FMRIPREP - Success stories

Hi there,

We would love to collect user experiences with FMRIPREP. The devs team would really appreciate that you spare 5 min to report that you used FMRIPREP successfully. We would very interested in hearing your stories and details like:

  • the overall context of the project (if you can tell about it),
  • FMRIPREP flavor and framework (python, docker, singularity, etc.),
  • where you run it (laptop, HPC, etc),
  • number of participants you processed,
  • MR protocol overview (again, if you can talk about it),
  • links to issues you may have reported,
  • if you’d like us to reach out in the future when we try to gauge FMRIPREP usage, etc.

Thanks a lot. It is very exciting to see interest beyond the building you work in.



I really enjoy using fMRIPREP. Here’s my experience:

  • Working on a large fMRI dataset of twins (N = 400+)
  • I use the Singularity fMRIPREP version on a CentOS-based HPC. We have SLURM as our scheduler, so I like to submit pre-processing jobs to fMRIPREP individually within a SLURM array (so, I have a list of subjects that fills in for the call of each job, and each call is to process one specific subject in my dataset).
  • Random story related to Singularity: I used to build the image on my Mac and that just tended to give poor results on the cluster. I installed CentOS on a VM and built the Singularity image there, and then it worked perfectly.
  • So far I have processed most of my dataset. I have a chunk of subjects who I am having trouble converting to BIDS (due to some weird nibabel issue?), so I’ll have to figure out how I can get those into the dataset to run them through the pipeline.
  • Related to the above, it would definitely be super nice if there was some sort of way to coerce non-BIDS subjects to work on fMRIPREP, or if you guys made a nicer BIDS converting tool that is better supported/documented than what’s currently available.
  • Reported memory issue on GitHub #836.
  • T1 scan, 8 EPI runs on a 3T
  • Yes, please!
  • task-based fmri experiment on familiar face processing
  • fmriprep 1.0.0 running in singularity (converted with docker2singularity)
  • run on Dartmouth’s HPC cluster (; each participant run as a single torque job with 24 cores
  • 18 participants, two sessions, total 22 runs + T1 + T2w, 1.25s TR, 2.5mm voxel size, fmap correction
  • aCompCor failed for some runs; upon re-run it worked (
  • you all know where to find me :wink:
  • A preliminary dataset from an fMRI experiment on social hierarchy
  • fmriprep from 1.0.0rc11 to 1.0.0, in Python 3.6.1 and pip (no docker or singularity)
  • Running on an HPC with CentOS Linux (Hoffman2 cluster at UCLA)
  • 9 participants so far (will scan more)
  • T1w, field maps and 9 functional runs
  • Got issue #858 and #868 (thanks for fixing them!). Another really really small issue I didn’t report is that the current niworkflows (0.2.2) seems to require svgo and cwebp, which is not in the documentation (so I got a few compression errors). It’s no big deal at all and I guess I just need to install them, but I think it would be nice if the documentation mentions them.
  • Sure!

Fantastic project! “Make it work, make it right, make it boring…” for fmri preprocessing, which previously was done heterogeneously and used a lot of secret ingredients.

  • the overall context : fMRI of story processing
  • FMRIPREP flavor and framework: standard fmriprep-docker on local MAC, also on openneuro
  • where you run it: currently on laptop, but trying to implement on singularity for the cluster
  • number of participants you processed: as of yet 10, will scan more, and ~40 in another project
  • MR protocol overview (again, if you can talk about it): continuous data as in resting state, but with meaningful content: standard T1 anatomy and functional scans (TR=2) on a GE scanner (-> no fieldmaps)
  • links to issues you may have reported: crash in anat_prepro, anatomical issue, entering of commands
  • if you’d like us to reach out in the future: sure
    Thanks a lot.

For my PhD at the University of Amsterdam, I am collaborating with a lot of researchers from different departments who are working with (f)MRI data. Additionally, I’m working at the faculty’s neuroimaging centre (“Spinoza Centre for Neuroimaging”, location REC) as ‘technical manager’ to improve and standardize data management, preprocessing, and analysis.

For my recent projects, I’ve converted my data to BIDS (using an in-house developed converter) and used fmriprep (with the fmriprep-docker package) to preprocess the data on our dedicated neuroimaging analysis servers (56 cores, 126GB RAM). These projects include two large datasets with together more than 500 subjects, whose (3T MRI) data includes anatomical data (T1-weighted), DWI data, and functional data (four tasks). I’m using both phasediff-unwarping (in some datasets) and topup-unwarping (in other datasets).

Next year, I hope to implement a protocol that converts all data acquired at our centre to BIDS and (optionally, if desired by the researcher) preprocesses the data using fmriprep.

You can contact me at the email address listed on my Github page.
Also, importantly: a big thank you for the fmriprep developers for this awesome tool!


This tool really is a god-send for bringing consistent analyses across labs (or even within labs) and in combination with BIDS, really breaks down the barrier of entry for analysis of data. At the same time, detailed documentation with references give a pathway for those new to FMRIPREP or analysis in general to pick up how the tool works as they use it. Finally, the choice of writing it in python instead of C++ (or some other low-level language) lowers the barrier of entry for those that wish to contribute to this software.

  • the overall context of the project
    • We study how exercise/cognitive training interventions impact brain health in older adults, as well as correlational studies that relate physical activity/fitness levels with brain health.
  • FMRIPREP flavor and framework (python, docker, singularity, etc.),
    • started using before v1.0.0 release, but the most recent version used for our data is v1.0.6 inside a singularity container
  • where you run it (laptop, HPC, etc),
    • Our HPC cluster installed singularity (last summer), so I’ve been heavily reliant on that.
  • number of participants you processed,
    • ~100 unique subjects (some people have been processed multiple times and some datasets reprocessed)
  • MR protocol overview
    • resting state (10 min), task (7-9 min, simon or flanker depending on project), and T1w & T2w, dwi (60 directions)
  • links to issues you may have reported,
  • if you’d like us to reach out in the future when we try to gauge FMRIPREP usage, etc.
    • Of Course!

EDIT: We are also starting a collaboration with another lab in Texas, and we are going to use BIDS for easier sharing of scripts/data and make analyses easier. MRIQC and FMRIPREP are going to play a crucial role in quality assurance and preprocessing.