When running fmriprep version 20.2.0 I’ve noticed that none of the confound files in my fmriprep output contain non_steady_state_outlier_XX columns. However, from running MRIQC earlier, I know that most confounds files should contain some non-steady-state columns. My fmriprep command is:
Hi, I will start with a question: Do you have sbref files in your data set?
I am not totally sure how this is done internally but here is an observation I made recently (I have sbref files in my dataset):
I ran fmriprep twice on the same dataset, once with and once without the argument: --ignore bref.
without the argument –ignore bref, there were no non-steady state outlier in the confounds_timeseries.tsv file. (of note: it is also indicated in the fmriprep HTML report in the summary section at the beginning of each run)
with the argument –ignore bref, the non-steady state volumes reappeared in the HTML report and as extra columns in the confounds_timeseries.tsv file.
That’s strange though, as I’ve never used --ignore sbref, and in the past my confound files have contained non steady state columns, unless I’m misremembering a different dataset.
I’m interested myself in the answer. I don’t have sbref files in my dataset, and I don’t have non_steady_state_outlier_XX columns in my confounds file.
I found this on the website:
This workflow estimates a reference image for a BOLD series. If a single-band reference (“sbref”) image associated with the BOLD series is available, then it is used directly. If not, a reference image is estimated from the BOLD series as follows: When T1-saturation effects (“dummy scans” or non-steady state volumes) are detected, they are averaged and used as reference due to their superior tissue contrast. Otherwise, a median of motion corrected subset of volumes is used. https://fmriprep.org/en/stable/workflows.html
If I understood it correctly, fMRIPrem computes non-steady state outliers either way, with or without sbref files. Am I wrong?
It is not exactly what I understand. From what I understand, the heuristic is the following:
if SBref is present, it will be used as reference and the non-steady state volume(s) won’t be calculated nor used.
if there is SBref, fmriprep will look for non-steady state volume(s) or dummy scans. If the latter are present, they will be used to create a reference volume.
if no SBref , non-steady-state volume(s) nor dummy scans are detected, then the mean of a motion corrected subset (the HCP uses volumes 11 to 20 for instance, one should look in fmriprep code to see what it is done) will be used as a reference.
So, to my understanding, you will only see the non_syeady_state columns in the confounds file if you are in case 2.
But I am not a fmriprep developper, that is only my take, I may be wrong.
If I am right, I would advise to change the html report in the case of SBref files: instead of writing non steady state volumes : 0, I would prefer: non-steady state volumes: not calculated.