help > Surf Ice vs SPM rendering - different outputs
Showing 1-2 of 2 posts
Oct 28, 2016 01:10 PM | Lydia Vinals
Surf Ice vs SPM rendering - different outputs
Dear Chris,
I've recently started using Surf Ice to do my surface renderings and I really love it! It's great!
I just had a question regarding discrepancies that I have observed between SPM renderings and Surf Ice renderings of the same results. For a particular contrast of interest, I have been saving a thresholded SPM .nii file. I then use this .nii file as an overlay in Surf Ice. The shape and extent of the activation clusters that I get in Surf Ice do not correspond exactly to those that I get when using SPM rendering. I have played around with the min and max values of the threshold in Surf Ice but I still don't seem to get the same rendering. I understand that these are visualisation tools and that they may yield slightly different renderings but I was just wondering whether there was a way to improve the correspondence between the two. I'm attaching an example where the same contrast in rendered using SPM and Surf Ice. As you can see, the Surf Ice rendering is oddly shaped and doesn't really reflect the same pattern of activation as the SPM rendering. Perhaps I'm doing something wrong?
Many thanks for your help,
Lydia
I've recently started using Surf Ice to do my surface renderings and I really love it! It's great!
I just had a question regarding discrepancies that I have observed between SPM renderings and Surf Ice renderings of the same results. For a particular contrast of interest, I have been saving a thresholded SPM .nii file. I then use this .nii file as an overlay in Surf Ice. The shape and extent of the activation clusters that I get in Surf Ice do not correspond exactly to those that I get when using SPM rendering. I have played around with the min and max values of the threshold in Surf Ice but I still don't seem to get the same rendering. I understand that these are visualisation tools and that they may yield slightly different renderings but I was just wondering whether there was a way to improve the correspondence between the two. I'm attaching an example where the same contrast in rendered using SPM and Surf Ice. As you can see, the Surf Ice rendering is oddly shaped and doesn't really reflect the same pattern of activation as the SPM rendering. Perhaps I'm doing something wrong?
Many thanks for your help,
Lydia
Nov 1, 2016 01:11 PM | Chris Rorden
RE: Surf Ice vs SPM rendering - different outputs
Lydia-
The NIfTI files saved by SPM are on discrete voxel grid, while Surf Ice interpolates these values to a 2D mesh. In other words, there is not a perfect one-to-one correspondence with each NIfTI voxel and each mesh vertex. I have a couple thoughts on this:
1.) You will probably get a stronger correspondence between SPM and MRIcroGL (which is a volume renderer based on voxels) than between SPM and Surf Ice.
2.) The depth of your surface mesh also plays a role, for example the surface of the the mesh mni152_2009.mz3 is much less eroded than BrainMesh_ICBM152.mz3, so the latter will probably catch signal that is not precisely on the surface.
3.) I would strongly recommend that for both Surf Ice and MRIcroGL you you load an UNTHRESHOLDED statistical map, and then apply your numerical threshold. If you use SPM to save a THRESHOLDED image, then you are artificially surrounding the activity with zero values, and this will make the interpolated activity look less significant and smaller than it is in fact (due to smoothing, voxels near clusters that survive thresholding tend to have values close to the threshold value, and far from zero). For FSL users, you would want to load your "zstat1.nii.gz" as an overlay, rather than the "thresh_zstat1.nii.gz" and then apply your statistical threshold.
The NIfTI files saved by SPM are on discrete voxel grid, while Surf Ice interpolates these values to a 2D mesh. In other words, there is not a perfect one-to-one correspondence with each NIfTI voxel and each mesh vertex. I have a couple thoughts on this:
1.) You will probably get a stronger correspondence between SPM and MRIcroGL (which is a volume renderer based on voxels) than between SPM and Surf Ice.
2.) The depth of your surface mesh also plays a role, for example the surface of the the mesh mni152_2009.mz3 is much less eroded than BrainMesh_ICBM152.mz3, so the latter will probably catch signal that is not precisely on the surface.
3.) I would strongly recommend that for both Surf Ice and MRIcroGL you you load an UNTHRESHOLDED statistical map, and then apply your numerical threshold. If you use SPM to save a THRESHOLDED image, then you are artificially surrounding the activity with zero values, and this will make the interpolated activity look less significant and smaller than it is in fact (due to smoothing, voxels near clusters that survive thresholding tend to have values close to the threshold value, and far from zero). For FSL users, you would want to load your "zstat1.nii.gz" as an overlay, rather than the "thresh_zstat1.nii.gz" and then apply your statistical threshold.