I have a question about the motion-correction algorithm works and how I can customize it for my needs!
The reason we use a mask around the cord is to make sure regions outside of the cord does not adversely impact motion-correction, right? What if I was indeed interested in these regions, what would you suggest?
Basically, if I were to motion-correct my GE-EPI images to investigate the activation not only within but also outside of the spinal cord, would you suggest:
to make my current mask (that is created using sct_create_mask command with the default size) diameter larger so that it covers a bigger portion of the image?
to make my current mask more customized (rather than a bigger cylinder or box) so that it includes structures I am interested in looking at (such as muscles)?
In that case would using mutual information as a metric be helpful?
Thanks a lot in advance and looking forward to hearing from you!
Regards,
Merve
Thank you so much for your question! I appreciate the level of detail you’ve provided in your post.
Disclaimer: I am not as personally familiar with motion correction for functional MRI, so I have also tagged my colleagues who will provide a follow-up response. But, I am replying just to get the ball rolling.
Just to make sure I understand: What is the motivation for considering regions outside of the spinal cord for motion correction? As far as I understand, motion correction consists only of X/Y translations, meaning that the if the cord is aligned across volumes, then the other tissues will be translated the same amount (and thus aligned as well).
But, maybe I am missing some sort of limitation to the cord-based alignment approach? I would be happy to hear your thoughts on the matter, so that I can improve my understanding of our users’ use-cases.
Basically, if I were to motion-correct my GE-EPI images to investigate the activation not only within but also outside of the spinal cord, would you suggest:
to make my current mask (that is created using sct_create_mask command with the default size) diameter larger so that it covers a bigger portion of the image?
to make my current mask more customized (rather than a bigger cylinder or box) so that it includes structures I am interested in looking at (such as muscles)?
To keep it simple I would go with “1”, but feel free to explore “2” as well if you feel confident it would produce better results
In that case would using mutual information as a metric be helpful?
Sure, that metric should work equally well than with a smaller mask-- feel free to explore other metrics as well (eg CC).
Thank you so much for the quick and very helpful replies!
I believe we are excluding the regions that are outside of our mask that we use for moco (so they are not aligned) - but now I would like to include them to investigate what is going on there, too!
Thank you! Will try both and compare.
So, MI should be better in a bigger mask? Did I understand that correctly?
Oh! That clarifies things, and actually perhaps there may be a misunderstanding here?
The mask option for moco (i.e. the -m option of sct_fmri_moco) shouldn’t actually crop the image – instead, it just tells the motion correction algorithm which parts of the image to use as visual landmarks when performing alignment. But, the actual transformations themselves will be applied to the full volumes. (See the figure in the fMRI tutorial for reference.)
The reason this is possible is that the motion correction only consists of axial affine transformations, so it is just as easy for us to apply these transformations to the full volumes as it would be to apply them to the spinal cord region.
I just double checked this myself, and indeed the “fmri_moco.nii.gz” image contains the regions outside of the mask. (i.e. the motion correction is done on a slice-wise basis for each of the full volumes)
Oh, interesting!
I knew the regions were not cropped, as you said, moco images contain the whole volume. But I thought regions outside of the mask are “not considered”, meaning, whatever my metric for registration is would be only “calculated” within that region, so if I am interested in a bigger area (please see below), the yellow mask would not be optimal. Is this correct?
