Hi Anastasia,
I fully agree with both Markus and Matt. I guess slightly eroding your FA image as Jesper indeed did for fnirt would have helped flirt, but the best approach is definitely using fnirt FA to FMRIB58 (following tbss1, tbss2 -T and tbss3)...
You can either define your ROIs in the MNI152-T1-1mm template and correct it manually on the FMRIB58 or you can define them directly on the FMRIB58.
Then, the only minor correction compared to what Markus has said is that I would inverse the warpfield generated by fnirt (*_to_target_warp.nii.gz) with invwarp and *then* apply the new warpfield from FMRIB58 to FA to your ROIs with applywarp (probtrackx in the standard space only accepts linear transformation - but it's going to change!)...
Hope this helps,
Gwenaelle
--- En date de : Mar 6.1.09, Matt Glasser <[log in to unmask]> a écrit :
> De: Matt Glasser <[log in to unmask]>
> Objet: Re: [FSL] T1 to b-zero registration for tractography
> À: [log in to unmask]
> Date: Mardi 6 Janvier 2009, 18h45
> This is actually a really good solution if you don't
> care about directly
> registering each subject's DTI image to their
> structural, and I have
> published results that used this technique with the earlier
> version of TBSS
> in the past.
>
>
>
> Back to the question below: there will not be any
> stretching of the
> registered image with six degrees of freedom, so if you are
> still finding
> that the registered image is stretched compared to the
> original image,
> something is not correct with how you are calling FLIRT.
> It may be,
> however, that because of EPI distortions the FA image is
> starting out
> non-linearly "stretched" relative to the T1. In
> this case, you will only be
> able to correct this with some distortion correction or
> nonlinear
> registration. I recently reviewed your discussion with
> Jesper a while back
> about using FNIRT to register the FA to the T1, as I had
> some data where no
> method of distortion correction was available. While FNIRT
> is not written
> for this application--EPI distortion is only in one
> direction (the phase
> encoding direction of the image), so it would be better if
> FNIRT was only
> allowed to make nonlinear deformations in that direction.
> However, as I
> recall Jesper got decent results registering an FA image to
> the T1 after
> eroding the bright rim of the FA image, and using the FNIRT
> config file he
> provided on the list I also got decent results (i.e. they
> were better than
> those I got from linear registration alone). They were not
> as good I would
> typically get with a real distortion correction method,
> however.
>
>
>
> If, for your application, you are able to go directly from
> MNI to diffusion
> space without need of structurals, I would use the method
> that Markus
> outlines below. That uses FNIRT in the way it is designed
> to be used and
> any EPI distortions will be corrected as well as can be
> done without real
> distortion correction as a part of the non-linear
> registration.
>
>
>
> Peace,
>
>
>
> Matt.
>
> _____
>
> From: FSL - FMRIB's Software Library
> [mailto:[log in to unmask]] On Behalf
> Of Markus Gschwind
> Sent: Tuesday, January 06, 2009 11:06 AM
> To: [log in to unmask]
> Subject: Re: [FSL] T1 to b-zero registration for
> tractography
>
>
>
> Hi!
>
> I also have/had this problem. I found that it is very hard
> to exactly
> determine an ROI in DTI space.
>
> Since I wanted to present the DTI data as group results I
> also needed the
> corregistration into standard space. That's why I
> finally run the tbss steps
> 1-4 on all my subjects DTI data in order to get a
> corregistration matrix
> from native-DTI into MNI space for each subject. Using
> applywarp you can
> then easily transfer every fdt_paths into MNI space. I
> found that these
> results are much improved compared to FLIRT.
>
> I don't know if you, in your case can learn anything
> from that.
> Maybe that FNIRTing the native-T1 to MNI template and using
> the tbss steps
> and finally concatenate those steps, or also draw your ROI
> in MNI space and
> apply the inverted warps in order to get them into DTI
> space...
>
> I would try something like this.
>
> Please let me know if you found the best solution!!
>
> Cheers, Markus
>
>
>
>
>
> 2009/1/6 Anastasia A Ford <[log in to unmask]>
>
> Matt,
>
> I tried registering my FA map to the T1 scan using both 6
> and 12 Dof. The
> resulting images look identical, with exception of a few
> voxels. The
> registered FA to T1 image is stretched out beyond the
> borders of the
> original T1 image. Is there anything you could suggest to
> improve the
> registration?
> You mentioned that I could try to use FNIRT for this
> registration. I tried
> FNIRT a while back, but the results were similar and I was
> advised against
> using FNIRT for this registration since FNIRT was not
> written for
> registration of images with different contrast. Has that
> changed recently?
>
> Thank you,
>
> Anastasia
>
>
>
>
> --
> Dr. med. Markus Gschwind, M.D.
> Laboratory for Neurology and Imaging of Cognition
> Dept of Neurosciences
> University Medical Center (CMU)
> 1 Michel-Servet - 1211 GENEVA - CH
>
> Tel 0041 (0) 22 379 5324
> Fax 0041 (0) 22 379 5402
> email: [log in to unmask]
> http://labnic.unige.ch
|