P.S.  Part of my motivation is that I'd like to avoid or minimize
smoothing where possible, so that the regions of statistically
significant differences in volumetric change between groups are as
clean as can be expected.   So I guess my overarching question is
whether the using the approach suggested by John Ashburner below
(which segments the subject averages and then creates a Dartel
template with the subject average segmentations alone), is a good as
it gets in this regard.

Thanks again,
David Romano

On Thu, Sep 11, 2014 at 5:32 PM, David Romano <[log in to unmask]> wrote:
> Dear SPM folks,
>
> I'm working with a population of children who were scanned at two and
> four years old, and I'm interested in a simple comparison of the
> volumetric changes corresponding to two subgroups within this
> population, for grey and white matter separately.  (No GLM analysis.)
>
> In looking over the workflow suggested below by John Ashburner in
> response to a query by Richard Binney, I'm wondering whether some
> hybrid between this workflow and the suggested pre-spm12b workflows
> might be warranted because of the amount of change that happens
> between the ages of two and four, which could result in a more
> pronounced blurring of the subject average images.  Specifically, I
> wonder whether first segmenting the native images and then warping
> them to subject average space would give better segmentations than
> segmenting the subject average images themselves.  For example, might
> it be preferable to proceed alternatively as follows?
>
> 1)  Run the pairwise longitudinal registration, generating subjects
> averages avg_A_B and deformations y_A and y_B, where A and B are the
> early and late native images, and suppose we could also generate the
> corresponding Jacobian images jy_*.
> 2)  Segment the native images, generating c1 and c2 images for each of A and B.
> 3)  Warp the segmented images into subject average space by applying
> the deformations from step 1 to the c1 and c2 images.
> 4)  Multiply  the warped wc1 and wc2 images by the corresponding jy_* to
> obtain modulated versions mwc1 and mwc2  which can then be subtracted
> (i.e., mwc1_B - mwc1_A, etc.) to
> obtain volumetric change images vc1 and vc2 in subject average space.
> 5)  Apply Dartel to the collection of image triples (mwc1_*, mwc2_*,
> avg_A_B), one for each of the early and late images of each
> subject, generating a template.
> 6)  Normalize and modulate the vc1 and vc2 images to template space,
> and then to MNI space.
>
> (I'm not sure how to obtain the Jacobian files from step 1 in spm12b,
> so maybe spm8 would be required to produce them.)
>
> Thanks in advance for your thoughts on this!
>
> Best regards,
> David Romano
>
>
> ---------- Original message ----------
> Subject: Re: Longitudinal VBM using DARTEL
> From: John Ashburner <[log in to unmask]>
> To: John Ashburner <[log in to unmask]>
> Date: Fri, 8 Mar 2013 11:19:43 +0000
>
>> For looking at GM atrophy from longitudinal pairs of scans, you could
>> do the following....
>>
>> For each subject
>>     Run the SPM12b pairwise longitudinal registration to generate the
>> subject average and Jacobian difference (jd)
>>     Segment the subject average, generating c1, rc1, rc2
>>     Use ImCalc to compute c1.*jd (possibly dividing the result by the
>> time difference to give the rate of atrophy)
>>
>> Run Dartel, aligning the rc1 and rc2 images from all subjects together
>>
>> Normalise and smooth the c1.*jd images
>>
>> Run stats
>>
>> Note that computing TIV in SPM8 is not as straightforward as it was,
>> because the fluid class (c3) does not just consist of CSF.  There is
>> also eyeball fluid, and a bit of other stuff too.
>>
>> Best regards,
>> -John