Using fmriprep with neurodesk on an HPC

This tutorial was created by Kelly G. Garner.

Github: @kel_github

This workflow documents how to use fmriprep with neurodesk and provides some details that may help you troubleshoot some common problems I found along the way.

An example notebook can be found here: https://github.com/NeuroDesk/example-notebooks/blob/main/books/functional_imaging/fmriprep.ipynb

Assumptions

• Your data is already in BIDS format
• You plan to run fmriprep using Neurodesk
• You have a copy of the freesurfer license file (freesurfer.txt), that can be read from the file system using Neurodesk

Steps

Launch Neurodesk

From the launcher, click the Neurodesktop icon:

Open fmriprep

Now you’re in Neurodesk, use the menus to first open the neurodesk options

and then select fMRIPrep. Note that the latest version will be the lowest on the dropdown list:

This will open a terminal window where fMRIPrep is ready and waiting at your fingertips - woohoo!

Setting up fmriprep command

You can now enter your fmriprep command straight into the command line in the newly opened terminal. Here is a quick guide to the command I have used with the options I have found most useful. Note that fMRIPrep requests the path to the freesurfer license file, which should be somewhere in your system for neurodesk to read - e.g. in ’neurodesktop-storage'.

export ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=6 # specify the number of threads you want to use

fmriprep /path/to/your/data \ # this is the top level of your data folder
         /path/to/your/data/derivatives \ # where you want fmriprep output to be saved
         participant \ # this tells fmriprep to analyse at the participant level
         --fs-license-file /path/to/your/freesurfer.txt \ # where the freesurfer license file is
         --output-spaces T1w MNI152NLin2009cAsym fsaverage fsnative \ 
         --participant-label 01 \ # put what ever participant labels you want to analyse
         --nprocs 6 --mem 10000 \ # fmriprep can be greedy on the hpc, make sure it is not
         --skip_bids_validation \ # its normally fine to skip this but do make sure your data are BIDS enough
         -v # be verbal fmriprep, tell me what you are doing

Then hit return and fMRIPrep should now be merrily working away on your data :)

Some common pitfalls I have learned from my mistakes (and sometimes from others)

1. If fmriprep hangs it could well be that you are out of disk space. Sometimes this is because fmriprep created a work directory in your home folder which is often limited on the HPC. Make sure fmriprep knows to use a work drectory in your scratch. you can specify this in the fmriprep command by using -w /path/to/the/work/directory/you/made

2. I learned the following from TomCat (@thomshaw92) - fMRIPrep can get confused between subjects when run in parallel. Parallelise with caution.

3. If running on a HPC, make sure to set the processor and memory limits, if not your job will get killed because it hogs all the resources.