High DVARS values seemingly unrelated to movement

[DCasteras]

Dear all,

I have been preprocessing two task-based fMRI sequences with fmriprep and I have noticed that the output (confonds_timeseries.tsv) of several participants marks a significant proportion of volumes as motion outliers, with the default parameters being FD > 0.5 mm or standardized DVARS > 1.5.
Upon examining the output HTML files, It appears that these participants have high DVARS values for most of the volumes:

With my limited experience with fmriprep, I did not expect such high DVARS values coming from participants who seemed relatively well-behaved in terms of motion. What could be the causes of such high variation in signal volume to volume? Would it be a reasonable approach to exclude participants based on mean FD (>0.25 or >0.50) and assume that a denoising pipeline will be able to handle noise unrelated to the task?

Thank you,

[DCasteras]

Dear all,

As a follow-up to my previous post, I wanted to share a solution I found in case anyone else encounters a similar problem.
Original preprocessing was done using the version 22.2.3 and updating to the newer 24.0 version did not resolve the issue.
However, reverting back to an earlier version we previously used in our lab (20.2.5) seems to have solved the problem effectively.
23.2.3:

20.2.5:

In the earlier version (20.2.5), the signal appears more stable (lower σ) while the movement assessment remains nearly identical.

This error might be specific to our data, but I wanted to share our findings regardless, hoping it could help others facing the same problem.

Best regards,

Daniel