NITRC CONN : functional connectivity toolbox Forum: help

RE: Lesion mask integration

Mohammad Samim Zaheer — Sun, 11 Oct 2026 15:09:08 GMT

<div>Dear Alfonso,</div>
<div> </div>
<div>Thank you for your detailed clarification. I understand that comparing a single subject to another cohort is not supported. However, if I want to export connectivity data for a specific ROI, such as the IFG, to perform a Crawford-Howell t-test outside of CONN, what steps should I follow?</div>
<div> </div>
<div>Is it possible to obtain this data? If so, could you advise on the appropriate extraction method and recommend which type of analysis I should use?</div>
<div> </div>
<div>For context, I have data from 10 GBM patients and 15 controls. I would like to compare each patient individually to the control group before conducting a group analysis.</div>
<div> </div>
<div>Thanks again for your help!</div>
<div>Samim</div>

RE: Only 1 mean file for 1 subject with 2 sessions

Mary Newsome — Sun, 11 Oct 2026 0:04:27 GMT

Thank you, Alfonso! I see how having one mean would make sense when all of the runs were collected at the same time and not expected to be different. I had an experience similar to that of Shayla's. For an intervention design, I think both of your ideas would help a lot. (The only thought I had about where to offer the option was what if in the rare occasion that the same T1 had to be used for both sessions for some reason (e.g., corrupted or accidentally not acquired), but then one could always copy the file and have it loaded in for each subject?) 
Also, thanks picking up on that error. Thankfully, we did catch that earlier.
All the best,
Mary

RE: Only 1 mean file for 1 subject with 2 sessions

shayla yonce — Sat, 10 Oct 2026 22:32:48 GMT

Hi Alfonso, 
 
I think automaticaly switching to this when the "multiple anatomical scans per subject" option is selected would be a great idea! It took me a long time to figure out that I need to run everything separately. 
 
-Shayla Yonce

multi-dataset resting-state fMRI

Shiva — Sat, 10 Oct 2026 21:50:37 GMT

Hi Alfonso,

I am combining 8 resting state fMRI datasets for a mega-analysis. I'm using CONN25b / SPM25 / MATLAB 2026 through batch scripts on a remote server (without GUI). I greatly appreciate your expert opinion on a few decisions.
My plan is to apply the same preprocessing, denoising, and first-level analysis pipelines separately to each dataset, then pool the participant-level results for a combined second-level analysis.

I am using CONN's default MNI preprocessing pipeline with the following settings:
Preprocessing: Realignment/unwarping, STC (except one dataset), ART (0.9 mm/5 SD), direct normalization (2 mm MNI), smoothing (8 mm FWHM).
Denoising: aCompCor (10 WM + 10 CSF, with 1st derivatives), motion (6 + 1st derivatives), scrubbing, session effects, detrending, and band-pass filtering (0.008–0.09 Hz).

I have some questions.
1. TR=0.9 / multiband / no STC: slice-timing info for dataset 5 is missing. so I am skipping STC for this dataset only. Do you find it reasonable?
Do you think there are other adjustments I should consider? like in motion scrubbing or ART?

2. smoothing kernel size: Dataset 4 has a voxel size of 1×1×2 mm (TR=2), while the others are around 3-3.5 mm. Would you still recommend 8 mm FWHM, or would 6 mm be better in our case?
In your previous reply, you explained that smoothing mainly accounts for differences in functional localization rather than improving SNR, but I'm still concerned that an 8 mm kernel would lead to a greater loss of data in this dataset compared to others.

3. coregistration: Do you recommend using the first functional volume or the mean functional image for coregistration?

4. aCompCor: I am currently using 10 WM + 10 CSF components (with first derivatives). My rational is to remove as much noise as possibel. Do you think this is appropriate, or would you recommend keeping the default 5 + 5?

5. pipeline: And finally :) do you have any recommendations or changes to the preprocessing or denoising pipeline?

