Dear Alfonso and CONN community,

I am a new CONN user (v25b) and would appreciate some guidance on setting up a task-based functional connectivity analysis with continuous behavioural regressors.

**Study Design**
I have two subjects (just to set up the analysis), each with two fMRI sessions:
- Session 1: Participants watch embarrassing videos
- Session 2: Participants watch joyful videos

I have continuous behavioural regressors (0-100 emotional experience ratings, collected from separate normative participants and HRF-convolved) for each session, sampled at TR resolution. I also have low-level audiovisual nuisance regressors (luminance, motion energy, RMS, spectral content) for each session.

**What I Want to Measure**
1. ROI-to-ROI FC during each session separately
2. How FC correlates with the continuous emotional experience rating in each session
3. The difference in FC between sessions
4. The difference in the FC-emotion relationship between sessions

**What I Have Done So Far**
I am using already-preprocessed data (SPM/CAT12) so I skipped CONN's preprocessing step. I have:
- Entered preprocessed functionals and structurals including c1/c2/c3 tissue masks
- Set up two conditions: 'embarrassment' (session 1) and 'joy' (session 2), each covering the full run
- Entered behavioural ratings and audiovisual regressors as 1st level covariates, with zeros for the non-applicable session
- Entered motion parameters (rp_ files) as 1st level covariates
- Run denoising removing: white matter, CSF, motion parameters, and audiovisual nuisance regressors (but NOT the behavioural ratings or condition effects)
- Set up an RRC analysis with AAL3 atlas (170 regions)

**My Questions**

1. In the RRC first level analysis tab, I cannot find where to specify which conditions and covariates to include in the connectivity model. The interface shows the ROI list and timeseries display but no conditions/covariates fields. Where are these specified in v25b?

2. Is RRC the correct analysis type for both my mean FC questions (Q1, Q3) and my continuous behavioural regressor questions (Q2, Q4)? Or do I need separate analysis types (e.g. temporal modulation) for the covariate questions?

3. For the behavioural regressor analysis, should the emotional ratings be included or excluded from denoising? I have currently excluded them from denoising so they remain available for the first level model — is this correct?

4. At the second level, what contrasts should I specify to test:
   a. Mean FC during embarrassment vs joy ([1 -1] contrast across conditions?)
   b. Difference in FC-emotion relationship between sessions?

Thank you very much for your help.

Best wishes

Dear Matthew,

The RRC analyses you have already set up are perfectly good for questions Q1 and Q3. In your second-level analyses, for Q1 you can simply specify the corresponding contrast across conditions (e.g. select one condition at a time to look at FC during joyful or embarrassing videos separately, or select a [1/2 1/2] contrast to look at FC on average across the two conditions), and for Q3 you would use a [-1 1] between-conditions contrast to look at the difference in FC across the two conditions. 

For Q2 and Q4 you simply need to define a new analysis (temporal modulation), in order to evaluate  the changes in FC that temporally covary with the emotional experience rating timeseries. First create a single "emotional experience rating" 1st- level covariate containing the associated rating measure separately for each session (irrespective of whether a session was joyful or emparrassing, that separation will be handled later directly by CONN). Then, when defining your temporal modulation 1st-level analysis simply select that 1st-level covariate as the modulatory term. Then, as before, for Q2 you would simply select one condition at a time (e.g. joyful) to evaluate the extent to which connectivity changes during the joyful videos covaried with the corresponding emotional-experience timeseries, and for Q4 you would use a [-1 1] between-conditions contrast to look at whether those emotional-experience modulated connectivity changes were different between the two types of videos. 

Regarding your specific questions about individual conditions, by default CONN will run each 1st-level analysis separately for each condition defined in your CONN project. If you prefer an analysis to be run only on a specific subset of conditions you can do that directly in the prompt that comes up when you start running your analysis. In there uncheck the option that reads "all conditions" and select instead the individual conditions that you would like this analysis to be run on.

And regarding your questions about denoising, in general there is no need to add the emotional ratings as part of the denoising factors, as those main BOLD signal effects will be directly taken into account and corrected for when computing the temporal modulation analyses, so your denoising strategy already looks perfectly fine. That said, I would consider the need to add some form of scrubbing covariate during denoising to control for the potential presence of outliers (that is generally not a bad idea given that FC analyses can be quite sensible to the presence of outliers). You can do this, for example, simply by running the 'functional outlier identification (ART)' preprocessing step on your data, and then during denoising adding the new 'scrubbing' 1st-level covariate to the list of nuisance regressors. After that you can simply evaluate using the Data Validity/Quality/Sensitivity scores whether adding the scrubbing covariates increase the quality of the denoised data or not (and if necessary also optimize other aspects of denoising for your particular dataset).

Hope this helps

Alfonso 

Originally posted by Matthew Hudson:

Dear Alfonso and CONN community,

I am a new CONN user (v25b) and would appreciate some guidance on setting up a task-based functional connectivity analysis with continuous behavioural regressors.

**Study Design**
I have two subjects (just to set up the analysis), each with two fMRI sessions:
- Session 1: Participants watch embarrassing videos
- Session 2: Participants watch joyful videos

I have continuous behavioural regressors (0-100 emotional experience ratings, collected from separate normative participants and HRF-convolved) for each session, sampled at TR resolution. I also have low-level audiovisual nuisance regressors (luminance, motion energy, RMS, spectral content) for each session.

**What I Want to Measure**
1. ROI-to-ROI FC during each session separately
2. How FC correlates with the continuous emotional experience rating in each session
3. The difference in FC between sessions
4. The difference in the FC-emotion relationship between sessions

**What I Have Done So Far**
I am using already-preprocessed data (SPM/CAT12) so I skipped CONN's preprocessing step. I have:
- Entered preprocessed functionals and structurals including c1/c2/c3 tissue masks
- Set up two conditions: 'embarrassment' (session 1) and 'joy' (session 2), each covering the full run
- Entered behavioural ratings and audiovisual regressors as 1st level covariates, with zeros for the non-applicable session
- Entered motion parameters (rp_ files) as 1st level covariates
- Run denoising removing: white matter, CSF, motion parameters, and audiovisual nuisance regressors (but NOT the behavioural ratings or condition effects)
- Set up an RRC analysis with AAL3 atlas (170 regions)

**My Questions**

1. In the RRC first level analysis tab, I cannot find where to specify which conditions and covariates to include in the connectivity model. The interface shows the ROI list and timeseries display but no conditions/covariates fields. Where are these specified in v25b?

2. Is RRC the correct analysis type for both my mean FC questions (Q1, Q3) and my continuous behavioural regressor questions (Q2, Q4)? Or do I need separate analysis types (e.g. temporal modulation) for the covariate questions?

3. For the behavioural regressor analysis, should the emotional ratings be included or excluded from denoising? I have currently excluded them from denoising so they remain available for the first level model — is this correct?

4. At the second level, what contrasts should I specify to test:
   a. Mean FC during embarrassment vs joy ([1 -1] contrast across conditions?)
   b. Difference in FC-emotion relationship between sessions?

Thank you very much for your help.

Best wishes