Hi Kaitlin

Regarding (1), try increasing in increments of 5, e.g. start with increasing the number of white matter components from 5 to 10 and see if that makes a difference, then try setting both WM and CSF to 10, etc. In general, with this or with any other denoising parameter, try to find the minimal change (e.g. the one that reduces the degrees of freedom of your data as little as possible) that results in the largest improvement in the QC measures (e.g. it makes the histogram distributions look better centered and similar across subjects)

Regarding (2) just the realignment parameters, like above I would try first just including first-order derivatives (i.e. 12 parameters), then try adding quadratic terms (i.e. 24 parameters).

Regarding (3), yes, there is no need to re-run the outlier identification preprocessing step or the Setup step, changing the outlier thresholds after preprocessing has already finished can be done directly in the Setup.Covariates (1st-level) tab selecting the option 'Covariate tools. Compute new/derive first-level covariates. Compute scrubbing', and after that selecting the option 'Covariate tools. Update changes from Setup to Denoising'. See https://www.nitrc.org/forum/message.php?... for details

Regarding (4), that looks fine already, I would not expect further reductions in this filter would have much beneficial effect (as even if you have a very fast TR -e.g. 200ms or lower- most cardiac and respiratory confounds should already fall outside of this filter) but of course there is no harm in trying that out and see if it helps. In general, looking at your data to try to gain some understanding on the possible "causes" of the biases that you are seeing in those QC histograms (e.g. you may look at the global signal residual timeseries plotted in the Denoising tab to see whether there are some obvious artifacts in there that you could address, like the initial scans showing large changes in the BOLD signal, or the residual signal being mostly fluctuations matched to respiratory cycles, etc.) can help select possible changes in the denoising strategy that one could hope would be better tailored to address those issues, but simply trying out different things and seeing what works best is a perfectly valid approach too. 

Hope this helps

Alfonso

Originally posted by Kaitlin Cassady:

Hi Alfonso,

 

Thanks for your message!

 

In terms of specifics...

1) How do I increase the number of aCompCor components? What number would you recommend increasing to?

2) Would you recommend increasing ALL motion parameters (i.e., timeseries, realignment and scrubbing) to 1st or 2nd order derivatives? What about the white matter and CSF?

3) For using a more conservative definition of outliers, this can be done during the ART pre-processing step, correct? Would I choose "Use conservative settings (95th percentiles...)" rather than the intermediate or liberal options? Can I perform this step without having to run the SETUP process all over again?

4) Would adjusting the band-pass filter be a good idea? Right now, I just used the default of 0.008 to 0.09.

 

Sorry for all these specific questions - I've just never run into these denoising issues before! My QC plots have always looked great in the past.

 

Thank you!

Kaitlin