Here is the profile of my datasets:

Dataset Scanner (3T) Voxel (mm) TR (s) Slice order Duration (min)
------- ------------ -------------- ------ -------------- --------------
1 Siemens 3×3×3 2.0 Interleaved 10
2 Philips 3.5×3.5×3.5 3.0 Ascending 5
3 Siemens 3.5×3.5×3 2.5 Interleaved 7
4 Siemens 1×1×2 2.0 Multiband 7
5 Philips 3.5×3.5×3.5 0.9 Multiband* 5
6 Siemens 3×3×3 2.0 Interleaved 8
7 Siemens 3×3×3 2.0 Interleaved 10
8 Siemens 3×3×3 2.0 Interleaved 10

Thank you very much for your time and guidance!

Best,
Shiva

RE: Only 1 mean file for 1 subject with 2 sessions

Alfonso Nieto-Castanon — Sat, 10 Oct 2026 16:22:51 GMT

Hi Mary,
The mean functional file is typically (when running the default MNI pipeline and using a single T1 file per subject) created during realignment and it represents the mean of all of the functional data (combined across all runs/sessions). This file is saved in the location of the first run functional data just by convention/simplicity, but it nevertheless does reflect the combined data across all runs, not just across the first run.
If, on the other hand, you are using multiple T1 images in order to accommodate possible anatomical changes in the subjects across runs/sessions, then typically the expectation would be to run preprocessing separately for each run/timepoint. If this was your case, then the error that you report would indicate that preprocessing has been run jointly instead of separately across all runs. Since this is a somewhat common point of confusion I am thinking that perhaps I could add something to the preprocessing pipeline that would allow users to easily run preprocessing separately for each session  (perhaps automaticaly switching to this when the "multiple anatomical scans per subject" option is selected?). In any case, let me know your thoughts and I will update this thread when I have these options ready for testing. 
Also, a quick but likely unrelated note: I noticed that one of your files reported in the error message was named auswau101-2_restingstate.nii, the "auswau" prefix often indicates that data already fully preprocessed (e.g. swau101-2_restingstate.nii, with prefix "swau") will have been used as source of yet a new preprocessing pipeline (hence the new "au" prefixes added). Please double-check and remember to reset your functional and structural tabs to their original states (raw/original data, probably the files "101-2_restingstate.nii" in the example above) before re-running preprocessing. 
Best
Alfonso
Originally posted by Mary Newsome:
<blockquote>
Hi Alfonso and everyone,
We have a Pre-Post design where each subject has 2 sessions. I'm wondering why there is a mean file created for the first session, but not for the second.
During Preprocessing for the second session, I'm getting error messages:  "Mean functional file not found..." and
"/raw_101-2/Resting_State_Analysis_Directory/auswau101-2_restingstate.nii ; run either realignment or ART first, or select the option "first functional volume as reference" -#coregtomean field- if you prefer to use the first functional volume instead of the mean functional volume as reference)"
Thanks for any help!
Best,
Mary
</blockquote>

RE: 2nd-level ROI-to-ROI results restricted to ROIs from separate 1st-level analysis

