Skip to content

Latest commit

 

History

History
111 lines (70 loc) · 3.81 KB

running.rst

File metadata and controls

111 lines (70 loc) · 3.81 KB
Raw

Running mriqc

"Bare-metal" installation (Python 2/3)

The software automatically finds the data the input folder if it follows the BIDS (brain imaging data structure) standard [BIDS]. A fast and easy way to check that your dataset fulfills the BIDS (brain imaging data structure) standard is the BIDS validator.

Since mriqc follows the [BIDSApps] specification, the execution is split in two consecutive steps: a first level (or participant) followed by a second level (or group level). In the participant level, all individual images to be processed are run through the pipeline, and the MRIQC measures <measures> are extracted and the individual reports (see The MRIQC Reports <reports>) generated. In the group level, the IQMs (image quality metrics) extracted in first place are combined in a table and the group reports are generated.

The first (participant) level is executed as follows: :

mriqc bids-dataset/ out/ participant

Please note the keyword participant as fourth positional argument. It is possible to run mriqc on specific subjects using :

mriqc bids-dataset/ out/ participant --participant_label S001 S002

where S001 and S002 are subject identifiers, corresponding to the folders sub-S001 and sub-S002 in the BIDS (brain imaging data structure) tree. Here, it is also accepted to use the sub- prefix :

mriqc bids-dataset/ out/ participant --participant_label sub-S001 sub-S002

Note

If the argument --participant_label is not provided, then all subjects will be processed and the group level analysis will automatically be executed without need of running the command in item 3.

After running the participant level with the --participant_label argument, the group level will not be automatically triggered. To run the group level analysis: :

mriqc bids-dataset/ out/ group

Examples of the generated visual reports are found in mriqc.org.

Depending on the input images, the resulting outputs will vary as described next.

Containerized versions

If you have Docker installed, the quickest way to get mriqc to work is following the running with docker guide <docker>.

Running MRIQC on HPC clusters

Singularity containers

Requesting resources

We have profiled cores and memory usages with the resource profiler tool of nipype.

An MRIQC run of one subject (from the ABIDE) dataset, containing only one run, one BOLD task (resting-state) yielded the following report:

Using the MultiProc plugin of nipype with nprocs=10, the workflow nodes run across the available processors for 41.68 minutes. A memory peak of 8GB is reached by the end of the runtime, when the plotting nodes are fired up.

We also profiled MRIQC on a dataset with 8 tasks (one run per task), on ds030 of OpenfMRI:

Again, we used n_procs=10. The software run for roughly about the same time (47.11 min). Most of the run time, memory usage keeps around a maximum of 10GB. Since we saw a memory consumption of 1-2GB during the the 1-task example, a rule of thumb may be that each task takes around 1GB of memory.

References