Dear Alfonso, dear CONN users,

I would like to use the rshrf toolbox to deconvolve the haemodynamic response function: https://github.com/compneuro-da/rsHRF

Unfortunately, it requires a bandpass filter to work. I have thus two options to apply in an analysis pipeline with CONN:

* Either after preprocessing, thus CONN would receive as input functional files already bandpass filtered images (and hrf deconvolved). Can this impact the CompCor denoising or is this OK?

* Either after CONN denoising, but then I have no idea how I can make CONN use the denoised+deconvolved files instead of just denoised, as CONN provides no way to specify custom additional post-processing steps (feature request? ;-) ). However, even if there was a way to do it after denoising, I think the denoising can remove important information to estimate the hrf, so this is probably not the best avenue (but that's only my thoughts?).

Does someone have any suggestion of what would be the best avenue, or the most correct technically?

Thank you very much in advance!
Best regards,
Stephen

Dear Stephen,

That is a very good question. My impression is that the best place to apply the hrf-estimation and deconvolution step would be after the standard denoising step (including aCompCor, scrubbing, and band-pass filtering). The rsHRF implementation code seems to similarly apply a number of denoising/regression steps before actually estimating the hrf, which I imagine are aimed at reducing possible biases in the hrf estimation step due to residual motion and physiological artifacts. 

In practical terms, I would probably suggest to process your data normally in Setup & Denoising, and have CONN export the denoised nifti files, then use the rsHRF toolbox to estimate and deconvolve the hrf on these nifti files, and then re-import the output files into CONN in Setup (but this time skipping CONN denoising steps entirely). I will see if I can provide some simple functionality to extend CONN's default denoising procedures in the next release to simplify this procedure, but in the meantime I believe this should work just fine. Please let me know your thoughts/comments and/or if you run into any issues 

Hope this helps
Alfonso

Originally posted by Stephen L.:

Dear Alfonso,

Thank you very much for your very helpful reply!

Yes indeed the rsHRF does its own denoising steps to possibly reduce biases for hrf estimation. These denoising steps are deactivable but doing so will result in biased estimation according to the authors (I tried and it was bad ;-) ).

Ok I will reimport the functionals as you suggest, but how is it possible to skip the denoising step? Simply by not clicking on "Done"? But then will the new functionals be used instead of the one denoised by CONN? Sorry to ask but I'm not sure how to disable CONN's denoising, I never did that before :-p

Thank you again Alfonso!
Best regards and cheers from Belgium!
Stephen

Dear Stephen,

If you want to skip all denoising procedures, in the Denoising tab (and before clicking on the 'Done' button) remove everything listed in the 'Confounds' list (to skip the regression of confounding effects), set the band-pass filter to [0 inf] (to skip filtering), and select 'No detrending' and 'No despiking' (to skip these steps as well). The only denoising step remaining will be the removal of session-specific BOLD-signal averages, which is the only one that cannot be unselected, but that should be just fine (your data will have zero average in any way already)

Hope this helps
Alfonso
Originally posted by Stephen L.:

Dear Alfonso,

Perfect, thank you very much for your fast replies! I will try that and let you know how it works out (hopefully this will result in a publication ;-) ).

Cheers from Belgium!
Stephen