Hello,
Out of curiousity, I wanted to test different interpolation methods in 3dvolreg to see which produced the best registration for my data – the results seem surprising to me and I was wondering if anyone had any insight into why things turned out this way (sorry in advance for the lengthy post)…
Part 1:
I ran the following command line:
3dvolreg -verbose -verbose -Fourier -twoblur 3 -twopass -coarse 1 4 -zpad 2 -prefix Vrun1 -base 'run1+orig[50]' -dfile run1ORT.1D run1+orig
and compared it with the output from running the same exact command line, but substituting “heptic” for “Fourier” for a group of 10 adults subjects.
I used three indices for comparison:
1. For each subject, the sum of the (RMSold-RMSnew) across all TRs. I obtained the RMSold and RMSnew values from the last two columns of the dfile and simply subtracted them.
2. For each subject, the average RMSold-RMSnew across all TRs.
3. For each subject, the maximum RMSold-RMSnew across all TRs.
All three indices for each subject (and thus for the group as a whole) are significantly larger (and I assume, better) for the heptic interpolation - I had predicted that Fourier would be better.
Part 2:
I am further confused by the fact that when I run the following command line (Fourier interpolation, single pass registration):
3dvolreg -verbose -verbose -Fourier -prefix Vrun1 -base 'run1+orig[50]' -dfile run1ORT.1D run1+orig
and compare it with the results of the command line in Part 1 above (Fourier interpolation, two pass), the numbers look almost identical. I was under the impression that the two pass registration would be significantly better. I also did this little experiment with a group of four child subjects (more movement in the data) and everything came out the same way.
Any ideas?
Thanks in advance!
Brianna