Problem with normalization | The R-fMRI Network

Submitted by Stephanie on Fri, 08/30/2013 - 15:23

Hi,

I am using the DPARSF 2.2 version using DARTEL for normalization. Unfortunatly after normalization my functional images (for almost all subjects) show random spots (a couple of voxels) with extremly reduced intensity values. This can be seen as black spots when checking "PicturesForChkNormalization". Prior to normalization, these spots are not visible.

I attached a pdf file showing my DPARSF setting and a couple of images from different subjects.

Thanks in advance for your help.

Regards,

Stephanie

Holes after using DARTEL normalization

Hi Stephanie,

I have asked this question in SPM maillist: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1202&L=spm&P=R71507&1=spm&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4

And here is Dr. John Ashburner's response: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1202&L=spm&P=R74748&1=spm&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4

"This is because the normalise to MNI space option is intended to be used with smoothing. The reason for the holes is that voxels in the original image are pushed to their new locations by scanning through the original and adding their values to the appropriate locations of the normalised version. To obtain the "unmodulated" normalised images, these are then divided by the number of voxels added at each point. With smoothing, it is a smoothed version of the image containing the total value of the projected voxels, which is divided by a smoothed count of the number of voxels. The reason for doing it this way that the average value at each point should be computed as the total count divided by the number of voxels.".

Best,

Chao-Gan

normalization with DARTEL using DPARSF

Thank you for responding so quickly. So if I understand correctly I have to smooth my data prior to normalization when using DARTEL. Is that right? Or is smoothing after normalization sufficient. So what are the implications then when using DPARSF for preprocessing?

Many thanks.

Best,

Stephanie

Re

Hi Stephanie,

In my understanding, for each voxel in MNI space (after normalization), DARTEL is scanning the corresponding voxels in original space and calculate a value, then put the value into that voxel in MNI space. For some voxels in MNI space, no corresponding voxels were found in original space for some reason (may due to DARTEL algorithm). In general, doing a regular "smooth" after DARTEL normalization can fix such a problem -- that's a way I am using, and it's also the so-called "default" parameter in DPARSF.

Best,

Chao-Gan

Dear Yan,

I've suffered from the same problem, but instead of this I used normalize by using epi templates... but I want to try with DARTEL.

You said that in DPARSF is a default mode to fix this problem, but checking with DPARSFA I had a couple of options.

Which fits better to the default in DPARSF? Calculate in MNI Space (warp by Dartel) or others?

One more thing, I've never perform from the defaults/template in DPARSFA, and I realized that I need to perform the analysis again, from the beginnig, if I want to fix the problem. Am I right?

Thank you once I again,

Best.
Juan

Hi Juan,

Yes, "Calculate in MNI Space (warp by Dartel)" is kind of "default".

You can try this option. But actually I am not sure what problem you are refering to.

Best,

Chao-Gan

Hi,

The same as pointed out by the person who created this topic. Exactly the same, random black spots, when I use dartel instead epi.

Thank you!

Ps: I run the default mode with dartel for MNI and I'm still having the same problem.

Cheers!

Hi,

As we discussed in the previous comments, the black dots (or even strips) are related to an issue in DARTEL. And according to John, he thinks smooth is kind of solution.

Best,

Chao-Gan

Hi Yan,

In your reply, you said "doing a regular "smooth" after DARTEL normalization can fix such a problem" . I wonder if I do smooth by DARTEL using DPARSFA after normalization, should I perform smooth again before caculater VMHC? If I want to calculate Degree Centrality, can I perform smooth by DARTEL before calculate DC?