We recently noticed a problem when running @SUMA_Make_Spec_FS specifically with the -NIFTI (or any -*IFTI option). This has been fixed as of AFNI version 17.0.16 (built March 21, 2017).
This issue affects group analysis with standard mesh surfaces. Using one of these options, node assignments to the standard mesh were variably shifted based on the dataset origin. The result of this is akin to a (likely moderate) misalignment across subjects. We recommend anyone who had used one of these *IFTI options to rerun the @SUMA_Make_Spec_FS script with the latest version.
From now on, we recommend using the -NIFTI option with @SUMA_Make_Spec_FS in order to simplify maintaining volumetric alignment with input datasets.
Special thanks here to Iain Dewitt, whose discovery of this bug and persistence was vital for our understanding.
For an example of the variation in the surfaces, see the attached images of the curvature map for a specific subject in both the asc and in GIFTI format. Similar variations could be expected across subjects.
Some details for those interested and some reminders of other issues regarding importing FreeSurfer surfaces and data into AFNI/SUMA format.
Standard mesh analysis. Standard mesh format data had not been imported properly with GIFTI or NIFTI options. MapIcosahedron was not taking into account a shift in the registered sphere GIFTI surface, and that bug has now been fixed. The effect was that nodal correspondence was lost among standard mesh surfaces, and the NIML datasets for those surfaces were variably shifted relative to each other. Processing without any options creates a non-shifted surface in FreeSurfer asc format. For an example of the variation in the surfaces, see the attached images of the curvature map for a specific subject in both the asc and in GIFTI format. Similar variations could be expected across subjects. MapIcosahedron has an option now that provides a quantification of the distortion distance from the correct position (-write_dist). In our preliminary testing, we have seen a mean of 8mm distortion.
The fix - MapIcosahedron has been modified to handle the shifting properly.
What this means → Single subject analysis is not affected for either asc or GIFTI output. However, group analysis with GIFTI format surfaces will have to be reprocessed by rerunning @SUMA_Make_Spec_FS. The default processing without GIFTI/NIFTI options does not produce this issue.
NIFTI volumes. By default, FreeSurfer shifts the volumes on input and in the output. The previous output used AFNI format that were not translated back to match the original space. By using the -NIFTI option (or -GIFTI,-GNIFTI or -IFTI), the NIFTI output volumes would be translated back to match the input data properly. The default AFNI format option would produce datasets that match the surfaces and segmentation but not the original dataset format.
Ranking. Recently, we changed the sorting done by AFNI to produce a ranked segmentation dataset that was consistent among subjects. For some subjects, certain regions were missing from some subjects. The ranked dataset had contained only those regions present for that subject. That made scripting more difficult. Having a consistent numbering scheme makes that easier for both display and analysis, but still use the original numbered format for regular analysis. See this recent link for more information:
Odd voxels and 1mm. If the input dataset is not 1mm or not an even number of voxels wide, the dataset would be scaled. An affine transformation computed by @SUMA_Align_to_Experiment can be used to correct this.
Overall recommendations:
Use the new NIFTI/GIFTI options with the new @SUMA_Make_Spec_FS.
Examine your data for consistent node anatomical positions across subjects.
Use the renumbered segmentation datasets for display, but use the original datasets for analysis.
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.
