How is the number of voxel used in the normalization?

[saslanmclean]

Hello

I used the calculation in MNI space. However due to memory and speed limitations, I try to use the program in the subject original space. In this case, how shall I properly choose normalization?

(I was using -mean_time -median_vol option before)

Best Regards
Serdar

[WillForan]

The short answer is continue to use -mean_time -median_vol.

But I'm not sure I totally get the question yet, and suspect there's something like an XY Problem. Or maybe an issue with the overloaded "normalization" term ("spatial normalization" = warping to e.g. MNI, temporal and/or value normalization == mean/median ?)

What's your input look like: matrix size, volume count -- output of something like 3dinfo -ad3 -n4 example_input.nii.gz, number of runs in a visit; MNI vs native space differences? Do you have any specific errors or benchmarks? tat2 is a thin wrapper around what I thought were pretty efficient AFNI commands.

tat2 will collapse across time (with the specified per volume summary stat) for a single 4D image regardless of the input 4D image's space (MNI, anat, native). The options work the same for orig as they do for MNI. You can give it whatever you want and, as long as you've at least motion corrected, the calculations should be sound. (Additional minimal preprocessing would be good too. Like slice-time correction, susceptibility distortion correction, despiking)

But, if you have multiple runs within a single visit to combine (eg. task-rest_run-1 and task-rest_run-2), they'll at least need to be aligned/co-registered to a common reference (e.g. mean of both as reference to 3dvolreg, mcflirt, antsMotionCorr) before going into tat2. I think SPM calls this realignment.

If you want to compare across subjects with native/orig-space tat2 output, you'll eventually need to either

1. warp the per visit output tat2.nii.gz to a shared template (likely MNI) or
2. extract native spaces metrics (e.g. 3dROIstats where mask is given ROI in warped to native space, or freesurfer aseg ROI warped to EPI-native space)

On the other hand, if you've already run fmriprep, the mni output of that (like *_task-rest_run-1_space-MNI152NLin2009cAsym_desc-preproc_bold.nii.gz) makes a good already-aligned input for tat2. And simplifies comparing across visits later. (I'm guessing since you're trying to avoid large computations, running fmriprep is a non-starter).

---

Or is the issue specifically the difference in size between the functional images in native space and in MNI space (3dinfo -n4 func_space-orig.nii.gz is smaller than 3dinfo -n4 func_space-mni.nii.gz)? In that case, you're maybe upsampling when going to MNI (?) and could use a MNI template with bigger voxel sizes?