Dr. Ashburner and all:
In the optimized protocol described in the Good et al paper, the used the
average of the segmented and smoothed images from the standard protocol. I
am at the point in the standard protocol where I am spatially normalizing
the subset of the whole brain(not segmented)images to the MNI whole brain
template. The Good et al article describes using the mask brain function at
this point for the reasons you indicate below.
I have been told not to use a the masking procedure by a colleague, what do
you all advise?
Rob McClure
>
> 4.) I have been told not to use a the masking procedure to weight the
> normalization to nonbrian tissue. Is this correct?
> If not what sort of mask should I use?
If you are spatially normalising grey matter to match grey matter, then
you should get the best results without any "brain masking". Incedentally,
the masking procedure weights the spatial normalisation so that non-brain
tissue has less influence on the final results.
-----Original Message-----
From: John Ashburner [mailto:[log in to unmask]]
Sent: Monday, January 14, 2002 11:55 AM
To: McClure, Robert (NIMH); [log in to unmask]
Subject: Re:
> I am making a template for optimized voxel-based morphometry using the
> "standard" Good et al. protocol and I have four questions about the
> normalization defaults in SPM:
>
> 1.) I've been told to use the number of non-linear basis functions to be 4
> x 5 x 4. Is this appropriate?
If someone has good evidence that 240 basis functions work best for
spatially
normalising your data, then use 240 basis functions.
The objective function for spatial normalisation is based on simultaneously
minimising two cost functions:
1) The sum of squares difference between the image and template
2) A measure of roughness of the warps.
In SPM99, the latter cost function is based on the sum of squares of the
gradients of the warp. Recently, there has been some evidence that this
cost function may allow relatively too much variability of high frequency
warps compared to those at lower frequencies, so maybe a penalty based
on bending energy (sum of squares of second derivatives of the warps)
would be more appropriate. Truncating the number of basis functions from
the default 1029 down to 240 may be the easiest way of approximating this
distribution using the spatial normalisation procedure of SPM99.
>
> 2.) I've been told to use the number on nonlinear iterations should be 8
or
> 12. How should this be determined?
I have always thought that the more iterations, then the better the
algorithm
should converge.
>
> 3.) The voxel size of my scans are .9371 x .9375 x 1.5 mm. I have been
> told to reslice the spatially normalized images to
> a voxel size of 2 x 2 x 2 mm, but would like to go to 1.5 x 1.5 x
> 1.5 mm. Is this appropriate?
If you are going to segment after spatial normalisation, then this will work
best on higher resolution spatially normalised images. The SPM segmentation
model does not handle partial volume effects particularly well. If your
spatially normalised images are higher resolution, then there are relatively
fewer voxels that contain a mixture of different tissue types.
There is no particular reason why the template images should have 2mm
resolution,
rather than 1 or 1.5mm.
>
> 4.) I have been told not to use a the masking procedure to weight the
> normalization to nonbrian tissue. Is this correct?
> If not what sort of mask should I use?
If you are spatially normalising grey matter to match grey matter, then
you should get the best results without any "brain masking". Incedentally,
the masking procedure weights the spatial normalisation so that non-brain
tissue has less influence on the final results.
Best regards,
-John
--
Dr John Ashburner.
Functional Imaging Lab., 12 Queen Square, London WC1N 3BG, UK.
tel: +44 (0)20 78337491 or +44 (0)20 78373611 x4381
fax: +44 (0)20 78131420 http://www.fil.ion.ucl.ac.uk/~john
|