470f-a14a-c33f2b66f728n%40humanconnectome.org.

[Ashlea Segal]

Hi Tim,

Thank you for your reply. The research involves a number of different disorders, and the seeds for each disorder have identified meanings. Comparison will be done within disorder which simplifies the problem. By stating the number of seeds, I simply wanted to emphasize that running the entire analysis pipeline using dense maps was not feasible. Ignoring the issue of multiple comparisons across seeds for the moment, is it appropriate to implement TFCE at a parcel level using the ptseries data, as opposed to vertex level (which is what was detailed course materials)?  

Thank you,

Ashlea 

You received this message because you are subscribed to a topic in the Google Groups "HCP-Users" group.
To unsubscribe from this topic, visit https://groups.google.com/a/humanconnectome.org/d/topic/hcp-users/PTtlfGT-kOg/unsubscribe.
To unsubscribe from this group and all its topics, 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/SN6PR01MB4991374FA1C591A5D2866E9D817D9%40SN6PR01MB4991.prod.exchangelabs.com.

[Harms, Michael]

 

Off the top of my head, I don’t see how it would make sense to run TFCE on parcellated data.

 

Cheers,

-MH

 

-- 

Michael Harms, Ph.D.

-----------------------------------------------------------

Associate Professor of Psychiatry

Washington University School of Medicine

Department of Psychiatry, Box 8134

660 South Euclid Ave.                        Tel: 314-747-6173

St. Louis, MO  63110                          Email: mha...@wustl.edu

 

From: Ashlea Segal <ashlea...@gmail.com>
Reply-To: "hcp-...@humanconnectome.org" <hcp-...@humanconnectome.org>
Date: Tuesday, March 30, 2021 at 12:06 AM
To: "hcp-...@humanconnectome.org" <hcp-...@humanconnectome.org>
Subject: Re: [hcp-users] Running PALM using Parcellated CIFTI data with TFCE

 

* External Email - Caution *

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/CAAQabAcNvYJhXU1j-LXic%2BrTj5eOTD47XW2CDOABkKcLKfEcGg%40mail.gmail.com.

 

---

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.

[Joseph Orr]

You can't run TFCE on parcellated data. The parcellated data doesn't have information about the locations - with dense data you read in the surface. So even if it made sense to do cluster detection in parcellated data (it doesn't), it wouldn't work. Below is an example of running PALM with parcellated data. If you have 1500 seeds and want to look at connectivity, maybe you should do -cifti-correlation. This is the closest thing I can think of to a conn analysis.

# Configuration file for PALM.
# Version alpha115, running in MATLAB 9.5.0.1067069 (R2018b) Update 4.
# 23-Sep-2019 10:48:32

-i cope1_N.ptseries.nii
-transposedata
-i cope2_R.ptseries.nii
-transposedata
-i cope3_N-R.ptseries.nii
-transposedata
-i cope4_R-N.ptseries.nii
-transposedata
-d ../design.mat
-t ../design.con
-corrmod
-corrcon
-logp
-n 5000

-o results_taskcue_MMP
-cmcx
-ise
-saveglm

--

Joseph M. Orr, Ph.D.

Assistant Professor

Department of Psychological and Brain Sciences

Texas A&M Institute for Neuroscience

Texas A&M University

College Station, TX

he/him/his

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/CAAQabAcNvYJhXU1j-LXic%2BrTj5eOTD47XW2CDOABkKcLKfEcGg%40mail.gmail.com.

[Coalson, Timothy Scott (S&T-Student)]

To clarify, TFCE uses information like "these touching elements are all on the same side of zero" to enhance activations despite the presence of noise.  When you already know where the boundaries of your areas are, it doesn't make as much sense to care about "these touching functional areas both activated positively" more than if they weren't touching, so TFCE doesn't seem like a good choice for most parcellated maps.

Parcellated cifti files do actually contain some information related to location, specifically the lists of vertices (or voxels) that are members of the parcel.

Comparing only within the same disorder group seems unusual (if I understood correctly), can you tell us what question you are trying to ask of the data?  Trying to predict severity of the disorder from the data?  Knowing the goal might make it easier to give useful suggestions.

Tim

---

From: Joseph Orr <josep...@tamu.edu>
Sent: Tuesday, March 30, 2021 8:58 AM
To: HCP Users <hcp-...@humanconnectome.org>
Subject: Re: [hcp-users] Running PALM using Parcellated CIFTI data with TFCE

 

To view this discussion on the web visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/CAOxRjYkPO62%3D4U07HaujoJuRBd4UvB%2Bgtgr7%2BMURusxfsAsWww%40mail.gmail.com.