Hi Gabor,
Thanks a lot for your detailed description. It is very clear and helpful!
Thanks again!
Regards
Li
________________________________
From: [log in to unmask] [[log in to unmask]]
Sent: Thursday, September 20, 2012 3:35 PM
To: Kong, Li
Cc: [log in to unmask]
Subject: Re: RE: RE: the problem of mapping VBM results to native space
Hello Li,
the necessary parameters can already be stored during segmentation step. The settings and namings depend on the type of segmentation. So forgive me the detailed description below, but it's really a little confusing...
With "segment", both mat files *_seg_sn and *_seg_inv_sn should be created and stored automatically when using the GUI interface (including the batch window). To apply these parameteres onto other images, use 1) "Normalize (Write)" and choose the corresponding mat file for "Parameter File" or 2) "Deformations" and choose "Imported _sn.mat".
With "New segment", you get a single mat file called *_seg8.mat. This file is not compatible with "Normalize (Write)". Actually I'm not sure what it's for, but anyway :-) To get the transformations during "New Segment", you have four options when using the GUI/ the batch editor, given at "Deformation fields": none, inverse, forward, inverse + forward. The forward file is called y_subject.nii, the backward file is called iy_subject.nii. In general, the backward file is much larger, don't worry. To apply the parameteres to other images, you have to use "Deformations", and choose "Deformation Field" as "Composition".
When using VBM8 toolbox by Christian Gaser, you also get only one *.mat file called *_seg8.mat. Again, this is not compatible with "Normalize (Write)", but again, you have four options to create deformation fields. Naming is quite similar to "New Segment", the file starting with y*.nii being the forward and iy_*.nii being the inverse. Choosing "High-dimensional: Dartel" as "Spatial normalization", they will actually be called y_r*.nii and iy_r*.nii, choosing "Low-dimensional: SPM default" the "r" is missing.
In case you don't use the GUI but work with m files / scripts directly, it might be necessary to add something to the code to create both the two mat files or deformation fields, dunno. John should be the expert for this :-)
Best,
Gabor
-------- Original-Nachricht --------
Datum: Wed, 19 Sep 2012 16:42:16 +0000
Von: "Kong, Li" <[log in to unmask]>
An: "Graf Öderland" <[log in to unmask]>
CC: "[log in to unmask]" <[log in to unmask]>
Betreff: RE: RE: the problem of mapping VBM results to native space
Hi Gabor,
Thanks for your idea!
You are right! I just segment raw image directly and write *_seg_inv_sn.mat to my cluster. Warped cluster locates in the right position of the raw image. Thanks again!
The cluster is got from normal VBM analysis in SPM8. I run VBM with defaut setting and did statistical analysis for smoothed GM. However, I did not got any *_seg_inv_sn.mat files (Even I changed the setting of "Deformation Fields" in VBM Batch Editor into "Template--> Image (Inverse)", I still can't get *_seg_inv_sn.mat ). So rerun my subjects step by step to get the *_seg_inv_sn.mat files.
So I think this cluster is no problem. Is it possible to get *_seg_nv_sn.mat files in VBM analysis??
Thanks for your help!
Regards
Li
________________________________
From: "Graf Öderland" [[log in to unmask]]
Sent: Wednesday, September 19, 2012 5:40 PM
To: Kong, Li
Cc: [log in to unmask]; [log in to unmask]
Subject: Re: RE: the problem of mapping VBM results to native space
Hello Li,
the normalization part is included in segmentation (for details, see Ashburner & Friston, 2005, Unified segmentation), so no prior normalization step is needed for your purpose. Based on the naming of the files and your description it seems you accidentally normalized the structural volume (getting a subject_sn.mat and a wsubject.img), and then entered the wsubject.img into segmentation, thus getting wsubject_seg_sn.mat and wsubject_seg_inv_sn.mat
Thus these parameters are the "normalization" parameters for an already normalized version of your structural volume. Therefore the back-projected cluster (using w*_seg_inv_sn.mat) will not match the structural volume s*.img of course, but it should go well with the normalized version ws*.img.
So just give a try and segment the subject.img, not the wsubject.img. You should then get subject_seg_sn.mat and subject_seg_inv_sn.mat (without the "w"), and this *inv*.mat should result in a nice back-projection from MNI space to subject space.
