I have been using FMRIPrep 20.2.7 ran from Docker and I am aware that distortion correction quality is an issue with FMRIPrep sometimes. I was wondering if there is any heuristic as to when distortion corrected images are still acceptable and what constitutes a failure. I see varying quality across participants (see images) and I am often not sure what to do.
First, I would check whether the direction of unwarping matches what you would expect based on the phase encoding direction. I would also see if there are any warps that are being corrected (e.g., focal points of indents or expansions that are not biologically plausible), in other words is the final product more brain like? Then I would check whether the unwarping leads to better EPI alignment with the t1 image, which should be another part of the HTML report.