For VBM, we are generally interested in tissue volumes. Therefore,
scaling the warped tissue maps by the Jacobian gives the amount of
tissue within each region, which can subsequently be compared by
fitting a GLM.
For fMRI, most people are interested in comparisons of the average
signal within each region, under the assumption that the regions are
well aligned across subjects. They are generally not interested in
the total amount of BOLD signal change within each region, although
it's possible that the latter may occasionally be a better model. If
the objective is to assess what happens on average (rather than to
compare groups of subjects), I suspect that there would be less
variance among maps of average signal change than there would be among
maps of total signal change. I would expect this decrease in residual
variance to lead to greater sensitivity.
For group comparisons, the situation is more complicated because of
the anatomical variability among populations of people. Formulating a
sensible question of the data (ie whether to ask about average BOLD
response, total BOLD response etc) requires knowledge about a good
model of the data with which to frame the question. If you specify
the question around a poor model, then it's more likely to be a dumb
question. The most accurate model of the difference between the groups
is something that could be determined by model selection.
Returning to the point. Computing the average intensity within a
region would be done by summing up the values in the region and
dividing by the number of voxels. However, this is slightly less
straightforward if the images have been warped. In this case the
approach would be to sum up the intensities multiplied by the Jacobian
determinants (which gives the sum of the values in the region if it
was projected onto the original scan) and divide by the sum of the
Jacobian determinants of the region (which gives a count of the number
of voxels contained in the region if it was projected onto the
original image).
A similar strategy could be used for generating smoothed versions of
the warped images. This involves smoothing the Jacobian scaled warped
data and dividing this by the smoothed Jacobians.
In practice though, the approach that is used is slightly different in
that rather than resampling the image according to the spatial
transform and scaling by the Jacobians. Instead, the "normalise to
MNI" option of Dartel will use the inverse of the transform to project
all the original values into their new location in the normalised
version. This exactly preserves the tissue volumes. This would then
be smoothed, and divided by a smoothed map of the voxel counts.
Best regards,
-John
On 21 December 2011 11:24, Michel Dojat <[log in to unmask]> wrote:
> Thanks John.
>
> So the problem may more be stringent if we stay in the population template
> space.
>
> Deformation fields locally create dilatation or compression of voxels. This
> directly impacts the corresponding grey levels which represent tissues
> composition in structural images or de-oxyhemoglobin concentration in case
> of functional images.
> I do not really understand why modulation makes sense only for the former
> case.
>
> Best Regards
> M
>
>> ---------- Forwarded message ----------
>> From: John Ashburner<[log in to unmask]>
>> Date: Tue, 20 Dec 2011 12:44:54 +0000
>> Subject: Re: [SPM] modulation for functional images
>> To: [log in to unmask]
>>
>> I don't think "modulation" makes so much sense for spatially
>> normalising fMRI. If you use the "normalise to MNI space" option
>> without "modulation", there is a correction included so that the
>> smoothing accounts for the volumetric changes.
>>
>> Best regards,
>> -John
>>
>> On 20 December 2011 10:29, Michel Dojat<[log in to unmask]>
>> wrote:
>>>
>>> Dear All,
>>>
>>> When applying deformation field calculated using DARTEL pipeline to
>>> functional MRI, is it needed to modulate in order to compensate,
>>> similarly
>>> to VBM applications, for local modification of the images ?
>>>
>>> thank for your reply
>>>
>>> M
>
>
>
>
|