Hi Alejandro,

If your data has already been preprocessed in FSL you should not be repeating the same preprocessing steps in CONN (particularly realignment, as the subject motion parameters will be incorrectly estimated from an already-motion-corrected dataset). I would recommend first importing your realignment .par files (generated during FSL motion-correction step) into CONN as a first-level covariate named 'realignment' (those will contain the correct subject-motion estimates), and then running the rest of your preprocessing pipeline (without step (2), and CONN will pick up those subject-motion estimates necessary for the outlier identification step from your 'realignment' covariate). 

Regarding denoising, that looks perfectly fine but don't forget to also add scrubbing to your list of potential confounder terms as outlier scans often account for a significant amount of noise.

And regarding your first-level analyses I would recommend using bivariate correlations instead of semipartial correlations for graph analyses (as the latter measures are not symmetric)

Hope this helps

Alfonso

Originally posted by Alejandro Garma:

Hi!

Hello dear CONN experts,

I have just started using CONN and have some simple questions to ensure I am processing my data correctly.

I have resting-state fMRI data from 100 subjects that I want to transform into 100 functional connectivity matrices for use with the Brain Connectivity Toolbox. I have partially preprocessed my data in FSL, performing slice timing, motion correction, B0 unwarping, and spatial smoothing. Then, I used ANTs to transform the data into MNI space.

Unfortunately, my images have low SNR and low CNR. Due to this, I want to apply additional preprocessing steps available in CONN, specifically aCompCorr and a 12 DOF covariate GLM with simultaneous band-pass filtering in the denoising step.

The current pipeline I am using in CONN is as follows:

1. Functional: Label current functional files as "original data"
   
2. Functional: Realignment (subject motion estimation and correction)
   
3. Functional: Segmentation (Grey/White/CSF segmentation)
   
4. Functional: Outlier detection (ART-based identification of outlier scans for scrubbing)
   
5. Functional: Label current functional files as "MNI-space data"
   
6. Structural: Segmentation (Grey/White/CSF tissue estimation)
   

Then, I perform only ROI-to-ROI analysis, using semi-partial correlation and HRF weighting. Finally, I compute the graph theory results using the atlas ROIs and a two-sided correlation coefficient with a threshold of 0.3, and export the adjacency matrices.

My first question is whether I am following the correct processing pipeline.

My second question is regarding my adjacency matrix: it does not appear to be symmetrical. Why is that so?

Thanks for your time, and I hope you have an excelent day!

---