what tools are there out there that I can use to manually refine an OK but not great fmriprep alignment?
When I worked on this last I sort of resolved a bigger problem of the alignments sometimes being really terrible: Fmriprep output badly misaligned?
The solution was simply to add the bold2t1w-init header option at running fmriprep.
But now I see that the alignments are still not there. I can remember doing it by hand long ago, but that was part of another package that I cannot use without creating a totally different workflow…
Here’s an example of what I thought was pretty OK, just mean of one EPI scan:
But then when I look at the parameter maps I compute elsewhere:
It’s clearly rotated forward-and-up.
So is there some kind of tool for making such fine adjustments?
Or could anyone explain how I could do it (if advisable) by directly adjusting the affine matrices for a scan/image (basically then just looking at plot_stat_map images like here)? surely someone has written such a tool already…
I use ITKSNAP’s manual & automatic registration tools for such partial coverage alignments. You can load one image as main and another as additional image. Then navigate to the registration menu and explore the GUI (e.g. see this very short demo of registration in ITK-SNAP 3.6 - YouTube ).
AFNI has a Nudge plugin, but it isn’t as pretty as the itkSNAP tool. For itkSNAP, we sometimes see left-right flipping of the data that probably depends on the initial orientation of the data.
From your original post, it seems you wanted to adjust the numbers in the transformation matrix itself. That ought to be doable because they are typically stored in a text file in most software, I believe. You have the choice of moving either the anatomical to the EPI or vice versa. Changing the affine matrix by a translation should be a simple addition or subtraction. Concatenating a translation from one of the above tools might be possible too. That would have the advantage of avoiding any additional interpolation, but it makes the transformation chain more complicated.
