I would not suggest skull stripping as the noise is estimated from the
background of the images. The "noise" in the scalp is probably not
really noise, but a result of squashing the soft tissue outside the
skull. I would also not suggest bias correcting first, as when you do
this, the noise is magnified in some areas more than others. If the
registration seems to need more bias correction, then the
regularisation can be adjusted accordingly.
Best regards,
-John
On 29 May 2013 01:15, Jeff Browndyke <[log in to unmask]> wrote:
>> Thanks, John. I appreciate the input. I did indeed have the later SPM12b version with the time interval model insertion function. I figured out after some experimentation that the time interval needed to be specified in year integers after trying days.
>>
>> After running the process, my results revealed lots of noise in the periphery, so I suspect something was askew from using FSL BET/Susan skull stripped and bias corrected data for initial DARTEL insertion. I'm going to try again with non-skull stripped originals to see if that was the issue.
>>
>> Warm regards,
>> Jeff
>>
>>
>> On May 28, 2013, at 2:48 PM, John Ashburner <[log in to unmask]> wrote:
>>
>>>> I have constructed the cl*jd files such that they are corrected by the time
>>>> difference between baseline and 1 year, giving me a map for the rate of atrophy.
>>>
>>> Check which version of the software you have. In revision 5430, I
>>> changed the pairwise code so that the Jacobians or divergences no
>>> longer needed to be divided by the time difference. If the user
>>> interface gives the options to "Save divergence" and "Save Jacobian
>>> Differences", then this is the older version, for which you should
>>> divide the results by the time difference. If it says "Save
>>> Divergence Rate" and "Save Jacobian Rate", then there's no need to do
>>> this division. The online help for the later version should also
>>> explain that the values have been divided by the time interval.
>>>
>>>
>>> A few questions. Should the DARTEL to MNI normalize (step #5) be done
>>> with the rate of atrophy maps with preserve concentrations
>>> (unmodulated) or preserve amount (modulated)?
>>>
>>> I'm not entirely sure here. Really, I should look into the equations
>>> for something called "parallel transport" (see
>>> http://en.wikipedia.org/wiki/Parallel_transport ) to see which
>>> approach gives an answer that is closer to the proper maths. I
>>> haven't done this yet though.
>>>
>>> I would guess that providing you report what you have done, then
>>> either way should be OK.
>>>
>>>> Also, when I went through step number #1 above I also produced
>>>> "dv_" images. It's my understanding that these are tensor-like maps
>>>> denoting the expansion/contraction between baseline and 1 year.
>>>> Can these be used in the above processing framework to produce a
>>>> TBM analysis of rate of expansion/contraction? If so, how? Any
>>>> special considerations in using these dv_ tensor images? Log
>>>> transformation, etc.?
>>>
>>> The divergence maps should not need log transforming. They
>>> essentially give the rate of volume change, which is a sort of
>>> logarithm anyway. They have values similar to the logarithms of the
>>> Jacobian determinants.
>>>
>>> The divergence is the trace of the Jacobian tensor. Consider the
>>> rotation and translation part of a simple affine transform. This can
>>> be parameterised by Jdef = expm(Jvel), which allows Jvel to have a few
>>> nice(ish) properties. Then we see that the trace of Jvel is equal to
>>> the log of the determinant of Jdef.
>>>
>>> Jvel = randn(3);
>>> Jdef = expm(Jvel);
>>> trace(Jvel)
>>> log(det(Jdef))
>>>
>>> It's more complicated for nonlinear deformations, which are computed
>>> from the initial velocity via Riemannian exponentials (see
>>> http://en.wikipedia.org/wiki/Exponential_map#Riemannian_geometry ).
>>>
>>>
>>> In practice, if the deformations are sufficiently small, the things I
>>> called "Jacobian rates" should be pretty close to what I called
>>> "Divergence rates".
>>>
>>> Jvel1 = randn(3)*0.01;
>>> Jvel2 = randn(3)*0.01;
>>> trace(Jvel2)-trace(Jvel1)
>>> det(expm(Jvel2))-det(expm(Jvel1))
>>>
>>> If you find Jacobian determinants easier to explain in a paper, and
>>> the deformations are small, then you could probably just as well use
>>> those.
>>>
>>> Best regards,
>>> -John
>>>
>>>
>>> On 25 May 2013 15:09, Jeff Browndyke <[log in to unmask]> wrote:
>>>>
>>>> John (and other SPMers),
>>>>
>>>> I've been implementing the following processing stream in SPM12b with
>>>> regards to the baseline and 1 year structural data:
>>>>
>>>> For each subject:
>>>>
>>>> 1.) Run the SPM12b pairwise longitudinal registration to generate the
>>>> subject average and Jacobian difference (jd)
>>>>
>>>> 2.) Segment the subject average, generating c1, rc1, rc2
>>>>
>>>> 3.) Use ImCalc to compute c1.*jd (possibly dividing the result by the time
>>>> difference to give the rate of atrophy)
>>>>
>>>> 4.) Run DARTEL, aligning the rc1 and rc2 images from all subjects together
>>>>
>>>> 5.) Normalise and smooth the c1.*jd images
>>>>
>>>> 6.) Run stats.
>>>>
>>>>
>>>> I have constructed the cl*jd files such that they are corrected by the time
>>>> difference between baseline and 1 year, giving me a map for the rate of
>>>> atrophy.
>>>>
>>>> A few questions. Should the DARTEL to MNI normalize (step #5) be done with
>>>> the rate of atrophy maps with preserve concentrations (unmodulated) or
>>>> preserve amount (modulated)?
>>>>
>>>> Also, when I went through step number #1 above I also produced "dv_" images.
>>>> It's my understanding that these are tensor-like maps denoting the
>>>> expansion/contraction between baseline and 1 year. Can these be used in the
>>>> above processing framework to produce a TBM analysis of rate of
>>>> expansion/contraction? If so, how? Any special considerations in using
>>>> these dv_ tensor images? Log transformation, etc.?
>>>>
>>>>
>>>> Warm regards,
>>>> Jeff
>>
>
> -----------------------------------------------------------------------
> Jeff Browndyke, Ph.D.
> Clinical & Research Neuropsychologist
> Durham VA & Duke University Medical Centers
>
> [log in to unmask] / [log in to unmask]
> office: (919) 286-0411 ext. 4656
> cell: (336) 264-4222
|