mohmdrezk / bids_fmri_scripts Goto Github PK

• make FAST the default instead of AR(1)
• pass HRF time
• dispersions derivatives
• possibility to pass a specific mask
• residuals of GLM for quality checks on sphericity
• implement MAC toolbox goodness of fit?
• other quality control (e.g from SPMup ?)
• refactor!!!

This could actually be turned into a kanban / project on github: @mohmdrezk let me know if you want to set one up.

Things to check:

• : test on several BIDS dataset from openneuro
• : test with octave
• : test with older matlab

Things to do:

• containerize with docker
• paralellize where possible (parfor loops).
• separate matlabbatch creation and matlabbatch running: could come in handy for containerization or when working on a HPC

Apparently the code base calls some functions that are not just SPM based.

e.g.
n=load_untouch_nii(fileName);
save_untouch_nii(n_dummies,fullfile(dummiesOuputDir,['dummies_',fileName]) )

Are they functions from vistasoft?

I guess we could add the whole thing as a dependency but that might be a bit of an overkill: depends how much we use things from that codebase.

Otherwise we could use SPM functions to do the same.

I think that one good way to refactor the code will be to make use of the spm_BIDS function to grab a lot of info from the data set

I think we definitely need to add a check to the SPM version otherwise people will run into trouble with the json reading problems.

• for preprocessing
• for first level GLM
• deal with prefix

There is one issue about the code in general. Since I was copying it from the older scripts, everything is based on macOS/linux directories. So these scripts wont work well with windows directories.
This also on my list, that the directories should use "fullfile" function rather than the strings with the "/" or "".

It would most sense to do this using the BIDS stats model extension.

I will go through it and I will try to suggest something.

https://docs.google.com/document/d/1bq5eNDHTb6Nkx3WUiOBgKvLNnaa5OMcGtD0AZ9yms2M/edit?usp=sharing

Also separate matlabbatch creation and matlabbatch running: could come in handy for containerization or when working on a HPC

make sure that this does not mess things with the prefix for realignment

I don't think it will be an issue but maybe just in case?

I have kept working on the pipeline and tried to adapt it.

Before I submit a new pull request you can check what I have done by

• add my repository as one of your remote to your own repository by typing this where you are in the folder of your own project on your local machine:
  git remote add remi https://github.com/Remi-Gau/BIDS_fMRI_scripts.git
• get the content of my repository onto your machine
  git fetch remi
• switch to one of my branches to look what I have done
  git checkout implement_BIDS_preproc

You will not be able to push changes to that branch until I give you write access but it can allow to see where I am at the moment before I make a pull request.

Of course you can make branch from this work I have done and make a PR on implement_BIDS_preproc on repo but I think that it makes more sense if we keep your repo as the place where most of the final merging happens.