Removing Dummy Scans - fMRIPrep

Summary of what happened:

Hi, I am new to using fMRIprep and I have a query about Dummy Scans.

If possible, I would ideally like to remove the dummy scans (i have 5) before running fMRIprep so that it closely matches how I would’ve preprocessed my data in SPM. However, I cannot see experience of people doing this. I have raw dicom files, that I am converting to .nii and putting into BIDS via Heudiconv and then running fMRIprep. Is there a way to correctly remove dummy scans (as you would in SPM), do you need to do this prior to using Heudiconv, or would this cause inconsistencies in the headers and json files? Does anyone have experience here?

Alternatively, I can see if that I could put --dummy-scans 5. However, it is not clear as to what this does.
Firstly, how are those 5 dummy scans then treated in the fmriprep processing steps? Secondly, in my first level analysis script in SPM, I adjust my task times to take into account that I have deleted the first 5 volumes - would i still do this, because it seems that using the tag doesn’t actually remove the dummy scans.

Hi @MRI_New,

It adds one-hot encoded regressors to the confounds files that can be imported into first level model denoising.

You would not adjust the event timings.

Best,
Steven

Hi, Thank you for getting back to me so fast. Does this mean that it is not feasible to simply remove the volumes before running fMRIprep?

Hi @MRI_New,

You can do that, but it is typically recommended to not mess with raw data.

Best,
Steven

Hi Steven,

Thank you for your help so far. We have looked into the data and unfortunately we must adjust the timings, as we only have information for the offset of the last dummy scan, which needs to be our time 0. We are replicating a previous study which deleted the dummy scans prior to manual spm preprocessing and we are missing the information regarding the onset for the dummy scans - leaving us with no choice but to adjust the timings.

My questions therefore are:

1, Do I run fMRIprep with “–dummy-scans 5”, then disregard the extra regressors, remove the first 5 lines from the motion regressor file, and conduct SPM first level analysis from the 6th volume onwards.
2, Or do I run fMRIprep without the dummy scan flag, and do as above.

3, Do you foresee any problems that the dummy scans would be present when conducting preprocessing, but disregarded when conducting first level analysis?
4, Do you have any other suggestions?

Hi @MRI_New,

Event timings should correspond to the BOLD file, so if your event timings are currently defined relative to the beginning of the non-dummy scans, I would adjust your event timings. You could remove the dummy scans but I am personally more of a data purist and advise against altering raw data unless absolutely necessary.

I am not sure I see the point of specifying the dummy scans argument if you are just going to disregard the extra regressors.

As long as the event timings are correct relative to the BOLD file and you regress out the effect of the dummy scans, should be fine.

Best,
Steven

Hi,

Thank you for getting back to us so quickly.

The events are relative to the end of the 5th dummy scan, not the beginning unfortunately. So, to have correct timings, we must make the end of the 5th Dummy Scan time-zero and look at the BOLD data from the 6th volume onwards.

What we would like to know is how to appropriately handle this in fMRIprep.

The previous study conducted preprocessing in SPM, by only conducting each preprocessing step from the 6th volume onwards, and the same for the first level analysis. We would like to conduct our preprocessing in fMRIprep so we are trying to match this as closely and appropriately as possible.

We wondered if we should remove the first 5 dicoms before fMRIprep but were advised that this would result in inaccurate meta-data going forward.

We are concerned that conducting fMRIprep on all the volumes and only conducting the first-level analysis on the 6th volume onwards would cause issues - as the dummy scans would be included in the motion correction, co-registration, slice timing correction (we have interleaved data), susceptibility distortion steps etc - even though we would be disregarding them for the first-level analysis. Would you think this is ok to do?

Hi @MRI_New,

fMRIPrep does not deal with event timings. That is only used in your GLMs.

Is there anything particularly concerning about these dummy scans that makes you think they would impact motion correction, co-registration, SDC throughout the rest of the timeseries? Have you looked through the fmriprep htmls to do quality assurance?

Best,
Steven

Although fMRIprep does not deal with event timings - we are trying to work out how best to handle the preprocessing so that the first-level analysis is not impacted.

We saw here that using the dummy scans flag impacts the slice timing correction within fMRIprep:
FAQ - Frequently Asked Questions — fmriprep version documentation :

“Hence, nonsteady states are discarded by the slice timing correction tool (in this case, AFNI’s 3dTShift ).”

This is why we were unsure as to whether we should include the dummy scan flag so that the preprocessing takes into account their existence, even though we will not be including the first 5 volumes in our first-level analysis later on
and if we didn’t use the flag, would the output of the preprocessing be inappropriate for first-level analysis that will not include all the volumes from the preprocessing.

I would personally include the dummy scans flag and adjust event timings.