Alfonso Nieto-Castanon — Sat, 10 Oct 2026 11:00:18 GMT

Hi Tyler,
In the Results GUI for RRC analyses, when selecting the "Group analysis results: individual ROIs" option, you have the option to define, separately for each 1st-level RRC analysis computed, the target ROIs that you would like to be displayed in this GUI. This option appears when right-clicking on the bottom results/Targets table. The default option in there is "targets = source ROIs only", which will use as target ROIs the same as the list of source ROIs defined for your 1st-level analysis, but you can switch that to "targets = selected ROIs only" if you prefer to manually define which specific ROIs may be shown as targets -so that the FDR correction is performed only across the selected ROIs). I may be misinterpreting what you are describing but one possibility would be to go to your 400 Schaefer ROIs results tab, right-clicking on the Targets table at the bottom of the display, and making sure the default "targets = source ROIs only" option is selected there (perhaps at some point you have selected there the DMN-only ROIs?).
Let me know if this was it, or please send me a couple of screenshots otherwise of your two analyses first- and second- level tabs to make sure I am interpreting your question correctly.
Best
Alfonso
Originally posted by Tyler Reekes:
<blockquote>
I am running into an issue with the 2nd-level results GUI for ROI-to-ROI analyses, specifically when using the "Group analysis results: individual ROIs" option. (CONN version 25.b.)
I have a seed ROI of interest that I am analyzing against the full set of 400 Schaefer atlas ROIs. In one 1st-level analysis, the seed is analyzed against all 400 Schaefer ROIs. I then created a separate 1st-level analysis using the same seed but restricted the target ROIs to only those belonging to the DMN.
The problem occurs when viewing the 2nd-level ROI-to-ROI results. The original analysis containing all 400 Schaefer ROIs appears to be restricted to only the ROIs included in the separate DMN-only analysis. In other words, the 400-ROI analysis is displayed as though it were the DMN-restricted analysis. This occurs even though the two analyses were set up separately at the 1st level.
I have checked the corresponding SPM files, and they appear to contain the correct dimensions/matrices (i.e., the expected number of ROIs for each analysis). Thus, it seems that the underlying first/second-level files may be correct, but the results displayed in the CONN GUI are not corresponding to the appropriate analysis. I have also tried deleting the corresponding 1st-level analysis files and rerunning the analyses, but this does not resolve the issue in the Results GUI.
Is there a specific file, or results cache that controls the ROI selection displayed in the 2nd-level ROI-to-ROI single ROI analysis GUI? Or, is there a recommended way to completely reset/rebuild the 2nd-level ROI-to-ROI results so that each analysis uses its own ROI set?
I would be happy to provide screenshots of the setup/results GUI, or examples of the relevant SPM files/matrices if helpful.
</blockquote>

RE: preprocessing data issues

Alfonso Nieto-Castanon — Sat, 10 Oct 2026 10:49:16 GMT

Hi Omar
In general during the functional realignment preprocessing step CONN/SPM perform realignment across multiple runs/sessions in two steps, realigning first each individual run separately to the first image in each run, and then following that by coregistering all runs to the first run. After that process CONN/SPM create a single "mean" image combining all time points (across all runs). When, later in a preprocessing pipeline, CONN uses the outlier detection algorithm to identify potential outlier timepoints, this step also creates a new/revised mean image by combining all time points (also across all runs) but now without the potential-outlier timepoints.
Hope this helps
Alfonso
Originally posted by Omar Maximo:
<blockquote>
Hi,
Here's something I noticed across all versions I have used:
When I processed subjects with 3 timepoints, I noticed that it only created a mean image for the first timepoint and not for the remaining 2 but still created outputs. Is CONN using a mean image from timepoint 1 to correct for motion and coregistration for the other 2 sessions? Why isn't CONN creating a mean image for each timepoint and using them accordingly? or is it creating a mean image across all sessions?
There is also an issue when you have multiple T1w images. When you tell CONN that you have multiple images, then it only preprocess subjects with one timepoint and get errors for the other subjects with multiple errors. Then, when I change the setting to one scan per subject, then the errors go away. But then again, that brings me back to my first point.
</blockquote>

RE: Single vs. session-specific T1 images in a repeated-measures pre/post rs-fMRI study

Alfonso Nieto-Castanon — Sat, 10 Oct 2026 10:41:58 GMT

Dear Liliana,
Yes, you are exactly right and the recommendation applies to pre-post designs as well (using a single structural when there is no expected anatomical differences between sessions helps by reducing the risk of introducing possibly-confounding effects, where differences in preprocessing options between the two sessions are confounded with intervention effects)
Hope this helps
Alfonso
Originally posted by Liliana Kataryne:
<blockquote>
I am analyzing a repeated-measures resting-state fMRI study and I would appreciate your advice regarding the structural images.
We have 22 participants who completed three experimental visits. In each visit, resting-state fMRI and a T1-weighted image were acquired both PRE and POST intervention. Therefore, each participant has six resting-state acquisitions and six corresponding T1-weighted images.
No anatomical changes are expected between these time points, since this is an acute intervention study.
For preprocessing in CONN, would you recommend using one T1-weighted image as the structural reference for all six resting-state sessions of each participant, or using the corresponding T1-weighted image for each session (session-specific structurals)?
I have seen your previous comments suggesting that a single structural image may be sufficient when no structural changes are expected across repeated measurements, but I was unsure whether the same recommendation would apply to this PRE/POST design.
Thank you very much for your help!
Best, Liliana
</blockquote>

