John,
Because DARTEL segmentation of my T1 images included quite a bit of the non-gray matter tissues around brain in the rc1 images, and the lower and posterior part of the brain still over stretched, I tried VMB8. VBM8 seems to produce much cleaner segmentation and the normalized brains (wmr and wrp1 images) showed little over stretching. However, I don't know whether and how I may apply the normalization parameters from VBM8 to the EPI images (the T1 images are coregistered to the EPI images). Can the forward deformation file from VBM8 be used to normalize the EPI images?
Thanks for your help.
Ben
-----Original Message-----
From: Xu, Ben (NIH/NINDS) [E]
Sent: Tuesday, June 22, 2010 3:51 PM
To: 'John Ashburner'; [log in to unmask]
Subject: RE: [SPM] Normalizing fMRI data using DARTEL
John,
Thank you very much! This is very helpful.
Regarding my last question (question 5), I got this error message while DARTEL was trying to create templates:
*******
Failed 'Run DARTEL (create Templates)'
Array dimension must match for binary array op.
In file "/usr/local/spm8/toolbox/DARTEL/spm_dartel_template.m" (v3172), function "spm_dartel_template" at line 101.
*******
It worked after I removed the T1 images that have difference number of slices. As the voxel size of all images are the same, I assumed that it was the slice number problem. If it was not, what does this error message mean? What might be the problem?
Thanks for your help.
Ben
-----Original Message-----
From: John Ashburner [mailto:[log in to unmask]]
Sent: Tuesday, June 22, 2010 12:26 PM
To: [log in to unmask]
Subject: [SPM] Normalizing fMRI data using DARTEL
I'm new in using DARTEL with spm8 to normalize fMRI data to the
MNI space. I have used DARTEL for structure (T1 images)
segmentation and normalization before, but not with EPI images.
I have read the DARTEL guide in SPM8 manual and related messages
in the spm message archives, but still not clear about the
specific steps. Here are my questions.
1. Do I need to segment and normalize multiple subjects' TI
images first to create the template and flow field images in
order to normalize a subject's EPI images? Can I run the
segmentation and normalization with one subject's T1 and EPI
images only? If the latter is possible, how to set up the steps
using DARTEL? (Single subject analysis is desirable if I want to
look at the data before other subjects' data are acquired.)
The aim is to bring all subjects in a study into common anatomical alignment. If the population of subjects is warped to their average shape, then findings should be less biased towards a particular template and the distortions should be less extreme. Therefore, the application of Dartel to a single subject may not work so well.
In principle, I could release a crisper template to which the data are aligned. However, what the segmentation considers as grey matter in a scan collected with some particular sequence may not be considered grey matter if a scan of the same subject was collected with another
sequence. Therefore a standard template may not be representative of
the images people have in their study.
2. If the structure segmentation and normalization must be done
with more than one subject's images, is the coregistration
between the subject's T1 and EPI images necessary?
It depends. If the subject does not move between collecting functional and structural data, then the information in the image headers may provide good alignment.
At no point does SPM use magic, so if the fMRI and anatomical scans are not aligned, then the software will not know this. Therefore, coregistration is recommended, as is the use of the Check reg button to ensure that the alignment is good.
3. Once the T1 segmentation and normalization are completed, how
to apply the normalization parameters to the EPI images (already
realigned and resliced)?
If the anatomical and functional data are aligned (according to Check Reg), then the software will use the information about their alignment in the headers. You should not need to do anything else.
Note the treatment of regions in the normalised data where data was not available in the original scans (ie in parts of the head not covered in the FOV of the original scans, or which were set to NaN in the contsrat images). The Dartel software will try to make an educated guess about what values should be put in these regions, but usually it is necessary exclude them from the analysis. Questions about this appear every so often on the mailing list.
4. A slightly different but related question: I have segmented
(using New Segment with the default settings) and normalized
(using preserve "amount") two subjects' TI images (mprage, voxel
size = 1x1x1). However, the c1 images clearly included part of
the non-gray matter tissues on the top and the posterior part of
the brain (see the attached image file "DARTEL New Segment").
These non-gray matter tissues also seemed to be carried over to
the template images (attached "DARTEL smwc1"). What parameter
settings I should try that may improve the segmentation step?
There isn't really anything in the software that will fix this.
However, I would eventually hope to get around to including a MRF model that will tidy the results up.
I haven't tried Christian Gaser's toolbox, but this may do a reasonable job of stripping off these mis-classified regions.
5. I also realized that in creating the "template," DARTEL
requires all T1 images to have the same number of slices. All my
T1 images have the same voxel size, but some subjects' images
have a few more slices than others although the brain and the
top of the skull are covered in all subjects even without the
extra slices. Is there a way to have DARTEL accept images with
different number of slices?
Dartel does not specifically require all original scans to have the same number of slices, although it is good practice to ensure that all scans were acquired using the same sequence. It is the "imported" data that should all have the same dimensions.
Best regards,
-John
--
John Ashburner <[log in to unmask]>
|