Such a 2 stage procedure is only really necessary if the deformation is
parameterised by a small number of parameters. If the deformations don't
need to also model variability outside the region in which you are
interested, then it is better able to model variability within that region.
According to my contacts within FMRIB, they are developing some nice nonlinear
warping algorithms, which use more parameters. When their new algorithms are
released, then such a masking procedure should no longer be required.
Similarly, there is also a new warping procedure within SPM5. You can access
it via TASKS->Tools->DARTEL, and there is some documentation about how to use
it in the man/manual.pdf file.
Best regards,
-John
On Friday 26 October 2007 17:58, Buyean Lee wrote:
> Dear Ged,
>
>
>
> I would like to ask one more similar question; can one safely use FSL
> processing in SPM5 analysis.
>
>
> For example, if one is interested in only the subcortical regions (e.g,
> volume or dopamine D2 receptors), I think one can use the following
> two-stage normalization in order to improve the accuracy of normalization
> of the subcortical regions.
>
>
> This is the description in the FSL website.
>
>
>
> "first_flirt - This script runs two-stage affine
> registration to MNI152 space at 1mm resolution (we will assume for
> these instructions that the image is named im1.nii.gz). The first
> stage is a standard 12 degree of freedom registration to the template.
> The second stage applies a 12 degrees of freedom registration using an
> MNI152 sub-cortical mask to exclude voxels outside the sub-cortical
> regions."
>
>
> I can easily observe that the normalized images by this two-stage
> registration is better fitting into the standard MNI space, if I focus on
> only the subcortical regions; but normalization of other areas such as
> cortex is worse than regular normalization.
>
> Since I am using mainly SPM5, I would like to know if I can safely use this
> FSL normalzation in SPM5 voxel wise analysis. Alternatively, can I run this
> kind of two-stage normalization in SPM5?
>
> Additional question is that if I can use this two-stage normalization,
> should I use the explicit mask (= the subcortical mask used in the second
> stage normalization of first_flirt) during the statistical analysis?
>
> Many thanks in advance.
>
> Buyean
>
>
>
>
>
>
>
>
>
>
>
>
>
> -----Original Message-----
>
> From: DRC SPM <[log in to unmask]>
>
> To: [log in to unmask]
>
> Sent: Thu, 25 Oct 2007 3:16 am
>
> Subject: Re: [FSL] sienax problem 512x512 scans
>
>
>
>
>
>
>
>
>
>
>
>
>
> Hi Buyean,
>
> > can one use SPM5 Display to reorient the image (Nifti) for FSL
>
> SPM5's reorientations get written to the NIfTI headers, so you could
>
> use FLIRT's -usesqform option to resample the reoriented images as
>
> Mark describes here:
>
>
>
> http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0710&L=fsl&P=40969
>
>
>
> However, I think Steve made the point that you should probably avoid
>
> the interpolation errors from such reorientation and resampling, by
>
> instead using only fslswapdim. As you said, it is limited to 90 degree
>
> rotations, but those have the benefit of being exact -- no
>
> interpolation is needed. Hopefully if you get the orientation correct
>
> to the nearest 90 degrees then the flirt calls within sienax should be
>
> successful. If not, you might be able to modify the sienax script to
>
> make use of the usesqform FLIRT option internally, but I haven't
>
> looked into that... (I hope this isn't the third idiotic email I've
>
> sent to this list recently... ;-))
>
>
>
> Note that manual reorientation is more important in SPM5 than in FSL,
>
> because at present the unified segmentation model in SPM5 doesn't
>
> model non-brain tissue very well, so its affine registration step
>
> needs a good rigid initialisation to succeed. (There are a few posts
>
> by John Ashburner on the SPM list with similar wording.) FLIRT on the
>
> other hand, uses a simpler, more general cost function, with a more
>
> sophisticated and robust optimisation routine (described in the tech
>
> reports, and the course slides last time I looked), which generally
>
> seems to be very good. In fact, I once played around with using FLIRT
>
> to generate the initial transformations for SPM5; this seemed to work
>
> well if the subjects can be successfully registered to MNI using rigid
>
> registration only, but gets complicated if you have to enable scaling
>
> for FLIRT to work well.
>
>
>
> Anyway, I'll stop waffling.
>
> Best,
>
> Ged
>
>
>
>
>
>
>
>
> ________________________________________________________________________
> Check Out the new free AIM(R) Mail -- Unlimited storage and
> industry-leading spam and email virus protection.
|