RE: Error "Index exceeds the number of array elements. Index must not exceed 2." in preprocessing step

Alfonso Nieto-Castanon — Sat, 10 Oct 2026 10:34:50 GMT

Dear Hyunjun,
Not sure what this may be about, it could potentially indicate a mismatch between the number of ROIs or first-level covariates defined across different subjects (e.g. as if different subjects were run through the Setup step with different settings each time?). If you could please share the script or more details about the procedure that you followed leading to this error that may be helpful.
Best
Alfonso
Originally posted by Hyunjun Park:
<blockquote>
Dear users,
I'm preprocessing my fMRI datasets using CONN connectivity. My datasets consist of two sessions and two different conditions were assigned to each session. I designed the script for it. While uploading images and setting up conditions were okay, I got the error message below during the preprocessing. Would you mind letting me know how I can fix this error, or sharing any previous posts regarding it, please? 
---------------------------- [ Error Message ] ---------------------------- Step 7/7: Updating Denoising variables Index exceeds the number of array elements. Index must not exceed 2.
Error in conn_process (line 1226) for n1=1:N2, CONN_x.Preproc.variables.dimensions{N1+n1}=max(CONN_x.Preproc.variables.dimensions{N1+n1},size(x2.data{n1},2)*ones(1,2)); end ^^^^^^^^^^^ Error in conn_process (line 55) case 'setup', conn_disp(['CONN: RUNNING SETUP STEP']); conn_process([0:4,4.5,5]); ^^^^^^^^^^^^^^^^^^^^^^^^^ Error in conn_batch (line 1361) conn_process Setup; ^^^^^^^^^^^^ Error in GBA_prepro_Batch_v1 (line 29) conn_batch('filename', projectFile, ... ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
------------------------------------------------------------------------ 
</blockquote>

RE: Sliding Window Analysis Storage

Alfonso Nieto-Castanon — Sat, 10 Oct 2026 10:21:43 GMT

Hi,
Often times people use sliding-window analyses in the context of RRC analyses, which considerably reduces the storage requirement as for each temporal window and subject we would just compute a single ROI-to-ROI connectivity matrix. Using sliding-window analyses in the context of SBC analyses is more demanding in terms of storage, as for each temporal window and subject we would compute one separate whole-brain connectivity map for each seed. This is perfectly fine when you have one or a few seeds of interest a priori, but if you have defined many seeds of interest in your SBC analyses I imagine this could easily add up to the large storage demands (>1Tb) that you describe.
Hope this helps
Alfonso
Originally posted by sfeuer45:
<blockquote>
Hello,
I have been trying to perform sliding window analysis using CONN after successfully completing gPPI analysis. With eight subjects, I attempted to complete sliding window analysis, where the window length is 20 TRs (pretty standard according to <a href="https://onlinelibrary.wiley.com/doi/full/10.1002/brb3.1255">Savva et al 2019)</a> and the window shift is 4 TRs (much higher than standard, because I was trying to avoid storage issues and this results in less windows). 
The result was 53 windows each for three sessions, across eight participants. I suppose this is quite a few, but this is less than I would expect if I stuck with standard values for window shift. However, when I ran 1st-level SBC analysis utilizing these windows, it took up 1.1 TB of storage on my machine.
This feels like a lot for such a widely-used method and would prevent me from doing any further analysis. Is it possible that I performed the analysis incorrectly? Or have others had this issue with sliding window analysis taking up a lot of storage? Happy to clarify any other parameters I used.
</blockquote>