I wonder where you got the cluster from though. Is it from some analysis of your segmented structural volumes? This would probably mean that you have to define the clusters again based on the correct segmentation and up-dated analysis.
Best,
Gabor
-------- Original-Nachricht --------
Datum: Wed, 19 Sep 2012 13:41:51 +0000
Von: "Kong, Li" <[log in to unmask]><https://service.gmx.net/de/cgi/g.fcgi/mail/new?CUSTOMERNO=40619594&t=de424934924.1348066691.e9fc8b62&to=%26lt%3BLi.Kong%40med.uni-heidelberg.de>
An: Gabor Oederland <[log in to unmask]><https://service.gmx.net/de/cgi/g.fcgi/mail/new?CUSTOMERNO=40619594&t=de424934924.1348066691.e9fc8b62&to=%26lt%3Boederland%40gmx.ch>
CC: "John Ashburner [[log in to unmask]]"<https://service.gmx.net/de/cgi/g.fcgi/mail/new?CUSTOMERNO=40619594&t=de424934924.1348066691.e9fc8b62&to=%26lrm%3B%5Bjashburner%40gmail.com> <[log in to unmask]><https://service.gmx.net/de/cgi/g.fcgi/mail/new?CUSTOMERNO=40619594&t=de424934924.1348066691.e9fc8b62&to=%26lt%3Bjashburner%40gmail.com>, "[log in to unmask]"<https://service.gmx.net/de/cgi/g.fcgi/mail/new?CUSTOMERNO=40619594&t=de424934924.1348066691.e9fc8b62&to=%26quot%3BSPM%40jiscmail.ac.uk> <[log in to unmask]><https://service.gmx.net/de/cgi/g.fcgi/mail/new?CUSTOMERNO=40619594&t=de424934924.1348066691.e9fc8b62&to=%26lt%3BSPM%40jiscmail.ac.uk>
Betreff: RE: the problem of mapping VBM results to native space
Hi John and Gabor,
Thanks for ideas!
Although I haven't got the right method to transform my mask from MNI space to native space, I think the "wsubject_seg_inv_sn.mat" I got is not right file. If this "wsubject_seg_inv_sn.mat" can bring the mask from MNI to native space, the file "wsubject_seg_sn.mat" should be able to bring my raw image "subject.img" to MNI space. I tested with spm_write_sn ('subject.img', 'wsubject_seg_sn.mat') and I expected warped subject (wsubject.img) should be in MNI space and similar to the "wsubject.img" I got from "Normalization". But it isn't.
So I am afraid I got this "wsubject_seg_inv_sn.mat" in wrong way. I first did VBM analysis with defaut setting and I did not got any *_seg_inv_sn.mat file. Therefore, I planned to re-run my subjects step by step. I first copied one subject to a folder for testing and did normalization and segmentation for this subject (first normalization and then segmentation). After normalization, I got one mat file "subject_sn.mat". "wsubject_seg_sn.mat" and "wsubject_seg_inv_sn.mat" were got from segmentation. Do you think this is the right way getting *_seg_inv_sn.mat???
My cluster.img locates in precentral gyrus. The peak of the cluster in in high GM density.The qualities of raw image, normalized and segmented image are OK, not bad.
I will enclose them (subject.img is the raw image; wsubject.img is warped image by normalization; L_precentral_t.img is the cluster from spm_T.img).
Thanks!
Li
________________________________________
From: Gabor Oederland [[log in to unmask]]<https://service.gmx.net/de/cgi/g.fcgi/mail/new?CUSTOMERNO=40619594&t=de424934924.1348066691.e9fc8b62&to=oederland%40gmx.ch>
Sent: Wednesday, September 19, 2012 2:15 PM
To: Kong, Li
Subject: Re: the problem of mapping VBM results to native space
Hello Li,
maybe you could attach some pictures to get an idea of your data? As John said, one option would be that your structural volume was reoriented after segmentation or the segmentation/normalization has a poor quality.
Something different which came to my mind: did you smooth your segmented data? Sometimes you get clusters where the probability of GM is rather low. So, where is the cluster located in your VBM data? Is the peak of the cluster at some position with high GM volume/density or somewhere with rather low GM volume/density? In the latter case, it might actually be that when back-projected to the original volume (no smoothing), the cluster seems to be lying outside the brain (because it stems from smoothed data). Dunno, just another idea :-)
Best,
Gabor
|