Hi all,

I adapted the human conn_batch_humanconnectomeproject script for use with the 7T HCP data (see attached script) and entered the 4 functional runs and movement parameters. Oddly, I am getting severe motion plots, with 1000s of artifacts which obviously is not correct since HCP has high-quality data.

Alfonso or others---do you see anything wrong with the script? 
Has anyone used the 7T data?
Does this have to do with radians or degrees in the Movement_Regressors.txt files?

Thanks!
Sarah

Hi Sarah,

I do not see anything wrong with the script, could you please share one subjects' files associated with the "realignment" and "scrubbing" first-level covariates? (the files might possibly be named "Movement_Regressors.deg.txt" and "art_regression_outliers_rfMRI_REST2_7T_AP_hp2000_clean.txt"?)

Thanks
Alfonso
Originally posted by Sarah Kark:

Dear Alfonso,

Thank you so much for your quick response.

I've attached the files here, compressed.

It looks like the composite (frame-wise displacement values) are normal but perhaps the flagging is related to the signal intensity. 

I see this for many subjects, composite motion is fine but 300+/900 frames for a given run will be flagged as outliers.

Best,
Sarah


Originally posted by Alfonso Nieto-Castanon:

Dear Sarah,

That is very curious, it seems that the issue arises from the global BOLD signal showing unrealistically wild fluctuations. My guess is that this may be caused by a failure of the global BOLD signal estimation procedure, rather than the data actually showing those wild fluctuations. This may happen, for example, when the BOLD timeseries is demeaned (since that affects the global scale of the BOLD signal which is used by SPM to determine an appropriate mask of "in-brain" voxels), but that is strange because, as far as I recall, the HCP data is not demeaned as part of the minimal preprocessing pipeline.  If you could please share the first few scans/acquisitions of the same subject&session that you selected for the "realignment" and "scrubbing" files, that may help better understand the issue here (the entire functional file is probably too large as an attachment, so you may just select for example the first 10 acquisitions for sharing using something like:

vol = spm_vol('rfMRI_REST2_7T_AP_hp2000_clean.nii');
spm_file_merge(vol(1:10), 'fileforsharing.nii');

Best
Alfonso
Originally posted by Sarah Kark:

Absolutely and thanks for your support on this.

I have extracted the files here.

Also attaching frames 10-18 since they look particularly erratic!

Hi Alfonso,

For reference, I got one of the scans for the 3T for this participant to run through with 0 outliers id'd (default thresholds) so maybe this has something to do with the 7T data inputted. Not sure what would be difference there, both are outputted from ICA-FIX pipelines.

Sarah

sorry the attachment appears empty, could you please re-send?
Thanks
Alfonso
Originally posted by Sarah Kark:

Dear Alfonso,

I tested this one, it should open frames 10:18 from REST1_7T_PA.

I wonder why the 3T data work? The input structural for the 3T data (T1w_brain_restore) is skull-stripped, unlike the 7T 1.6mm structural--would that make a difference in art?

Many thanks!
Sarah

Hm, the file seems to fail once downloaded through the forum.
Trying a compressed version here.

Hi Sarah,

I was able to take a look at the data and yes, as suspected, this data has been demeaned (about 50% of voxels show negative "activation" values at any given timepoint), which precludes the use of ART (or any process that requires SPM-based global BOLD signal estimation). I would suggest, if possible, to simply use instead the minimally-preprocessed HCP data.

Alternatively, you could try to "fix" this issue simply by adding some constant value to all of your functional BOLD timeseries, for example using something like:

vol = spm_vol('rfMRI_REST2_7T_AP_hp2000_clean.nii');
for nt = 1:numel(vol)
   data = spm_read_vols(vol(nt));
   mask = data~=0;
   data(mask)=data(mask)+10000;
   spm_write_vol(vol(nt),data);
end

which will add a constant value of 10000 to all of your BOLD timeseries (except out-of-brain voxels); note: please remember to back-up your data (or modify the lines above to act on a copy of your data) before running the above commands since they will overwrite your original functional data files

Hope this helps
Alfonso

Originally posted by Sarah Kark:

Interesting, thank you! 

So to clarify,

The conn_batch_humanconnectomeproject.m sample script (for 3T dataset) references the rfMRI_REST1_LR_hp2000_clean.nii.gz files, which are the ICA-FIX data from HCP (that then need further segmenting, ART ID, and smoothing if applicable in cONN). This script seems to run successfully with the clean 3T data, are those files then not demeaned but the 7T data are demeaned?

If I employ your alternative, would adding 1000 or so to those scans for art to be able to run still result in valid connectivity data if then carried through the rest of the pipeline?

Thank you!
Sarah

Hi Sarah,

Yes, the 3T data in rfMRI_REST1_LR_hp2000_clean.nii files (see left image in attached figure for an example) is not demeaned, as it can be seen by the all-positive values and the level of anatomical detail of the images. In contrast the 7T data (see right image in attached figure showing the first acquisition from the file you shared) is demeaned, with a distribution roughly centered of activation values and without almost any anatomical detail. 

And yes also, adding 10000 to those values (not 1000, which may still keep a considerable number of voxels with average negative BOLD signal values) should work just fine, as those average values are disregarded in all subsequent connectivity-based analyses (but keep in mind that if you want to do other preprocessing steps which, like ART, use the anatomical information available in the functional images -e.g. coregistration, normalization, etc.- those would also fail, as that constant 10000 value that we are adding is not a real substitute for the actual anatomical information in the original functional data)

Hope this helps
Alfonso
Originally posted by Sarah Kark:

Dear Alfonso,

I have taken your suggestion and added a constant to all functional volumes.

As a test, I entered the used the conn batch script to test this on n=4 before my full sample (n=137).

The QA_NORM_functional .jpg looks blown out (brain looks all-white, see attached image), is this because of how CONN scales the image intensity for these plots and adding a constant of 10000 has blown it out of it's normal scale so it just looks white but the underlying data are still intact? 

Other QA metrics look ok.

Please find the attached file containing my .m script and QA plots.

If the distribution of the r-connectivity values looks okay and I, for example, can see the DMN when I seed the PCC, is it safe to assume the functional registration QA plots are just "blown out" because of the constant I added?

Thank you!
Sarah

Dear Sarah,

That looks perfectly fine. The all-white functional images simply reflect that these images lack any sort of anatomical detail (but that is expected, we just added a constant 10000 value for all voxels inside the brain so that is what those plots are showing; cf. in a typical scenario, without demeaning, different voxels would show different intensities because the average BOLD signal depends on tissue class - so it is possible to tell apart gray/white/csf tissue visually).

In any way, that looks all perfectly fine, the functional and anatomical data appears to be properly coregistered and in MNI-space, the FC distributions look appropriately centered and homogeneous across subjects, so I do not see any reason for concern

Hope this helps
Alfonso
Originally posted by Sarah Kark:

Terrific, thank you for your swift help and explanations--always a pleasure!