statistic model session wise

[Rageve]

until you'll fix the issue with the already smoothed images from fmriprep when selecting use-aroma (..._preproc_smoothAROMAnonaggr_bold.nii.gz), I continued with the standard processing of the fmriprep unsmoothed images including copying and smoothing. Now I run the statistical model with the standard motion regressors as confounds.
Unfortunately, the first level statistic was concatenated over two session with 2 runs each (4 runs in one model). However, I would like to compute the first level model session wise (the same day/visit), that means two runs for session1 (ses-V0) and two runs for session 2 (ses-V9).
Should I modify the Nodes flags to run this model? Currently the definition is as follows.
"Nodes": [
{
"Level": "Run",
"Name": "run",
"GroupBy": [
"run",
"session",
"subject"
],

[github-actions]

We found the following entries in the FAQ which you may find helpful:

• General: How can I run bidspm only certain files, like just the session 02 for example?
• Statistics: How can I see what the transformation in the BIDS stats model will do to my events.tsv?
• Statistics: How should I structure my data to run my statistical analysis?

Feel free to close this issue if you found an answer in the FAQ. Otherwise, please give us a little time to review.

This is an automated reply, generated by FAQtory

[Remi-Gau]

OK from the top of my head I think this is something I have not yet implemented because no one had asked for it.

Specify contrasts at the session level is not yet present here:

bidspm/src/stats/subject_level/specifyContrasts.m

Lines 60 to 74 in 05d92e9

	switch lower(node.Level)
	case 'run'
	[contrasts, counter] = specifyRunLvlContrasts(contrasts, node, counter, SPM);
	case 'subject'
	if ~checkGroupBy(node)
	continue
	end
	[contrasts, counter] = specifySubLvlContrasts(contrasts, node, counter, SPM);
	end

It vaguely relates to: #809

Just to be clear the way this will most likely be implemented is to have all runs in single model and then compute contrasts session wise.

[Rageve]

I think it can be a problem if lets say session 1 is placebo and session 2 is verum and the sessions (both two runs) are presented in counterbalanced order over subjects. In this case it is easier to define the contrasts session wise and not combined over all runs. Zitat von Remi Gau ***@***.***>:

OK from the top of my head I think this is something I have not yet implemented because no one had asked for it. Specify contrasts at the session level is not yet present here: https://github.com/cpp-lln-lab/bidspm/blob/05d92e9ebcf854e42d2cd12db55495479704239c/src/stats/subject_level/specifyContrasts.m#L60-L74 It vaguely relates to: #809 Just to be clear the way this will most likely be implemented is to have all runs in single model and then compute contrasts session wise. -- Reply to this email directly or view it on GitHub: #1007 (comment) You are receiving this because you authored the thread. Message ID: ***@***.***>

Ralf Veit PhD Institute for Diabetes Research and Metabolic Diseases of the Helmholtz Center Munich at the University of Tübingen Otfried Müller Str. 47 fax: +49-7071-295706 phone: +49-7071-2987703

[Remi-Gau]

Yes I think we agree on that. I may be phrasing it poorly.

Now I just need to find an example dataset to test and work on this. As I do I will most likely ask you for feedback to make sure we are on the same page.

[Remi-Gau]

Note to self.

Check the bids stats model Zoo or the fitlins repo to they have multisession examples with cross session contrasts.

[Rageve]

thanks for all your effort. For example currently, we analyze one study with two sessions and each session has two runs. The participants see in each session pictures with high-calorie sweet, high-calorie savory, low-calorie sweet and low-calorie savory content. We can create contrasts for each regressor and each condition (i.e. only high calorie, only sweet ...) and the difference between conditions (i.e. high- vs low calorie). In this study we have two groups, one with verum and one with placebo. The subjects will be measured before the treatment (ses-V0) and after the treatment (ses-V9). When we combine all sessions in one first level model, one interesting contrast will be the interaction between condition and treatment (let's say high- vs low-calorie in ses-V0 versus ses-V9 for example 1 -1 1 -1 -1 1 -1 1) in the two groups. The question is which contrast do you define in the first level model or in the second level model. The factor group can be only defined in a second level model. The factor session and/or condition (and their combinations) can be defined in the first or the second level model. In another study we use insulin or placebo spray in ses-V0 or in ses-V9 in counter-balanced order (within subject design). In this case it might be better to use a first level model separately for the different sessions and use the contrast images based on the assignment of the two sessions in the second level model. Hopefully things become now more clear. Zitat von Remi Gau ***@***.***>:

Yes I think we agree on that. I may be phrasing it poorly. Now I just need to find an example dataset to test and work on this. As I do I will most likely ask you for feedback to make sure we are on the same page. -- Reply to this email directly or view it on GitHub: #1007 (comment) You are receiving this because you authored the thread. Message ID: ***@***.***>

Ralf Veit PhD Institute for Diabetes Research and Metabolic Diseases of the Helmholtz Center Munich at the University of Tübingen Otfried Müller Str. 47 fax: +49-7071-295706 phone: +49-7071-2987703

[Remi-Gau]

Wow. Thanks that definitely helps get the idea of the analysis you have in mind.

I have checked and there seems to be no example for such model in the model Zoo repo.

bids-standard/model-zoo#44

So this will the perfect occasion to also add one so that people in your situation have some guidance to follow.

Will be a bit busy in the coming days but I will make the time to work on this.

[Remi-Gau]

@Rageve
Finally took some time this weekend to implement session level analysis.

I tested it on a couple of datasets from openneuro to run some kind of test / restest analysis across sessions.

The scripts are there: https://github.com/cpp-lln-lab/bidspm/tree/main/demos/openneuro

And the example for the BIDS stats model are here:

• https://github.com/cpp-lln-lab/bidspm/blob/main/demos/openneuro/models/model-ds000114_desc-testRetestVerbal_smdl.json
• https://github.com/cpp-lln-lab/bidspm/blob/main/demos/openneuro/models/model-ds000224_desc-glasslexical_smdl.json

[Rageve]

Hi Remi, Many thanks for your support. bidspm is increasingly becoming a valuable tool. Zitat von Remi Gau ***@***.***>:

@Rageve Finally took some time this weekend to implement session level analysis. I tested it on a couple of datasets from openneuro to run some kind of test / restest analysis across sessions. The scripts are there: https://github.com/cpp-lln-lab/bidspm/tree/main/demos/openneuro And the example for the BIDS stats model are here: - https://github.com/cpp-lln-lab/bidspm/blob/main/demos/openneuro/models/model-ds000114_desc-testRetestVerbal_smdl.json - https://github.com/cpp-lln-lab/bidspm/blob/main/demos/openneuro/models/model-ds000224_desc-glasslexical_smdl.json -- Reply to this email directly or view it on GitHub: #1007 (comment) You are receiving this because you were mentioned. Message ID: ***@***.***>

Ralf Veit PhD Institute for Diabetes Research and Metabolic Diseases of the Helmholtz Center Munich at the University of Tübingen Otfried Müller Str. 47 fax: +49-7071-295706 phone: +49-7071-2987703

[Remi-Gau]

Setting up those models can be challenging so let me know if you are struggling.

[Remi-Gau]

glad it is useful though