Whi fMRIPrep use only opt_comb from Tedana?

,
1

Dear experts,

In the current state, when combining multi-echo fMRI, fMRIPrep is using opt_comb from Tedana.
However, if Tedana is already installed in the image, why to not also use the denoising of the same Tedana, or at least to have this denoising optional?

Is there’s some interference of fMRIPrep processing and Tedana denoising?

2

That is a recommendation from the tedana developers, including myself, for a couple of reasons. First, tedana supports multiple different decision trees, dimensionality estimation methods, etc. and we didn’t want to add all of that functionality to the fMRIPrep command line. Second, tedana’s denoising is not a “solved” issue, and it requires that users review the classification results and potentially modify them.

Instead, we have proposed an fMRIPost-tedana BIDS App, which will take in fMRIPrepped data and apply tedana to that. I haven’t had much time to work on it, but now that fMRIPost-AROMA is working pretty well, it should be possible to implement fMRIPost-tedana without too much work.

1 Like
3

Hi @nbeliy

Apart from what @tsalo has indicated, I would add that fMRIprep also returns multiple additional nuisance regressors. Ideally, nuisance regression should be done in a single step (e.g. using 3dTproject in AFNI). Otherwise, performing another nuisance regression step with these extra nuisance regressors in the tedana denoised data could reintroduce some spurious effects.

Check these papers:
https://doi.org/10.1016/j.neuroimage.2013.05.116
https://doi.org/10.1002/hbm.24528
https://doi.org/10.1155/2013/935154

Therefore, since fMRIprep is specifically designed not to perform any denoising step, it is better to return the optimally combined dataset.

2 Likes
4

Thanks for the quick answers.

So, in order to run tedana denoising we need first determine best parametrisation for Tedana denoising, then replace in fmriprep the echo combination (using fMRIPost-AROMA)?

@CesarCaballeroGaudes can you provide more info on this (or point to relevant fmriprep documentation section?). Is there’s a way to combine these nuisance regressors with tedana’s ones?

1 Like
5

You can check this website from the fMRIprep documentation
https://fmriprep.org/en/stable/outputs.html#confounds
The file ***_desc-confounds_timeseries.tsv includes multiple regressors that are commonly used for denoising through nuisance regression.
You can then create a design matrix for nuisance regression including the relevant ones along with the rejected components from tedana, ica-aroma, etc. You can implement the nuisance regression with FEAT in FSL or 3dTproject / 3dDeconvolve in AFNI.
For instance with 3dTproject, you can add as many -ort files with sets of nuisance regressors, along with Legendre polynomials (-polort) and Fourier terms (i.e. low-, high- or band-pass filtering), and so forth. 3dTproject will perform this nuisance regression in a single step.

6

Thanks @CesarCaballeroGaudes , it’s much clearer now)